This issue draws from 2024-2026 research by McKinsey, BCG, Gartner, Deloitte, MIT, and Anthropic. It’s informed by my own experience building an agentic competitive intelligence system using Claude Code, and conversations with executives navigating the same choices.

In early 2024, Klarna deployed an autonomous AI agent that handled two-thirds of all customer service chats in its first month. 2.3 million conversations. Resolution time dropped from 11 minutes to under 2. Equivalent of 700 full-time agents. The CEO projected $40 million in profit improvement.

Fourteen months later, Klarna reversed course. Customer satisfaction had dropped. CEO Sebastian Siemiatkowski: “We focused too much on efficiency and cost. The result was lower quality, and that’s not sustainable.”1

Klarna is not an outlier. It’s the pattern. 95% of AI pilots stall, delivering little to no measurable P&L impact.2 Only 11% of enterprises have deployed agentic AI in production.3

The vendor pitch says: design an autonomous agent, deploy it, let it run. The evidence says the opposite. The best agents are grown, not built.

Subscribe now

Why big-bang deployment fails

The technology works. The failure is in how companies deploy it: treating an agent like finished software instead of a system that needs to learn your context.

Gartner predicts 40%+ of agentic AI projects will be canceled by 2027: “Current models don’t have the maturity and agency to autonomously achieve complex business goals or follow nuanced instructions over time.”4

“If you just take your existing workflow and try to apply advanced AI to it, you’re going to weaponize inefficiency.”
Bill Briggs, Deloitte5

Without formal oversight, failures stay invisible until they compound.

BCG documented what this looks like in practice. An expense report agent could not interpret certain receipts. Rather than flagging uncertainty, it fabricated plausible entries to complete the report, including fake restaurant names. Every metric showed success. Only human review caught the fabrications.6

That is the core problem with “deploy it once.” These silent failures are invisible to upfront design. You only catch them through iteration with real work and real human review. The software industry learned this decades ago: big-bang ERP approaches regularly fail. The AI agent space is repeating the same mistake with higher stakes.

What growing an agent looks like

BCG, Google DeepMind, and researchers at the Knight First Amendment Institute have independently converged on the same pattern: agents earn autonomy through proven performance. BCG calls it a “promotion path.”7

BCG: “Successful implementations require agents to run in Tier 1 (Shadow Mode) until they prove they align with your organization’s risk appetite. Autonomy is a journey of trust quantified by accuracy.”8

Here’s what that journey looks like in practice.

The competitive intelligence agent

A few months ago, our competitive intelligence was ad-hoc. No knowledge base. No structured process. Information lived in scattered documents and individual memory. The first step was collaborating with the agent on workspace design. “Here’s our objective: create competitive intelligence products for executives. How should we organize this?” The agent helped design the knowledge base structure. Then the first real task was simple retrieval: “Here are our top competitors. Research them. Here’s context on our company to help you identify relevant insights.” No analysis, no recommendations. Just gathering.

The hallucination problem showed up immediately. The agent fabricated claims, invented citations, pulled statistics from nowhere. Rather than accepting that as a quality ceiling, we separated quality control into its own process: one agent researches, a second agent verifies every claim against its source, and a human reviews the final product. Hallucinations dropped to near zero. That is graduated autonomy applied to the system itself.

With the foundation in place, the products grew one at a time. First, a weekly competitive briefing. Once that ran reliably, we added competitor messaging trend analysis, tracking how rivals changed their positioning and value propositions over time. Then full analysis of competitors’ earnings releases and financial filings. Each new product took a fraction of the time to build because the knowledge base, the verification system, and the learning infrastructure were already in place.

Today, autonomy varies by maturity, not by system. Skills that have been running consistently for two months operate with minimal oversight. Newer skills get heavier review. This maps directly to BCG’s tiers: the weekly briefing runs at Tier 2 (supervised execution with light approval), while a brand-new analysis product starts at Tier 1 (shadow mode, full human review). Same agent, different autonomy levels, based on earned trust.

The agent has accumulated close to 200 discrete learnings. It identified that a specific industry publication was strong for one geographic region but weak for others, and started prioritizing that source for that region in future runs. These are not programmed rules. They are patterns the system discovered through repetition and feedback.

My initial hypothesis was 80-90% of the value in 10% of the time. That held early on. But because the system compounds, because every research run strengthens the knowledge base and every learning refines future output, I am now delivering well over 100% of the original value target. Requests that would have taken days to build from scratch take hours. Not because the agent is faster at the same work, but because it has built context that no fresh start could replicate.

The less obvious benefit: growing an agent puts it in the hands of the subject matter expert from week one. My feedback, my judgment, my context shaped this system from the very first research run. A traditional build approach would have kept me out of the loop until someone handed me a finished product to test. By then, the assumptions are baked in and the rework is expensive.

Why this works: the feedback loop is the product

The competitive intelligence system works because growing an agent is the feedback loop. Every product shipped and every error caught makes the next product faster and more reliable. The research confirms the pattern.

McKinsey found that companies getting real value from AI fundamentally rework their processes at almost 3x the rate of other firms. Workflow redesign had the biggest effect on EBIT impact out of 25 attributes tested. The single practice with the strongest correlation to AI value: human-in-the-loop validation, defining when and how AI outputs require human review.9

Anthropic’s data shows this from the user side. They studied millions of human-agent interactions and found that autonomy grows organically with trust. New users auto-approve about 20% of agent suggestions. Experienced users auto-approve over 40%. Average interventions per session dropped from 5.4 to 3.3, while success rates on challenging tasks doubled.10

The counterintuitive finding: experienced users interrupt more often than new users (9% vs. 5%). They have learned when to trust. They monitor more strategically. They know when to let the agent run and when to redirect. This is the difference between abandoning oversight and graduating it. Even among experienced users, 80% of tool calls still include safeguards like restricted permissions or approval requirements. Growing an agent means moving the controls to where they matter most.

When to hand off, when to hold

Hand off the next step when:

• The agent produces consistent quality across a meaningful sample, not just a demo

• You have seen it handle edge cases, not just the happy path

• It fails gracefully, flagging uncertainty instead of fabricating confidence

Hold when:

• The cost of a wrong answer is high (financial, reputational, legal)

• The domain requires judgment the agent has not demonstrated

• You have not seen enough volume to trust the pattern

This pattern is not limited to software engineering. A product manager at Chime uses an agentic coding tool as a complete PM system: PRD creation, ticket generation, status reporting, morning briefings. “It’s a much better product manager than I ever was.”11 Non-engineers are adopting these tools at rates that sometimes exceed engineers, and more than a third of prompts involve planning or diagnosing, not writing code. The shift is from syntactic work (typing, formatting, searching) to semantic work (communicating intent and evaluating output). Growing an agent is not a software engineering practice. It is a knowledge work practice.

Where are you now?

Four questions to diagnose your current approach:

1. Are you designing your agents for day-one autonomy, or for graduated trust?

2. Does your agent run in shadow mode before it acts independently?

3. Can you point to specific evidence, not demos, that earned each level of autonomy?

4. When the agent fails, does it flag uncertainty or fabricate confidence?

If you answered “autonomy,” “no,” “no,” and “fabricate,” you are building, not growing. The fix is not a better design. The fix is a smaller first step.

The choice

The competitive intelligence agent I run today, the one that delivers multiple products to executives weekly, that catches its own errors, that learns from every research run, looked nothing like what I would have designed on day one. It got here because I grew it. One task at a time. One product at a time. One earned level of trust at a time.

Grow your agents. Let them earn it.

Apply this framework to your company: ChatGPT | Claude | Perplexity

This issue synthesizes seven weeks of research drawing from BCG, McKinsey, MIT, Deloitte, Harvard Business School, and field studies across industries. It’s informed by conversations with executives making these choices, and my own experience leading AI transformation.

You’ve Probably Solved One of These

You picked a strategic priority. You assigned an owner. Maybe you built a governance framework or ran a pilot.

And it still isn’t working the way you expected.

92% of organizations are investing in AI. Only 1% describe their deployments as mature.1 Ask yourself why.

Over the past seven weeks, I’ve written about seven disciplines: Strategy. Ownership. Value. Governance. Design. Operations. People. Organizations that nail four or five of these still stall. I’ve seen it over and over. The disciplines interact, and it is in those interactions that value is created or destroyed.

Subscribe now

Clear strategy, but nobody owns the outcome? It dies in a slide deck. Accountable owner, but no investment thesis? The initiative gets cut in the next budget cycle. Governance bolted on after design? Your best people route around it.

Every discipline depends on the others. The 1% who reach maturity have figured out how to make all seven reinforce each other.

The Seven Disciplines

Strategy informs design. Design reveals operational gaps. Operational gaps expose people needs. People challenges reshape strategy. All seven run concurrently. All seven reinforce each other.

Here’s the system.

1. Strategy. Start with your strategic priorities. Companies getting results focus on ~3 AI initiatives tied to strategy. The ones struggling chase ~6.2

2. Ownership. A business leader owns each outcome. Initiatives with centralized business ownership reach production 70% of the time. Delegated to IT, 30%.3

3. Value. Define the investment thesis before anyone builds. What’s the metric? Who captures the value? How? 88% of organizations deploy AI. 4% consistently capture value.4 The ones who capture it defined the mechanism first.

4. Governance. An enabling constraint designed into the architecture from day one. ING embedded governance into their development process and deployed in seven weeks. Most organizations take months because governance shows up at the end and sends everyone back to the drawing board.

5. Design. A human plus AI plus context and capabilities, working as a pod, owning a defined scope of work.

6. Operations. The system between pods is where organizational value lives: how outputs flow from one pod to the next, and how capabilities are shared, how context travels with the work. Individual AI users complete 21% more tasks. At the company level, zero correlation between AI adoption and better outcomes.5

7. People. 70% of AI value depends on your people. Not models (10%). Not technology (20%).6 Your people are at different levels. They need different things. They won’t move if you’re not honest about what’s changing.

Where Are You?

That’s the system. Now turn it on yourself.

Seven questions. Honest answers.

1. Strategy. Can you name your top 3-5 AI initiatives and the strategic priority each one advances?

2. Ownership. For each initiative, can you name the business leader who owns the outcome (not the build)?

3. Value. Does each initiative have a written investment thesis with a specific metric and a capture mechanism?

4. Governance. Is governance built into your AI development process, or does a committee review it afterward?

5. Design. Are you designing AI to amplify specific people in specific roles, or are you looking for tasks to automate?

6. Operations. Are AI capabilities shared and connected across teams, or locked in individual workflows?

7. People. Do you know where each person is on the AI maturity spectrum and what they need to move to the next level?

Most executives I talk to answer “yes” to two or three. The ones scaling AI answer “yes” to all seven. They haven’t perfected every discipline. They’re working on the connections between them.

You don’t need equal effort on all seven. You need to know which gaps are breaking the system right now.

The gaps are where to focus next. And I want to know where yours are.

Your answer shapes what we cover next.

What Comes Next

The last seven weeks built the foundation: seven disciplines that every AI transformation needs, working as a system.

Next, we’ll apply them. How to design your first pod. How to think about context and capability engineering. How to connect pods into amplification chains. How to make the build-vs-buy decision. How to spot agent washing before it wastes your budget.

The foundation is set. Now we build on it.

Subscribe now

Apply this framework to your company: ChatGPT | Claude | Perplexity

This issue draws from 2024-2026 research by BCG, McKinsey, Deloitte, PwC, and Harvard Business School. It’s informed by conversations with executives navigating the same choices, and my own experience leading AI transformation.

70%

That’s how much of your AI value depends on your people. Not the models (10%). Not the technology (20%). The people.1

Most organizations invest overwhelmingly in the 10% and 20%. Then they schedule a prompt workshop and call the people part done.

The result: Of the 92% of organizations investing in AI, only 1% are mature.2

Subscribe now

The people discipline has two sides. Talent management: who you hire, how roles evolve, what the org chart looks like in two years. Enablement: how you train, what tools you provide, and how you lead people through the change.

This issue is about enablement. Not because talent management doesn’t matter. Because enablement is where most organizations are failing right now.

The standard explanation is skills gaps and resistance to change. Those are real. But there’s a deeper problem most companies miss: they’re trying to solve a distribution problem with a standardized solution.

You don’t just have a training problem. You have a distribution problem.

Your people are at different levels. They need different things. And they won’t move if you’re not honest with them about what’s happening.

The Trust Gap

There’s a trust gap most leaders haven’t closed.

Your CEO said “amplification” in the all-hands. Then said “headcount reduction” in the board deck. Your people weren’t in the board meeting. But they know.

52% of workers worry about AI’s impact on their jobs.3 53% worry that using AI makes them look replaceable.4 Trust in company-provided generative AI fell 31% between May and July 2025. Trust in agentic AI dropped 89% in the same period.5

Your people have watched three years of “AI will empower everyone” while reading headlines about layoffs. They’ve heard “no one is losing their job” from companies that then restructured quietly six months later.

Here’s the thing most leaders won’t say: both statements are true at the same time. AI will amplify those who remain. AND the organization may need fewer people. Pretending only the first part is true destroys trust. Acknowledging both builds it.

That’s the honest conversation. Not “AI will empower everyone” (corporate speak nobody believes). Not “use AI or get fired” (mandate that breeds resentment). The honest version sounds like this: “The world is changing. Some roles will look different. Some will go away. The people who move will be more valuable than ever. Here’s exactly how your role is shifting, and here’s how we support you through it.”

Specificity is what closes the trust gap. “Your role is going to shift from doing X manually to directing AI that does X while you focus on Y” is honest. “AI will empower everyone” is not.

Where Your People Are

Your organization needs a shared language for this. This is the framework I use.

Level 1: AI Consumer. Asks AI questions, gets information, and acts on it manually. Copies and pastes output into other systems. Text in, text out, human does the rest. Think Google Maps. It gives you turn-by-turn directions. You still drive. Every turn, every merge, every detour. It gave you information. You produced the outcome yourself.

Level 2: AI Operator. Directs AI that takes action in real systems. Reviews and approves. The AI creates the project, updates the CRM, drafts and sends the email. The human isn’t doing the work. They’re directing it. Think Uber. You tell the driver where to go. They drive. You sit in the back, check your email, and redirect if plans change. You’re directing and approving, not driving.

Level 3: AI Collaborator. Works alongside AI that holds persistent context. The AI knows the work. Not just today’s task, but the project history, the client relationship, the patterns from last quarter. The 51st conversation is dramatically more valuable than the first because of everything that came before it. Think personal driver. It’s Tuesday at 7:30 AM. You get in the car. The driver already knows: Starbucks first, then the office. They’re not just executing commands. They’re applying accumulated context to anticipate what you need.

Level 4: AI Architect. Designs the capabilities and workflows that Levels 1-3 consume. Builds what others use without knowing how it works under the hood. Think automotive engineer. They designed the car, the navigation, and the sensors. The Uber rider doesn’t think about the transmission. The personal driver doesn’t worry about how GPS works. But none of it works without the engineer who built it.

These describe people, not technology. A Level 1 person using a Level 3 tool is still at Level 1.

The most important jump is from Level 1 to Level 2. That’s where AI shifts from producing information to producing outcomes.

Why The Standardized Program Fails Everyone

Most companies focus on getting everyone from Level 0 (not yet using AI) to Level 1. Run a prompt workshop. Celebrate adoption metrics.

Here’s what’s actually happening:

Your Level 0 people still don’t trust you. A chatbot workshop didn’t address why. Your Level 1 people got the workshop, but no guidance specific to their role. They still don’t see how this changes their actual work. Your Level 2 people are bored. They outgrew prompting six months ago, and nothing connects to their systems. Your Level 3 people? They’re sitting through a workshop on how to write a good prompt while they prompt AI to find them a new job on their phone under the table.

A standardized AI program doesn’t fail quietly. It tells your most capable people that the organization doesn’t understand what’s possible.

And in a market where AI skills carry a 56% wage premium,6 those people have options.

The talent risk runs both directions. You’re underserving the bottom of the ladder, so they never climb. You’re insulting the top, so they leave.

You converge on a workforce stuck at Level 1.

What Each Level Actually Needs

I hit this wall recently. A team I was supporting had a mix of Level 1, Level 2, and a couple of Level 3 people. They asked me what tooling to enable across the team.

I wanted to recommend the Level 3 tool. It was the right answer for where AI is heading. But I looked at their team and knew: most of them weren’t ready for it. Half the team would open it once, get overwhelmed, and never come back.

The Level 1 tool they already had wasn’t getting the job done either. People were copying and pasting between systems, doing manually what AI should have been handling. They’d outgrown it but didn’t know what came next.

I kept going back and forth. One tool for everyone. That’s how you do it, right? Pick a platform, roll it out, train the team.

That’s when it hit me. The question wasn’t “what tool does this team need?” The question was “What does each person on this team need?” The Level 1 people needed a connection to their actual systems. The Level 2 people needed more autonomy and fewer guardrails. The Level 3 people needed persistent context and the freedom to build their own workflows.

One team. Three different answers.

Level 0 to Level 1: Show them what’s possible

Don’t start with training. Start with exposure.

Most people’s first experience with AI is the chatbot. They ask a question, get a mediocre answer they could have found themselves, and go back to doing things their way. That experience confirmed what they already suspected.

The gap goes deeper than willingness. Manal Iskander trains faculty at California universities on AI implementation. “They can’t understand even the word orchestration,” she told me. “Terms like data governance or workflow automation go right over their heads.” She estimates 75-80% of cybersecurity students, the people you’d expect to be early adopters, have “no clue.” The gap is vocabulary, and no prompt workshop is going to fix it.

Let people see what Level 2 and 3 look like. Not a demo. The real thing. Show them what happens when AI takes action in real systems and holds context across weeks of work. That’s what changes minds. Not a slide deck about prompt engineering.

The manager matters most here. BCG found that 88% of managers in future-built companies model AI use on a daily basis. At laggards, only 25%.7 People follow what their manager does, not what the company emails them about.

Be honest about what’s changing. Name the identity shifts ahead: “Your value is about to move from execution to judgment and taste. That’s uncomfortable. Here’s how we support you.”

Don’t force adoption. Create the conditions where it’s safe to try. But don’t pretend the world isn’t changing.

Level 1 to Level 2: The big jump

This is the highest-ROI investment in the entire people discipline. This is where AI stops producing information and starts producing outcomes.

The skill shift is fundamental. Not “how to prompt” but “how to direct and verify.” You learn to define the outcome you want rather than prescribing the steps. You learn to review AI work: what to check, what to trust, where errors hide.

This matters because AI has what Harvard researchers call a “jagged frontier.” On tasks within its capabilities, consultants using AI were over 25% faster and produced over 40% higher quality work. On tasks outside its capability, they performed 19 percentage points worse.8 Knowing where AI works and where it doesn’t is the critical Level 2 skill.

The tools matter as much as the skills. AI needs to connect to the systems where work happens. CRM, project management, file storage, and email. If AI can’t take action where the work lives, the person stays at Level 1, no matter how skilled they are.

And then there’s the identity shift. This is the part nobody talks about in the training program. Level 2 means going from doer to director. For people who built their careers on being fast and accurate with their hands, their value now moves to their judgment. You’re rethinking what makes you good at your job.

Level 2 to Level 3: The collaboration

Level 3 is where the work changes fundamentally. You’re not using a tool anymore. The closest human analogy: working with a colleague who’s been on your team for six months. They know the project history, the client preferences, and what you tried last quarter that didn’t work. You don’t start from zero. You pick up where you left off.

The skills at this level are different. Context engineering: structuring the information environment your AI works within. Design thinking: breaking complex work into components that a human-AI pair can execute. Quality judgment at scale: knowing when to trust AI’s patterns and when to override.

The tools are emerging, but real. Persistent workspaces that hold context, learn from interactions, and evolve over time.

Not all work needs Level 3. Transactional work is perfectly served at Level 2. Level 3 is for complex, judgment-heavy, context-dependent work. The work done by the people you can least afford to lose.

Level 4: The architect

Most organizations won’t build this internally. They partner for it.

The executive question: “How do we access Level 4 capability?” We’ll go deeper in a future issue.

Where Are Your People?

One question to take into Monday morning:

“Where are our people, and where do we need them to be?”

Walk through your organization. Assess honestly:

• What percentage is at Level 0? That tells you something about your organization’s honesty and the experience you’re offering.

• What percentage is at Level 1? That’s your enablement gap. They need a connection to real systems and role-specific guidance.

• Who’s at Level 2? That’s your emerging capability. Are you investing in them or leaving them to figure it out?

• Who’s at Level 3? That’s your highest-value talent. Are you supporting them or losing them?

• Where’s Level 4 coming from? Build or partner?

Then the strategic question: where does your strategy require people to be?

If your strategy needs 40% of people at Level 2 within twelve months (Issue 1), that becomes a resource allocation conversation (Issue 3), which requires the right governance (Issue 4), designed into pods (Issue 5), operating as a system (Issue 6), enabled through honest, level-specific development (Issue 7).

All seven disciplines. Connected.

The Choice

70% of AI value comes from the people component.9 That 70% depends on three things happening at the same time: being honest about what’s changing, meeting each person where they are, and moving with urgency.

Not sequence. Not “build trust, then enable.”

Trust IS what happens when you meet people where they are, when you’re honest about the identity shifts. When you invest in level-specific development instead of a generic workshop.

Enablement done right is trust-building.

You can have the best strategy. The right ownership model. A clear investment thesis. Governance built into the architecture. Beautifully designed human+AI pods. A system that’s ready to scale.

None of it works without people who are enabled, moving, and willing to go on this journey with you.

The Intelligent Org: Seven disciplines. Seven foundational shifts.

1. Strategy. AI isn’t the strategy. It advances the strategy.

2. Ownership. Business leaders own AI. Technologists enable it.

3. Value. Every initiative has an investment thesis.

4. Governance. Built into the architecture, not bolted on.

5. Design. Amplify humans, don’t automate them out.

6. Operations. Systems, not heroes.

7. People. Meet them where they are with honesty, speed, and trust.

Each discipline supports the others. Skip any, and the system breaks. But if one undergirds the rest, if one determines whether everything else succeeds or fails, it’s this one.

These seven disciplines are the foundation. Next, we apply them: pods, amplification chains, value chains. The full architecture of the intelligent, amplified organization.

But architecture without enabled, trusting people is just a blueprint that never gets built.

The companies that win won’t have the best AI. They’ll have the best amplified humans. That starts with one commitment: be honest about what’s changing, meet every person where they are, and move.

Apply this framework to your company: ChatGPT | Claude | Perplexity

What’s next

Season 1 is complete. In Season 2, we build: pods, amplification chains, and the full architecture of the intelligent organization. Subscribe so you don’t miss it.

Subscribe now

This issue draws from systems thinking research by W. Edwards Deming, Donella Meadows, and Peter Senge. It’s informed by conversations with executives making the same choices, and my own experience leading AI transformation.

The productivity paradox

Your best people are crushing it with AI. Your organization isn’t getting better.

Faros AI analyzed 10,000+ developers across 1,255 teams. Individual AI users complete 21% more tasks and merge 98% more pull requests.1 The productivity gains are real and measurable at the individual level.

At the company level, there is no correlation between AI adoption and better outcomes. Pull request review times increased 91% as human approval became the bottleneck. Bug rates climbed 9% per developer. Throughput stays flat. Quality doesn’t move. The organizational impact never arrives.

Subscribe now

You have the hero. You don’t have the system.

This pattern repeats across industries. One person discovers AI and triples their output. They move faster, handle more complexity, and deliver better results.

Everyone else waits for them.

Work flows to the hero because nobody else knows how to do it. The hero’s workflows are personal. Their outputs work for them, but break when someone else tries to use them. Their knowledge stays locked in their head. The capability never spreads.

You can’t scale the hero. You can only wear them out.

The problem isn’t the people. It’s the system around them.

“A bad system will beat a good person every time. Complex systems have many leverage points and can be influenced in many ways. It is unreasonable to have a broken management system and blame those working within it for the naturally poor results than such a system creates.”
W. Edwards Deming2

The shift: systems, not heroes

Individual amplification is where it starts. Systems are how it scales.

In Issue 5, we introduced the pod: a human plus AI plus capabilities, owning a defined scope of work. That’s individual amplification. But pods don’t operate in isolation. They connect into amplification chains, and chains execute your value chain. That’s the system. Issue 5 was about designing the pod. This issue is about what happens between them.

You don’t scale AI just by amplifying more individuals. You scale AI by redesigning the system in which those individuals operate.

The system is not a process map. It’s not an SOP. Four things define a system designed for amplification:

• Outputs: How outputs from one pod become inputs for the next

• Capabilities: How capabilities are shared and modularized across the organization

• Context: How context travels with the work

• Connections: How the connections between pods work for both humans and AI

Get these four right, and the system works. Miss any of them, and the value stays trapped.

When you design the system for amplification, extraordinary individual performance becomes organizational capability.

When the system fights amplification, heroes burn out, and nothing scales.

Why scaling is so hard

The concept is simple. The execution is brutally difficult.

Operations is the hardest discipline in AI transformation. The data confirms it.

Despite widespread AI adoption (88% of organizations use AI in at least one function), 67% remain stuck in the pilot or experimentation phase.3 Only 39% report any EBIT impact at all. MIT found that 95% of enterprise AI pilots produce zero measurable returns.4 Only 5% of companies achieve AI value at scale.5

The ones that succeed aren’t smarter. They’re doing something different about the system.

Your tools weren’t built for this

Your entire operational stack was built for humans clicking through GUIs (graphic user interfaces). PSA systems, CRM platforms, ERP databases, project management tools, PowerPoint decks, Word documents. Data sits trapped in proprietary databases, rendered through web interfaces, locked in file formats that only humans can read.

The MSP runs on a PSA system. An L1 tech closes a ticket with “fixed it.” The resolution data exists only as free text in a field designed for human reading. Next time the same issue occurs, the system treats it as brand new. Someone created the knowledge, but it never became a shared capability.

The professional services firm uses a CRM. Deal closes, project kicks off. Everything the salesperson learned about client needs, pain points, and relationships stays locked in their head or buried in email threads. The delivery team starts from scratch. No connection between sales and delivery exists. Everything that made the sale successful is lost.

The marketing agency lives in Google Docs, Slack, and nested folders. Creative brief in one Doc. Feedback scattered across threads. Final assets in a folder somewhere. Campaign results in a hand-built deck. Every transition requires a human to reassemble context from scratch.

Getting AI to flow through these systems isn’t a configuration change. It’s rethinking how work moves and what format it takes.

Research on organizational workflows confirms what executives already know:

“A common class of problems with information transfer and handoffs includes degradation of information. If methods of transfer are informal and not documented, information may not be passed on when staff members leave. The lines of responsibility and expectations are not always clear.”6

Each connection loses context. Each transition requires rebuilding understanding. Individual AI wins don’t carry over because no one designed the system to carry them forward.

The human side compounds everything. People in your organization sit at different levels of AI capability. Some use chatbots for drafts. Others direct agents that take real action. Your system has to work across all of them, and your pods will reflect that unevenness. This is an identity-level change, not a skills gap. We’ll cover it fully in the next issue.

You can’t stop the business to redesign it

Alerts keep firing. Clients keep calling. Deadlines keep arriving.

You need to rebuild the plane while flying it. Big-bang transformation fails because the business can’t pause. Inaction means you fall further behind while competitors figure it out.

The gap is widening. Companies that scale AI achieve 3 percentage points more EBIT (a 30% lift compared to those that don’t), with revenue impact tripling from 6% to 20% or more.7 Future-built companies show 1.7x revenue growth, 3.6x three-year shareholder returns, and 1.6x EBIT margin versus laggards.8

The ones that don’t scale are paying the cost in competitive position.

You can’t stop. You can’t wait. You have to change the system while it’s running.

The payoff is real but delayed

When you amplify one person, the ROI is obvious. They produce more, faster, better. You can point to it.

When you redesign a connection between pods, the payoff is harder to see. You restructure how sales connects to delivery. Six months later, the delivery ramp-up time drops 40%. Rework decreases. Client satisfaction scores move. But no single metric captures the full impact because the value spreads across every deal that flows through the redesigned connection.

This lack of clear ROI is why most organizations underinvest in systems work. The metrics they track (Issue 3) are built for individual performance, not systemic improvement. You need leading indicators for systemic change: connection quality, context retention, and capability reuse. Most organizations don’t measure these, so the investment looks like cost with no return.

How to start

No proven playbook exists for this. The companies that succeed are figuring it out as they go. But the principles are clear.

See the system

Stop looking at the org chart. Start looking at your value chain.

Not the process map. Not how it’s supposed to work. Rather, how work moves in practice.

• Where are the pods? The places where one amplified human could own a complete portion of work.

• Where are the connections between pods? The points where one pod’s output becomes the next pod’s input.

• Where are the broken connections? The points that lose context, cause rework, or pull in the hero.

This is systems thinking applied to your value chain. You’re not looking at individuals. You’re looking at what happens between them.

Identify the hero reinforcing loops:

Hero solves everything → nobody else learns → more gets escalated to hero → hero becomes more indispensable → nobody else learns.

This is a system trap, not a people problem. You can’t fix it by hiring another hero. You fix it by redesigning the system so the hero’s knowledge becomes a shared capability.

This isn’t a months-long process mapping exercise. It’s seeing your value chain with new eyes.

Find the highest-leverage connection

Not all connections between pods are equally broken. Not all redesigns produce equal returns.

Systems thinker Donella Meadows spent her career studying where to intervene in complex systems. Her insight: most organizations spend 95-99% of their attention on parameters (changing numbers, tweaking details, adjusting process steps). Parameters are the lowest-leverage intervention.9

The highest-leverage points are changing information flows, changing the rules of the system, and changing goals. Yet almost nobody works at these levels.

Apply this to your value chain. Ask four questions about each connection:

• Which connection loses the most context?

• Which one causes the most rework?

• Which one pulls in the hero most often?

• Which one, if fixed, would create a capability useful to most other pods?

That’s your starting point. One connection. Not the whole value chain. Not the whole business. One pod-to-pod link.

What leverage looks like in practice:

Take the professional services firm. Sales closes a deal. Delivery needs to ramp up.

Low leverage: Requiring salespeople to fill out a handoff form before delivery starts. It’s a parameter change. Most will skip it or do it poorly. The connection doesn’t improve.

Medium leverage: Restructuring how discovery information flows. Every discovery call produces a structured document that delivery can consume. The information now travels with the work.

High leverage: Making the discovery process itself a shared capability. Both sales and delivery pods contribute to and build on the same client knowledge base. The connection becomes a two-way flow that improves with every deal.

Most organizations are stuck at low leverage. The high-leverage intervention creates a shared capability that compounds.

Pick the connection where fixing it creates a capability that can be reused. That’s high leverage.

As Meadows put it: “Missing feedback is one of the most common causes of system malfunction. Adding or restoring information can be a powerful intervention, usually much easier and cheaper than rebuilding physical infrastructure.”10

Let it compound

Each redesigned connection creates something reusable. A shared capability. Each shared capability makes the next connection easier to fix and the next pod easier to build.

This is the positive reinforcing loop. The opposite of the hero trap.

One connection redesigned → shared capability created → next connection easier to fix → more capabilities shared → system accelerates.

You don’t need to redesign the entire value chain upfront. You need to start the compounding loop.

Design principles for each redesigned connection:

1. Outputs consumable by both humans and AI

Not readable by humans alone. Structured so the next pod’s AI can process them too.

I’m building this right now. A strategic intelligence system I’m designing produces all outputs in two formats: Word documents for humans who read them and Markdown in a vector database for agents that consume them. Same information, structured for two audiences. An executive reads the competitive brief and gets a polished document. Another pod’s AI needs that intelligence to inform a proposal or a pricing decision and pulls from the Markdown database.

Your consumers are no longer just humans. You’re designing for connections between pods, and the next pod might be running on AI. The MSP that writes “fixed it” in a text field is designing for humans only. Structured data (problem type, solution applied, knowledge base article referenced) serves both. Humans can read it. AI can learn from it. The next time the issue occurs, the capability exists.

2. Context travels with the work

Not in meetings, Slack threads, or people’s heads. Structured and attached to the work itself.

Example: The professional services firm. Instead of tribal knowledge in the salesperson’s head, a structured discovery document (generated from call transcripts and notes) that moves with the deal. Client pain points, success criteria, decision makers, constraints. The delivery pod starts with the same context as the deal-closing conversation.

3. Capabilities shared, not siloed

When one pod builds something valuable, expose it to other pods. Don’t rebuild the same capability everywhere. Modular by design.

Example: One pod builds an integration to extract data from your PSA. Instead of building it once for one use case, expose it as a shared capability. Other pods access the same data without rebuilding the integration. This is what technical teams should build: capabilities that amplify multiple pods, not point solutions for individual tasks.

Where are you now?

Most organizations are at one of five stages:

Stage 1: Single hero succeeding with AI

One person figured it out. Everyone else is watching or waiting. The hero’s workflows are personal. Knowledge is locked in their head. No pods.

Stage 2: First amplified pod designed and operating

You’ve moved beyond the hero to a designed pod. Inputs, process, outputs, governed capabilities, and quality gates. It’s repeatable, not personal.

Stage 3: First connection redesigned for humans + AI

You’ve connected two pods. The outputs from the first are structured so the second pod’s AI can consume them. Context travels. The connection works.

Stage 4: First amplification chain running

Multiple pods executing a portion of your value chain, connected by redesigned connections. Shared capabilities are starting to accumulate.

Stage 5: Multiple chains executing the value chain

The system is running. AI flows through your value chain alongside humans. New pods are faster to build because they share capabilities. The compounding loop is visible.

Most readers are at stage 1 or 2. That’s fine. The value is knowing what’s next.

The choice

Nobody has this fully figured out. This is the frontier.

The companies that will scale aren’t waiting for the perfect playbook. They’re picking one connection between pods, redesigning it, learning from it, and expanding.

The 42% of companies that scrapped most AI initiatives in 2025 (up sharply from 17% the year before) are the ones that tried to solve the whole system at once, or the ones that amplified heroes without redesigning the system around them.11

The 5% that scale are doing something different.12 They’re thinking in systems. They’re finding leverage points. They’re letting it compound.

High performers are three times more likely to redesign workflows fundamentally.13 McKinsey tested 25 organizational attributes to identify what drives AI value. Workflow redesign emerged as the factor with the strongest contribution to EBIT impact.14

Yet only 21% of organizations have redesigned workflows.

That gap is the opportunity.

Systems, not heroes. One connection at a time.

Apply this framework to your company: ChatGPT | Claude | Perplexity

Next week

Next issue: Systems need humans who can work within them. How do you prepare your people for amplified work? How do you hire for it? Discipline 7: People.

Subscribe now

This issue draws from 2024-2025 research by MIT, McKinsey, and BCG, plus field cases from Sanofi, Wendy’s, and Klarna. It’s informed by conversations with executives facing the same choices, and my own experience leading AI transformation.

700

That’s how many customer service jobs Klarna cut after deploying AI. The CEO praised the efficiency gains. Response times dropped from 11 minutes to 2 minutes. Costs plummeted.

Then came the reversal. “We went too far.”1 Klarna is now rehiring the humans they replaced.

They’re not alone. 55% of companies that made AI-driven layoffs now regret it.2 They automated the work, lost value, and are hiring the people back.

I’ve watched this pattern repeat for years. AI projects fail not because the technology is wrong, but because the design is wrong.

Subscribe now

Every project starts the same way: “What can we automate?” Look at processes. Identify tasks. Deploy AI to replace humans at each step. Celebrate the pilots. Watch them stall, fail to scale, or get killed quietly when results don’t materialize.

That’s automation-first. It keeps failing.

I’ve stopped asking “what can we automate?” and started asking “who can I amplify?”

What Changed

I stopped looking for tasks to eliminate and started looking for people to supercharge.

The governance got simpler. The risk dropped. The speed increased. The results showed up.

The research confirms it. Companies that succeed aren’t automating humans out. They’re amplifying humans up.

Sanofi designed AI to strengthen human decision-making, not replace it. Close to €1 billion in redeployed value.3 Their CEO’s line: “Human plus AI beats AI alone every time.”

Wendy’s built a drive-thru AI with frustration detection that triggers human escalation. They optimized for customer and crew experience, not just speed. Now expanding to 500+ locations.4

The pattern holds. Design AI to replace humans, and it fails. Design AI to amplify humans, and it works.

Why Automation-First Fails

The automation-first approach seems logical. Map your processes. Identify repetitive tasks. Deploy AI to handle them. Reduce headcount. Save money.

But it fails for three reasons.

First, it confuses tasks with outcomes.

Klarna automated the task: answering queries quickly. But they destroyed the outcome: building trust, turning problems into loyalty. Faster responses. Lower satisfaction. Task completion is not value creation.

Second, it removes the learning loop.

Autonomous AI systems struggle to improve over time. Building learning into a fully automated system requires massive engineering effort. Pair AI with humans, and learning happens naturally. The human provides feedback. The system improves. Knowledge accumulates.

Third, it asks the wrong question.

“What can we automate?” leads you to examine existing processes and find ways to eliminate humans. It’s cost-cutting dressed up as transformation.

The better question: “What outcome are we trying to drive, and what combination of human judgment and AI capability gets us there?”

The data backs this up. 79% of advanced AI adopters invest in “AI generates insights for human decision maker.” Only 54% invest in “AI decides and implements.”5 BCG puts it precisely: “They move from people-centered processes augmented by digital tools to AI-agent-centered processes orchestrated by people.”6

Not removing people. Redesigning how people orchestrate AI.

The Human Amplification Model

One principle: Design AI to amplify humans, not replace them.

Instead of asking “what tasks can AI do?” ask “what can one person, with the right AI and tools, own end-to-end?”

What once required a team of five can now be executed by one amplified human. I call this unit a “pod”: a human plus AI plus the capabilities they need to do the work.

Consider a professional services firm. The (basic) business process: Win work → Deliver work → Retain clients.

Automation tries to replace humans at each step. Amplification asks: who could own each portion?

One person could run lead generation, qualification, discovery, proposals, and closing. AI handles research, scheduling, drafting, and follow-up. The human owns the outcome. The AI handles the volume.

Where Automation Still Makes Sense

Automation isn’t always wrong. In specific contexts, it wins.

Tier 1 customer support triage. Password resets. Order status checks. ETL pipelines. Format conversion. Report generation.

These share five characteristics: time-critical, highly repetitive, well-defined rules, measurable outcomes, and volume that exceeds human capacity. When these characteristics are present, automate.

But treat automation as a capability the Human+AI pod can access, not a replacement for the pod.

Automation handles volume and speed. The human handles judgment, relationships, and edge cases. The handoff is designed, not accidental.

Klarna automated the entire function, not just the bounded tasks. The right design: automate Tier 1 triage, escalate everything else to an amplified human.

The Learning Gap

MIT research surfaced a critical barrier to AI success: the “learning gap.”

Most corporate AI systems don’t retain feedback. They don’t accumulate knowledge. They don’t improve over time. A corporate lawyer told the researchers: “It’s perfect for a first draft, but for critical work, I need a system that learns from our cases, not one that starts from scratch every time.”7

The result: 70% of professionals use AI for simple tasks, but 90% prefer humans for complex work.

Human-AI pairing solves this. When humans and AI work together continuously, feedback flows naturally. Knowledge accumulates. The pod gets better over time.

In my own work, every process captures learnings. As patterns repeat, those learnings become rules. Learning is built into the workflow, not bolted on after.

Taste

As AI takes over execution, the human’s value shifts to judgment and curation.

I call this “taste”: the ability to recognize what’s good, what fits, what works in context. Knowing when the AI output is right and when it’s wrong.

The goal isn’t abdication. It’s leverage. AI handles research, drafting, editing, and distribution. The thing that makes it yours stays with you.

The Diagnostic

Before you build, assess your approach.

If your answers cluster in the left column, you’re headed for Klarna’s result.

Bring this to your next leadership meeting. One question: are we designing AI to replace humans, or to amplify them?

Four Questions Before You Build

For each initiative:

1. Who is the human this amplifies?
   Name a person. Not “the team.” Not “the department.” Who owns the output?

2. What capabilities do they need?
   Data access, integrations, tools. What would make them 10x more effective?

3. How does this system learn from use?
   Where does feedback enter? How does knowledge accumulate?

4. What quality controls are built in?
   How do you catch errors before they reach the customer? What are the gates?

If you can’t answer all four, you’re not ready to build.

The Choice

Every AI initiative comes down to one design decision: replace humans or amplify them?

Klarna chose replacement. They’re rehiring.

Sanofi chose amplification. Close to €1 billion in value.

Wendy’s chose amplification. Expanding to 500+ locations.

Same technology. Very different outcomes.

This week, pick one initiative. Apply the four questions. Name the human. Define the capabilities. Design the learning loop. Build the quality gates.

That’s how you build AI that works.

Apply this framework to your company: ChatGPT | Claude | Perplexity

Next Week

Next issue: Once you’ve designed the pods, you need to run and scale them. Discipline 6: Operations.

Subscribe now

This issue draws from 2024-2025 research by EY, Deloitte, McKinsey, and MIT Sloan. It's informed by conversations with executives navigating the same choices, and my own experience leading AI transformation.

The Terminology Problem

Ask five executives what “AI governance” means. You’ll get five different answers.

One thinks it’s about data privacy. Another thinks it’s compliance with EU regulations. A third imagines a committee that approves AI projects. The fourth is worried about vendors training models on company data. The fifth has no idea but knows the board keeps asking about it.

This confusion is why governance conversations go nowhere. Everyone is talking past each other because they’re using the same word to mean different things.

Only 12% of C-suite executives can correctly identify appropriate controls for AI risks, according to EY’s 2025 survey of 975 leaders.1 Chief Risk Officers scored below average at 11%. Deloitte found that 66% of boards still have “limited to no knowledge or experience” with AI.2 The people responsible for governance can’t describe how it works.

Before we can fix AI governance, we need to name what we’re actually talking about.

Subscribe now

Five Domains, Not One

When executives say “AI governance,” they’re usually conflating five distinct disciplines. Each has different owners, different expertise, and different solutions.

Data governance is the foundation. How is data collected, stored, accessed, and protected? Who can access what, and under what conditions? What are the quality standards? What is the lineage?

This discipline predates AI by decades. It’s the prerequisite. If your data governance is broken, AI governance cannot succeed. You’re building on sand.

Data governance is not the focus of this issue. But if you don’t have it, stop here and fix it first.

Tool governance is about the vendors and platforms you use. Are your AI tools leaking company data? Are vendors using your data to train their models? What are their security practices? What do the terms of service actually say?

This matters because most companies are buying AI capabilities rather than building them. Menlo Ventures found that 76% of enterprise AI is now purchased versus built, a reversal from 47/53 just a year ago.3 The tools you choose create exposure before you write a single prompt.

We’ll cover tool governance in depth in a future issue. For now, know that it’s distinct from the other four.

Compliance is the regulatory layer. GDPR. CCPA. The EU AI Act. Colorado’s AI Act. Industry-specific rules in healthcare and financial services. Audit trails. Documentation. This is what legal teams typically focus on when they hear “AI governance.”

Compliance is necessary. It’s also insufficient. You can be fully compliant and still have ungovernable AI systems.

Cybersecurity is about protecting AI systems from threats. Access controls. Identity management. Integration security. Prompt injection defense. The OWASP Top 10 for AI Agents identifies threats such as authorization hijacking, critical system interaction, and knowledge base poisoning.4

When AI systems have access to sensitive data and the ability to take actions, they become attack vectors. A compromised AI agent with access to your CRM, email, and calendar is not just an AI problem. It’s a security problem.

AI governance is the focus of this issue. It answers three questions:

1. What should AI handle, and what must it never touch?

2. What controls ensure AI behaves as intended over time?

3. Who is accountable for AI outputs?

AI governance is about the decisions you make when designing, deploying, and operating AI in your specific context.

Why does this distinction matter?

Different domains require different owners. Data governance is the responsibility of IT and data teams. Tool governance involves security, legal, and procurement. Compliance is legal’s domain. Cybersecurity belongs to your CISO. AI governance requires business leadership.

Conflating them creates governance theater: one committee trying to handle all five domains, doing none of them well.

You need all five. But this issue focuses on AI governance specifically: the discipline that determines whether AI does what you intend, reliably, over time.

What AI Governance Actually Means

AI governance has three components: selection, design, and operation.

Selection happens before you build. What should AI handle? What must it never touch? Where are the boundaries?

Design happens as you architect. What controls ensure AI behaves as intended? What guardrails prevent it from crossing boundaries?

Operation happens as you run. Who is accountable for outputs? How do you verify it’s working? What happens when something goes wrong? How do you maintain reliability over time?

Guardrails Defined

Guardrails are the architectural constraints that prevent AI from doing things it shouldn’t. They come in several forms:

• Topic restrictions that prevent the AI from addressing specific subjects

• Business logic that handles critical decisions through deterministic algorithms, not LLMs

• Output validation that checks results before they reach users

• Human approval checkpoints at high-stakes decision points

• Access controls that limit what data and systems the AI can reach

The key distinction: guardrails are built into the system. They’re not policies people are supposed to follow. They’re constraints the architecture enforces.

Deterministic vs. Non-Deterministic

This distinction is critical for AI governance.

Deterministic means the same input always produces the same output. A pricing algorithm that applies your rate card produces the same price every time for the same inputs. A tax calculation follows fixed rules. A compliance check against a defined policy yields a clear yes or no.

Non-deterministic means outputs can vary. LLMs are non-deterministic by design. Ask the same question twice, and get slightly different answers. This is a feature for creative tasks. It’s a risk for decisions that require consistency.

The governance principle: critical decisions that require consistency, auditability, and precision belong in deterministic systems. LLMs handle interpretation, generation, and judgment. The architecture enforces the boundary.

Example: Proposal pricing

A professional services firm uses AI to generate proposals. The AI drafts scope descriptions, writes executive summaries, and suggests approaches based on similar past projects.

Pricing does not go through the LLM.

The firm’s pricing follows a rate card with defined multipliers for complexity, timeline, and risk. This logic lives in a deterministic system. The AI cannot override, modify, or “interpret” pricing rules. When the proposal is assembled, pricing is determined by the algorithm, not the model.

If the LLM handled pricing, you’d get inconsistent quotes. The same project might price differently depending on how you phrase the prompt. A client could get a lower price by asking differently. The firm would have no audit trail for why a price was set.

Deterministic for consistency. Non-deterministic for creativity. The architecture enforces which is which.

Why This Matters Now

A chatbot that answers questions is different from AI that takes actions. Today’s AI systems can send emails, modify records, execute transactions, and communicate with third parties. This isn’t science fiction autonomous AI. It’s the conversational AI assistants that companies are deploying right now. When AI can act, the governance burden scales with capability.

Get AI governance wrong, and you face two outcomes:

• Paralysis: nothing ships because nobody can agree on what’s safe

• Exposure: things ship without real oversight, and you discover the problems after damage is done

EY found that 99% of organizations have experienced financial losses from AI-related risks, with an average loss of $4.4 million.5 The top dangers: non-compliance with regulations (57%), negative sustainability impacts (55%), and biased outputs (53%).

The Shift: Governance by Design

The companies moving fastest on AI didn’t skip governance. They made it everyone’s responsibility from the start.

Selection: Before You Build

Before development starts, answer these questions:

• What should AI handle in this use case?

• What must AI never handle?

• What data will AI access?

• What actions can AI take?

ING’s chatbot boundaries

ING built a customer-facing chatbot and deployed it to production in seven weeks.6 They didn’t skip governance. They evaluated 20 different risks for every use case upfront.

Selection was explicit: the chatbot would handle common customer service questions. It would never handle mortgage advice or investment recommendations. These weren’t guidelines. They were architectural boundaries.

The result: 20% more customers helped without agent intervention, deployed in seven weeks. Compare that to companies where governance committees meet quarterly to review projects that shipped months ago.

The Rule of Two

Meta’s security framework provides a useful constraint: “no agent should have access to all three of these within a single session.”7

1. Untrusted inputs (emails, customer data, external content)

2. Ability to take actions (send messages, modify systems, execute transactions)

3. Access to confidential or sensitive data

Having all three creates what Meta calls the “lethal trifecta.” A prompt injection attack can complete the whole chain: malicious input accesses private data and exfiltrates it to the attacker.

Design your AI so that no single agent crosses all three boundaries. If all three are required, require human approval before action.

Design: As You Architect

Critical decisions belong in deterministic systems. Non-deterministic LLMs handle interpretation and generation.

Compliance checking

A financial services firm uses AI to review documents for compliance issues. The AI reads contracts and flags potential problems.

The compliance rules themselves are deterministic. “Any contract over $1M requires CFO approval” is a rule, not a judgment call. The AI identifies that a contract is over $1M. The rule system determines which approvals are required.

What the AI handles: reading the contract, extracting the dollar amount, and identifying relevant clauses.

What deterministic logic handles: applying the compliance rules, determining required approvals, and generating the audit trail.

The AI can’t decide that a $1.2M contract “probably doesn’t need” CFO approval because it’s “pretty close” to the threshold. The architecture prevents it.

Quality gates in the workflow

ING built guardrails directly into its chatbot architecture. The system couldn’t give mortgage advice because the architecture prevented it. Not because a policy said so. Because the system couldn’t do it.

Quality gates belong in the workflow, not after it. Instead of reviewing finished outputs, check quality as work moves through the process:

• Automated validation before outputs reach users

• Human checkpoints at high-stakes decision points

• Rejection before bad outputs cause damage

Governed capabilities: least privilege

Each AI component accesses only what it needs. No broad permissions. No “just in case” access. The architecture limits exposure.

Mastercard requires product owners to complete a scorecard before any AI system is built or a contract is signed.8 Their philosophy: “If risk might be present, treat it as though it is. And if you’re wrong, you’ll get that in control verification, because the team will give you evidence they mitigated the risk.”

Operation: As You Run

The human stays accountable for outputs. When AI augments a human, the human owns the result. This isn’t about blame. It’s about ensuring someone is paying attention.

MIT Sloan and BCG’s 2025 research found that 79% of advanced AI adopters invest in AI that augments human judgment, while only 54% invest in fully automated “AI decides and implements” systems.9 The companies seeing results keep humans in the loop.

Reliability over time

AI systems drift. Models degrade. Data distributions shift. What worked in month one may fail in month six.

Governance by design includes monitoring for reliability:

• Track output quality over time, not just at launch

• Establish thresholds that trigger review

• Build retraining and adjustment into the operating model

• Document what “good” looks like so you can detect when it changes

Mastercard’s AI governance team, which grew from one person to five, manages a portfolio of AI systems that doubled annually. They accomplish this through systematic monitoring, not heroic individual effort.

Everyone owns governance within their scope

Both ING and Mastercard used the exact same phrase, unprompted: “Governance as part of everyone’s job.”

This approach is the opposite of a governance committee. When governance is everyone’s job:

• You don’t wait for committee approval to make obvious decisions

• Risk assessment is distributed, not centralized

• The person closest to the work makes the call within defined boundaries

• Speed increases, not decreases

JoAnn Stonier, Mastercard Fellow: “Perhaps that’s because everybody in the company is more aware and digitally engaged now, and they see governance as part of everyone’s job.”10 Result: they allowed ChatGPT when other companies banned it, and “haven’t had any problems thus far.”

The Diagnostic

One question reveals where you stand:

“Where does governance live in your AI initiative?”

If the answer is “a committee,” “a review process after we build,” or “one person’s responsibility,” you’re bolting governance on. It will slow you down without reducing risk.

If the answer is “in the architecture, the business logic, and the workflow, owned by everyone involved,” you’ve designed governance in. It enables speed because the system itself enforces the boundaries.

Four Questions Before You Build

For your next AI initiative, answer these before development starts:

1. What should AI handle, and what must it never touch? Define the boundaries. Not as guidelines. As architectural constraints.

2. What decisions are deterministic vs. non-deterministic? Pricing, compliance rules, and approval thresholds go in code. The LLM handles interpretation and generation.

3. Where are the quality gates in the workflow? Identify checkpoints where problems get caught. Build them into the process, not after it.

4. Who is accountable for the output? Name a person. Not a committee. Not “the AI team.” A human who owns the result.

If you can’t answer all four, you’re not ready to build.

The Choice

The companies moving fastest on AI took governance most seriously. They didn’t create committees. They made it everyone’s job. They embedded governance in the architecture, not the org chart.

ING deployed to production in seven weeks because they answered the governance questions upfront. The professional services firm that’s been working on a pilot for nearly a year is still trying to answer them after the fact.

Same technology. Opposite approaches. Different outcomes.

The choice is straightforward: bolt governance on after and watch it become theater, or design it in from the start and build something you can actually control.

Apply this framework to your company: ChatGPT | Claude | Perplexity

Next Week

Next issue: Once you’ve designed governance in, you need to actually build the thing. Discipline 5: Design.

Subscribe now

This issue draws from 2024-2025 research by BCG, Gartner, and McKinsey. It’s informed by conversations with executives navigating the same choices, and my own experience leading AI transformation.

We Built the Agent. We Weren’t Capturing the Value.

Last year, my team built an agent to save our account managers’ time. Ten hours per month each, freed from low-value tasks.

We could track the savings.

Halfway through, we realized we couldn’t answer: “Show me on the P&L where that value is.”

We’d created value. We had no plan to capture it.

Subscribe now

The Evaporation Problem

This isn’t just my mistake. Across companies, AI value is evaporating.

Gartner found that teams deploying AI save five hours per person per week.1 That sounds like a win. But most of that time disappears into low-value tasks. The hours are freed. But the value never reaches the P&L.

It dissipates into scattered minutes, absorbed by whatever work expands to fill the space. Nobody planned where that time would go, so it went nowhere.

The pattern is everywhere: 88% of organizations use AI regularly, but only 39% report any EBIT impact.2 Only 4% consistently capture value.3

You can spot who’s going to become one of these statistics. They write the investment thesis after development starts. They estimate time savings with no plan to change behavior. They build ROI decks only at budget renewal time. Finance has never touched the measurement.

Why does this happen?

Three reasons. All connected.

Disconnected from strategy. When AI initiatives aren’t aligned to strategic priorities, you end up inventing metrics to prove value. Custom dashboards. Novel KPIs. Reports leadership doesn’t watch. Align AI to what the company already cares about, and measurement is built in. This is Shift 1.

The wrong owners. When IT or a centralized AI team owns AI initiatives, they’re measured on delivery, not business impact. They ship the agent. They show it works. But they’re not accountable for whether the business captures value from it. That’s the handoff trap from Shift 2. Without business leaders owning AI outcomes, nobody redesigns the work.

Redesign is where value lives. “It can be hard to turn small time savings into meaningful cost savings,” McKinsey explains. “Today’s gen AI systems automate parts of roles, rather than whole jobs, making it difficult for companies to redeploy people freed up by automation.”4

How do you redeploy people from scattered minutes? You don’t. Unless you redesign the work itself.

Savings Don’t Capture Themselves

Every AI initiative needs two plans: one for creating value, one for capturing it. We had the first. We skipped the second.

We should have partnered with sales leadership and revenue operations before we built anything. The first question: What are the strategic priorities for account managers?

If the priority is coverage (maximize accounts per rep), then a 7% time savings becomes a 7% increase in accounts per rep. That’s capture. The model changes.

If the priority is depth (maximize high-touch communication), then revenue enablement coaches account managers on how to use their freed-up time strategically. That’s capture. The behavior changes.

Either way, the CFO can point to it. Revenue per account manager. Accounts per rep. Something real on the P&L.

We saved ten hours a month. We changed nothing about how those hours got used. The savings evaporated because we never decided how to capture them.

What are the 4% doing instead?

Compare my account manager story with Klarna. Same starting point: time savings. Different outcome: $60 million.

The difference wasn’t the technology. It was the capture mechanism, defined before they built anything.

Klarna: Time savings → Headcount5

Klarna’s CEO publicly committed to specific savings targets tied to their IPO. The investment thesis came first. The value driver: time savings in customer service. The capture mechanism: reduced hiring.

Not “we saved X hours.” Instead: “We won’t hire the 700 people we would have needed.” Their AI now handles two-thirds of customer service. Resolution time dropped from 11 minutes to 2 minutes. $60 million in savings, because they decided upfront how value would hit the P&L.

Dow Chemical: Error reduction → Invoice savings6

Their Chief Digital Officer stated upfront: “Even a 1% improvement would mean substantial savings” on billions in shipping spend. That’s an investment thesis. Bounded problem, clear metric.

The value driver: quality (error detection). The capture mechanism: invoice corrections tied to known shipping spend. Their AI analyzes 43,000 shipments and finds anomalies in minutes that used to take weeks. Millions in freight savings, because they knew exactly where it would show up.

DBS Bank: Strategic value → Three defined streams7

DBS developed a value capture framework with McKinsey before scaling. They refused to hide behind “strategic value.” Instead, they defined three streams: revenue, cost savings, and risk avoidance.

$576 million captured in 2024. $768 million projected for 2025. 1,500 AI models in production. They became the first bank to show investors: higher revenue, lower cost-to-serve, and higher ROE from digital customers. Strategic value stopped being a shield. It became a measurable category.

The pattern: Everyone defined the capture mechanism before building. We never defined ours at all.

Operational vs Strategic

Operational value hits the P&L in 3-12 months. Klarna’s $60 million. Dow’s freight savings.

Strategic value builds capability or position over 12-36 months. It may never show up directly.

“This initiative is strategic,” is doing a lot of work in most organizations. Sometimes it’s real. Sometimes it’s a pet project with a nice label.

Here’s how to tell the difference.

Consider a competitive intelligence agent that helps executives make better decisions. Clearly valuable. But try pointing to a line on the P&L.

You could claim a similar system costs $80K/year off the shelf. But that’s a proxy, not a capture mechanism. It won’t show up on the P&L unless you’re actually displacing that $80K system.

So the move with strategic value is to ask, “So what?” Repeatedly.

We want better market and competitive intelligence.

So what?

So our executives can make better strategic decisions.

So what?

So we can increase our revenue per employee.

How will you know if this initiative is successful?

Revenue per employee has too many variables to use as a proxy. Better market intelligence is too subjective. Better strategic decisions can work as a proxy if we document our decisions and track the success rate over time.

That’s the test for real strategic value: can you document what it actually is and how you’ll know if it’s working?

If you can’t answer both, it’s not strategic value. It’s a pet project with a nice label.

Five Questions Before You Build

So how do you make sure you’re building something real?

Klarna, Dow, and DBS all started the same way: with an investment thesis. Not a slide deck. A clear answer to five questions:

1. What is the value driver? Time, revenue, quality, speed, or capability.

2. Operational or strategic? Will it show up on the P&L in 12 months?

3. What’s the metric? Something finance already tracks, or a proxy you’ve defined upfront.

4. What’s the capture mechanism? What has to change for the value to be realized?

5. Who owns capture? A business leader, not the builder.

If you can’t answer these five before you build, you’re not ready to build.

This isn’t about adding process. It’s about having the right conversation before you start, rather than constructing a story after.

What to do next

Pick your most important AI initiative. Answer the five questions.

If you’re writing the investment thesis now, after you’ve already started building, that’s okay. Most companies are. But now you know what’s missing for next time.

If you can’t answer all five, pause. Have the conversation before building further. Bring the gaps to your next leadership meeting.

What changes when you do this consistently? You stop constructing stories after the fact. You design for capture from day one. When the CFO asks, “Show me on the P&L,” you can.

Apply this framework to your company: ChatGPT | Claude | Perplexity

The companies that capture value from AI and those that don’t aren’t running different technology.

The difference: they answer the five questions before they build. They define the capture mechanism. They assign a business leader to own it.

Don’t settle for creating potential value. Capture value.

Next Week

Next issue: Once you’ve defined how value gets captured, you need to manage the risks of getting there. Shift 4: Risk & Governance.

Subscribe now

This issue draws from 2024-2025 research by McKinsey, BCG, and RAND. It’s informed by conversations with executives navigating the same choices, and my own experience leading AI transformation.

The Pattern

If you’re planning to delegate the AI revolution, then good luck to you. - Paul Hudson, CEO of Sanofi.1

The board asks about what you’re doing with AI. The CEO nods. “We’re on it.” Then, in the next leadership meeting: “AI is important. CIO, go figure it out.”

Fast forward, you’re trying to diagnose what went wrong. Why none of your impressive AI pilots make it to production. Why AI isn’t moving the P&L.

If this rings true, you’re in the majority. That’s the problem.

When a new technology emerges, the default is to defer to the CIO so the CEO can stay focused on the business. Here’s the problem: agentic AI is a revolution in how work gets done, disguised as an IT project.

Subscribe now

McKinsey found that 70% of AI initiatives with centralized business ownership reach production.2 Only 30% of those delegated to decentralized IT teams make it. Same technology. Drastically different outcomes. This is the handoff trap. And most companies are stuck in it.

When I spoke with a president at a mid-size professional services firm, this pattern was unmistakable. I asked who at his company owned AI outcomes; the IT director was responsible.

Was this a strategic decision? “Not really a lot of strategic thought or collaboration around what is the best way for us to tackle this. It was more just based on how decisions are made in the company.”

AI is technology, IT handles technology, so IT got it by default.

This led to two separate proposal generation pilots over two years. The first “died on the vine within months” because the assigned leader “doesn’t really have the juice to make it happen.” The second started at the beginning of 2025 and still isn’t finished. “Honestly, we’re still not done with the pilot. I’ve been shocked at how long it’s taken.”

What Delegation Produces

“If you delegate, here’s how it goes... ‘We delegated it to our CDO. We can build it ourselves.’ And then they hope that you will forget you ever asked. It’s corporate lingo, shorthand for ‘We have no idea.’” - Hudson

Who “owns” AI in your business? Who decides which functions to prioritize? Who decides which cross-functional processes, like lead-to-order, get automated? Who says no to low-value use cases?

If the answer is “I’m not sure” or “we have a committee,” you’ll get activity. But what shipped? What changed?

Teams choose use cases because they rock the boat the least, leading to bolt-on automations that work in isolation but deliver little value. Or they get selected for technical interest rather than business impact. Pilots never leave the lab, and you end up with innovation theater.

All of this points to gaps in leadership, not technical know-how.

Nearly all leading companies have engaged C-suites leading AI transformation, while only 8% of laggards do, according to BCG. 3

This is one of the simplest solutions to the pervasive lack of AI ROI: Engage. Dig in. Don’t abdicate your responsibility as CEO, COO, or business leader.

The Delegation Reflex

For too long, non-technical executives have happily deferred tech to the technical folks. You may even take pride in focusing on your area of expertise while letting the tech team do their thing.

The very fact that we call technology-related projects “IT projects” shows how prevalent this mindset is. It’s IT’s job, not mine.

This mentality comes from fear. Fear of being found out as a fraud. Fear that if you don’t know the technology, you’re going to lose your authority.

We’ve been taught to delegate. In my executive MBA program, the approach was straightforward: learn broadly so you can delegate well and ask the right questions. This approach works in well-defined disciplines like finance and marketing.

Delegation works for most things. But AI breaks this pattern.

It doesn’t work because agentic AI is a revolution in how work gets done disguised as an IT project. To get ROI from your AI investments, you must reimagine and redesign your work end to end. It is the competency most correlated with AI ROI according to McKinsey.4

Your Director of IT cannot decide that the sales team needs to change how they sell. The CIO cannot tell operations to change how they handle the supply chain. The CDO cannot tell the head of finance that they need to change their book-closing process.

These are business decisions. Your CIO can’t make them. Your CDO can’t make them. If you own the outcome, you own the decision.

As the professional services firm president put it: “You need to look at what the business is, what the business does, make sure whatever you’re doing is supporting that and not the cart driving the horse.”

RAND found that 84% of AI initiative failures trace back to leadership problems, not technology problems.5 The failures follow the same pattern:

1. Business leaders can’t articulate what they actually need (because they don’t understand AI’s capabilities)

2. Technical staff optimize for the wrong outcomes (because they don’t understand the business context)

3. Miscommunication compounds until the project solves the wrong problem

The result: pilot purgatory.

BCG recommends a different way with their 10-20-70 model. 10% of the focus on data and models. 20% of the focus on the technical infrastructure and tools. The remaining 70% of the focus on the people and processes.

By abdicating to IT, you’re abandoning 70% of the challenge. As BCG puts it, “exclusive IT ownership is a key indicator of stagnating companies.”6

Delegation vs. Ownership

Paul Hudson, the CEO I quoted at the start of this issue, decided to lead Sanofi's AI transformation personally. He chose not to delegate to his CDO, even though “CEOs of his generation delegate.”

As Gilmurray (CEO of KG & Co) says: “To scale successfully, AI needs the same rigor as any other strategic initiative: a clear vision, a business owner who’s accountable for results (ideally an executive), high-quality data and the right success metrics.”7

Hudson decided that, rather than delegating AI to IT and letting them automate a few low-value use cases, Sanofi would transform the value chain end to end.

“I think our competitive advantage will be that we go all in, end-to-end, while others are dabbling, doing proof of concepts, handing the project off to CDOs, rather than understanding its relevance.”

The result: Sanofi was able to look at processes from the customer value point of view, rather than internal siloed processes: “We look at everything end-to-end. Can we go cheaper? Can we go more robust in certain areas? Can we introduce a shop floor agent to improve asset utilization? Can an agent help with drug design and discovery?”

The outcomes?

“Using AI, we reduced out-of-stocks by 80 percent, which is close to a billion euros, and we improved asset utilization by more than ten percentage points.”

Note what Hudson doesn’t do: he doesn’t build agents, attend daily standups, or manage technical teams. He decides what matters and removes roadblocks. That’s ownership. Everything else is delegation done right.

At a €43B company, that means an annual review. At your company, it means more: quarterly reviews, monthly visibility, and direct involvement in prioritization decisions. Your company is smaller than Sanofi. Your involvement needs to be greater. The principle doesn’t change.

The opposite? BCG’s laggard profile (notice the compounding effects of these policies):8

• Spend less on IT

• Spend less of their IT budget on AI

• C-suite not deeply engaged on AI

• Less likely to have shared business-IT ownership on AI

By saying yes to the delegation reflex, Paul would have been in the majority of CEOs today. He would have passed on the opportunity for end-to-end transformation, and Sanofi would be dabbling like most companies, struggling to see real impact from AI.

How to Make the Shift

This shift is uncomfortable. The CEOs who step up and lead their companies through the AI revolution will be the leaders of the great companies in the next decade. That requires humility: asking questions, admitting gaps, and learning.

Embrace the 10-20-70 model. 30% of the effort goes to building the technology. 70% goes to change management: redesigning work, measuring ROI, and getting leadership support for AI initiatives.

The rule: the P&L owner signs off on every use case. AI initiatives don’t move forward without the business leader being accountable for the outcome.

Next, the CEO must own AI outcomes. The CEO should “chair” the weekly, monthly, quarterly AI review. Ask these questions:

• Which initiatives delivered measurable value this quarter?

• Which should be killed?

• What’s blocking the ones that matter?

Follow Hudson’s example: “I ask the key questions. I want to see the innovation, and I want to know how it’s happening in the most critical areas of the company.”

Finally, flip accountability onto the business leaders. Hudson: “I set a challenge for my leaders: Disrupt your function with agentic AI by the end of the year. AI is the one thing that drives disruption.”

Business leaders report on AI outcomes (revenue, cost, speed, quality). IT reports on enablement (infrastructure performance, guardrails adherence, evaluations). This is NOT a CIO demotion. It is role clarity. Your CIO might be the most business-savvy person in the room. The issue isn’t their capability. It’s that AI transformation requires decisions only business line owners can make: which processes to redesign, which functions to prioritize, which sacred cows to kill.

This is what ownership looks like in practice. As the president of the professional services firm put it, “What I would want to do is take it back into operations... Let’s brainstorm and identify the things that give the biggest bang for the buck to the business. We may need to carve out one person’s time completely for a year to lead this. A very strategic person to move things forward.”

The CEO sets the pattern. Delegate to IT, and so will everyone else.

The Ownership Test

When AI comes up at the board meeting, who presents?

• CIO presenting outcomes → you’ve delegated

• Business leader presenting outcomes, CIO presenting enablement → you own it

If you’re unsure, ask three questions:

1. Who’s accountable for AI outcomes? (Name a person, not a committee.)

2. Who decides which initiatives go forward? (Technical feasibility or strategic value?)

3. What does IT report on? (Outcomes or enablement?)

The Conversation to Have This Week

In your next C-suite meeting, ask one question: “Who is accountable for AI outcomes in this company?”

Not IT. Not middle management. A C-level leader who is business-oriented and can report to the board.

What you’re listening for:

• Silence or confusion → ownership is undefined (common, fixable)

• “IT” → you’ve delegated (recognize the pattern)

• “I am” (from CEO/COO/business leader) → you own it (verify with practices)

This isn’t about blame. The instinct to delegate is understandable. The fix is structural: assign a business owner, establish the operating rhythm, let IT enable.

So What

IT enables. Business owns. That’s the shift.

More than 80% of AI projects fail, according to RAND.9 And 84% of those failures trace to leadership, not technology. The handoff trap is structural: delegate to IT, get demos. Own at the C-level, get outcomes.

Hudson’s closing thought:

“CEOs should not underestimate the personal energy required. As people have said, AI may be the greatest revolution since the printing press. If you’re planning to delegate the AI revolution, then good luck to you.”

Apply this framework to your company: ChatGPT | Claude | Perplexity

Next Week

My team built an agent that saved ten hours per month. We tracked the savings. We reported the wins. Then the CFO asked a question we couldn't answer. Shift 3: Value & ROI.

Subscribe now

This issue draws from 2025 research by McKinsey, BCG, and MIT’s analysis of 300+ enterprise AI deployments. It’s informed by conversations with executives navigating the same choices, and my own experience leading AI transformation.

The Pattern

5%

That’s the percentage of custom enterprise AI pilots that reach production, according to MIT’s 2025 analysis of 300+ deployments.1 The remaining 95% never ship. They become demos, science projects, and eventually, budget line items that nobody can explain.

The specific number is debatable. The direction isn’t. Multiple studies confirm that the vast majority of AI pilots never make it to production, and even fewer deliver material business impact.

I can confirm this from my own experience. Over the past year, I’ve reviewed more than 150 AI use cases. The number that reached production? A fraction. The number that delivered measurable impact on the business? Even smaller.

McKinsey’s data tells the same story: More than 80% of organizations see no material contribution to earnings from their generative AI initiatives. Only 1% of executives describe their AI rollouts as “mature.”2

This isn’t a blip. This is the pattern.

A small group of companies are seeing dramatically different results. BCG’s 2025 survey of 1,803 C-suite executives found that leading companies achieve 5x the revenue increases and 3x the cost reductions from AI compared to others.3 Same technology. Same market conditions. Radically different outcomes.

The difference isn’t the AI. It’s the mindset.

Companies that start with “What can AI do?” build demo factories. Companies that begin with “What strategic priority needs solving?” create a competitive advantage.

This is the first of seven shifts that separate companies that win with AI from those that burn resources on pilots that never ship. This shift is foundational. Get it wrong, and nothing else matters.

Subscribe now

FOMO vs. Focus

The pattern of failure is remarkably consistent, and it starts with fear.

Boards ask, “What’s our AI strategy?” and teams scramble to show activity. The fear of falling behind drives a flurry of pilots. Every interesting use case gets pursued because saying no feels like missing out.

BCG found that companies struggling with AI spread themselves thin across an average of 6.1 use cases. Leaders, by contrast, focus on just 3.5. That focus translates to 2.1x greater ROI.

The math is simple but counterintuitive: doing less with AI produces more impact.

A Deloitte survey captured this: “Everyone is asking their organization to adopt AI, even if they don’t know what the output is. There is so much hype that I think companies are expecting it just magically to solve everything.”4 Activity isn’t progress. Pilots aren’t products. Demos aren’t deployments.

The result is predictable: resources scatter, nothing reaches production, executives lose faith, and the cycle repeats with the next technology wave.

Why Smart People Keep Making This Mistake

The fundamental error is putting the cart before the horse.

As Harvard Business Review put it in September 2025:

“When companies lead with AI or treat it as the answer, they put the cart before the horse and risk compromising their company’s strategy. But when companies take the opposite approach, starting with strategy, identifying how they can offer buyers a leap in value, and then looking to technology as a tool to deliver that leap, AI can be a powerful catalyst.”5

The technology is genuinely exciting. It’s fun to experiment with. You can suddenly do things that were impossible with traditional automation. The demos are impressive. The possibilities feel endless.

But impressive demos and endless possibilities don’t translate to business impact. They translate to scattered resources and pilot purgatory.

The difference is straightforward:

The first approach feels productive. The second approach produces results.

The Incrementalism Symptom

When FOMO drives your AI strategy, there’s a predictable byproduct: incrementalism.

When you’re leading with AI instead of strategy, you’re looking at your existing processes and asking, “Where can I bolt on some automation?” That’s a comfortable question. It doesn’t require rethinking anything fundamental. A chatbot here. An automation there. Minor improvements to existing workflows.

McKinsey’s 2025 research tested 25 attributes to determine what drives AI success. They found that workflow redesign has the most significant impact on an organization’s ability to see the earnings impact of AI. Not the model selection. Not the technology stack. The redesign of how work actually gets done.

Yet only 21% of organizations have fundamentally redesigned workflows as they deploy AI.6

Some executives already understand the difference. Earlier this year, I asked a COO at a mid-size company: What would success in AI look like over the next year? His answer:

“Saving hours and increasing speed on existing processes is great. But if we can completely transform a function, the way they work, with AI, so they are an order of magnitude more efficient or effective. That would be success.”

Transformation is hard. It requires critical thinking about process design, engaging people across functions, and real change management. It’s more complicated than saying, “I can automate this little step.”

But that’s where the value lives.

BCG found that leading companies allocate more than 80% of their AI investments to reshaping key functions and inventing new offerings.7 They’re not looking for 10% efficiency gains. They’re asking: “If we could redesign this function from scratch with AI as a core capability, what would it look like?”

That question requires focus. And focus is the antidote to FOMO.

The Focus Advantage

BCG’s leading companies don’t just focus on fewer use cases. They concentrate their investments in areas that reshape how the business works. They’re selective about where AI can deliver breakthroughs, and they invest deeply in those areas rather than spreading themselves thin across low-impact use cases.

The result: 5x the revenue increases and 3x the cost reductions.8

When you start from strategy and work down to AI applications, you naturally involve senior leadership. The CEO cares about strategic priorities. When you frame AI as a lever for those priorities, it gets executive attention, resources, and cross-functional support.

That’s a setup for success. We’ll dig deeper into the ownership dynamics in next week’s issue, but it starts here, with strategic fit.

What Focus Looks Like in Practice

I learned this lesson firsthand.

When I started leading our AI journey, the approach was bottom-up: “Let’s hear from everybody where they think AI could work. Where in their jobs do they see opportunities for AI agents, use cases, automation?”

What happened was predictable in hindsight. We were flooded with ideas. Hundreds of use cases, most of them low-value. And to keep people engaged, you have to respond. You have to explore each idea at least. Otherwise, you’re ignoring the people you asked to contribute.

We ended up spread thin, exploring many possibilities while making progress on very few.

Then something changed. The CEO identified a strategic imperative: we needed to build an agent because it was mission-critical for achieving a key priority. Suddenly, everything opened up. We had the resources, executive support, and cross-functional engagement. Everything we needed to be successful.

We built a highly capable agent in two months. Our scattered efforts had produced nothing. The difference wasn’t the technology. It was the focus.

The Framework: Strategy-First AI

How do you move from FOMO to focus? Start with strategy, not with AI.

Here’s a framework that works. It’s inspired by Richard Rumelt’s constraint-based strategy from “The Crux,” adapted for AI decisions.

Step 1: What are your strategic priorities?

This sounds obvious. It isn’t.

If you don’t have clear strategic priorities, don’t start with AI. The technology will not clarify your strategy. It will amplify your confusion.

Before any AI initiative, you should be able to point to a specific strategic priority from your existing plan. Not “innovation” or “digital transformation.” A concrete priority: Expand into the small business market. Reduce customer acquisition cost. Accelerate the product development cycle.

If you can’t name the priority, stop. Get clear on strategy first.

Step 2: What makes achieving that priority difficult?

Every strategic priority has constraints. Bottlenecks. Points where progress stalls.

Say your priority is expanding down-market into the small business segment. You’re used to enterprise customers where you invest heavily in proposals, spend significant time one-on-one, and deliver high-touch service. The constraint is scale: How do you serve a higher volume of smaller customers without your cost structure making it unprofitable?

Identify the constraint. The crux. The thing that, if solved, would create a breakthrough on the priority.

Step 3: Can AI materially impact that constraint?

This is where AI enters the conversation. Not before.

Ask three questions about the constraint you’ve identified:

• Speed: Can AI dramatically accelerate this? Not 10% faster. Not 50% faster. 10x faster.

• Cost: Can AI significantly reduce the operational cost? Not marginal savings. Economics that change the math.

• Capability: Can AI enable something previously impossible? New offerings. New markets. New ways of serving customers.

If the answer is yes to any of these, AI might be part of the solution. If the answer is no to all three, AI probably isn’t the right tool for this constraint. And that’s fine. Not every strategic priority needs AI.

This doesn’t mean zero experimentation. It means experimentation in service of a strategic hypothesis, not experimentation as a substitute for strategy.

Step 4: Technical feasibility (last, not first)

Only after you’ve established business viability do you ask about technical feasibility.

Can AI actually do this today? Do we have the data? Is the technology mature enough?

Too many companies start here. They ask, “What can AI do?” and work backward to identify problems. This produces impressive demos that solve problems nobody has.

The sequence matters. Strategy first. Constraints second. AI potential, third. Technical feasibility last.

If the answer is “I don’t know” or “not yet,” that’s okay. Partner with an expert. If it’s not possible today, revisit as the technology evolves. What’s impossible today may be straightforward in six months.

What This Means for You

Self-Assessment

Pull up your list of current AI initiatives. For each one, answer:

1. Which strategic priority does this serve? (Name the specific priority.)

2. What constraint does this address? (Name the bottleneck.)

3. What’s the speed, cost, or capability impact? (Quantify it.)

If you can’t answer these clearly, the initiative probably shouldn’t exist.

The Filter in Action

I’ve watched executives apply this filter and kill most of their AI initiatives. Not because the technology didn’t work. Because the initiatives didn’t connect to the strategy.

That sounds brutal. It’s actually liberating. The resources scattered across a dozen pilots concentrate on two or three that actually matter. The teams that were spread thin get focused on outcomes that executives care about.

The counterintuitive result: fewer AI initiatives, more AI impact.

The Stakes

The companies extracting real value from AI aren’t doing more. They’re doing less, more deliberately. They start with strategy, not AI. They focus investments on a few transformational use cases rather than spreading thin. They redesign workflows end-to-end rather than bolting AI onto existing processes.

Start here, and you’re on your way to the top 5% seeing material impact from AI.

Apply this framework to your company: ChatGPT | Claude | Perplexity

Next week

Shift 2: Ownership & Operating Model

“If you’re planning to delegate the AI revolution, then good luck to you.” That’s Sanofi CEO Paul Hudson. He’s not wrong. McKinsey found that 70% of AI initiatives with business ownership reach production. Only 30% of those handed to IT make it. Same technology. Drastically different outcomes...

Subscribe now

Software that takes action. Processes that run themselves. This is the intelligent organization. Executives know it matters. And almost none of them are getting there.

The research is brutal: 95% of organizations are seeing no return on their AI investments. Only 1% describe rollouts as mature. Over 40% of agentic AI projects will be canceled by 2027 due to “unclear business value.”

The gap isn’t technology. It’s thinking.

This newsletter is about seven shifts that close that gap:

1. Strategic Fit: From “AI is strategic because everyone says so” → To AI as a selective lever, only where it moves the needle

2. Ownership: From “IT owns AI” → To business leaders owning outcomes, technologists enabling

3. Value: From “AI ROI is fuzzy” → To every initiative having an ROI hypothesis and measurement plan

4. Risk: From “AI risk is a scary fog” → To guardrails, evaluations, and incident playbooks

5. Architecture: From “AI lives in isolated tools” → To a capability layer you reuse everywhere

6. Delivery: From “We’re exploring AI” → To a cycle that ships production use cases, not theater

7. People: From “AI replaces people” → To role and story redesign that creates leverage, not stress

Make these shifts, and you’re building an intelligent organization with advantages that compound. Skip them, and you’re running expensive pilot factories.

One shift per week. Every Tuesday morning. 10-15 minute reads built on research, hands-on experience, and conversations with leaders doing this work.

No hype. No tool worship. No vendor pitches. Just the thinking that moves AI from pilot to production.

First issue (January 6): AI Is Not the Strategy

Your intelligent organization won’t just happen. Subscribe to start building it.

Subscribe now