AI Insights
AI creates speed, not automatic value. Your operating model decides whether that speed becomes business impact.
This page pulls together my most essential perspectives on AI transformation. Below, I’ll share how AI productivity works, how to recognize and escape "AI Theater", and how to design your organization so that AI speed actually translates to business outcomes.
AI creates speed, but speed is uneven. When you accelerate one part of a system—like generating code or drafting copy—without addressing the downstream review, deployment, and adoption constraints, you build inventory and queue time rather than actual business value.
In many organizations, adding AI looks like massive productivity. Teams are writing code faster, drafting reports in seconds, and spawning marketing copy by the gigabyte. Yet, the end-to-end delivery times to customers remain virtually unchanged. The reason is a simple system-level mismatch.
AI productivity only matters when it improves the core constraint of your value stream. If your bottleneck is code review, security approval, or customer adoption, making engineers write code faster just creates a large backlog of unreviewed code. It adds inventory and congestion, not value. To move from activity to impact, you must identify your system constraint and aim AI there.
AI Theater is the illusion of progress measured by activity metrics like seats allocated, prompt libraries built, or pilots launched. Escaping it requires shifting focus from outputs to outcomes, tracking system cycle times, and measuring sponsor-ready evidence of actual business value.
It is easy to get caught up in the excitement of new models and fancy agent demos. Organizations build committees, distribute prompt swipe files, and run hackathons. They measure success by the number of tool licenses purchased or prompts submitted. This is AI Theater—it creates the appearance of progress without impacting top-line or bottom-line metrics.
To escape the theater, you must start measuring what matters. Stop tracking tool adoption as a success metric and start tracking cycle time, value-added time, customer adoption, and actual business outcomes. AI adoption should be treated as a product adoption exercise where teams pull in tools to solve concrete constraints, rather than pushing tools by executive mandate.
While writing code (the inner loop) has accelerated by orders of magnitude, the outer loop—including code reviews, security checks, integration, and release—remains slow. This mismatch creates acceleration whiplash, where developers build code faster than the organization can absorb, review, or validate it.
AI tools like GitHub Copilot have dramatically sped up the developer inner loop: writing, testing, and debugging code on a local machine. But code is not value until it is running in production and serving customers. The outer loop—the process of getting code from a developer machine to production—is where the real constraints live.
When the inner loop accelerates but the outer loop remains manual and slow, you experience "acceleration whiplash." Pull requests grow larger, code review queues back up, quality becomes fragile, and engineers get frustrated. Shifting from local developer productivity to system-level flow is the only way to resolve this bottleneck.
Realizing AI value requires treating internal AI contexts as products, governing in the flow of value, and funding initiatives in stages as confidence increases. Through Organizational AI Coaching, we redesign workflows to ensure AI productivity actually impacts the end-to-end constraint.
Value realization is not about building multi-year AI roadmaps or creating large, centralized governance gates. In high-uncertainty environments, certainty is built through evidence. We steer AI investments by managing a portfolio of small bets, using staged funding to release budget only when confidence indicators are met.
This requires an operating model change. We treat internal AI context—such as prompts, tools, and fine-tuned models—with the same discipline we apply to consumer-facing products. We establish clear ownership, build telemetry, and redesign workflows to ensure that AI capabilities are pulled where they actually resolve bottlenecks.
The absolute Bible on economic-based product decision rules, queue management, and batch-size optimization. Essential for understanding how speed in code creation can lead to queues if downstream flow is neglected.
A brilliant work on overcoming "resistance"—the mental blocks that prevent us from finishing and shipping our work. Highly relevant as we shift our focus from starting AI experiments to actually delivering value to production.
A practical guide to building organizational discipline and execution focus. Moving from AI pilots to business impact requires operational focus and accountability systems like those mapped out in Traction.
AI Theater is when an organization measures AI success by activity—such as seats allocated, prompts typed, or demos built—rather than actual business outcomes. It creates the appearance of progress without driving revenue, cost savings, or cycle-time improvements.
Measure AI value by tracking system-level outcomes: end-to-end cycle time, release frequency, change failure rates, and customer value indicators. Avoid local productivity metrics like lines of code generated, which ignore downstream review and testing bottlenecks.
Because value is constrained by the slowest part of the end-to-end stream. If engineering speeds up but product discovery, code reviews, deployment, or customer adoption remain slow, you simply pile up work-in-progress (WIP) and create longer queues, not faster value delivery.
Insights Database
Browse my complete archive of perspectives and diagnostic articles on moving AI investments from activity to impact.
AI work creates value when it starts with a real business constraint, not a tool wishlist. Before building the agent, run a small test, look for evidence, and make sure the problem is worth solving.
Most AI efforts start with tools and demos. A better move is to use product discovery and Lean Startup habits to aim AI at a real business constraint, test the riskiest assumption, and learn before you burn.
AI accelerates team delivery, but it also collapses procedural complexity. As team-level dependencies thin out, the mechanics of scaling frameworks become optional—but first principles of flow and managing WIP become absolute.
When AI coding increases the arrival rate of pull requests, asking reviewers to work faster is the wrong response. Use end-to-end feature flow, spec-driven development, and the Theory of Constraints to improve reviewability and business throughput.
AI isn't killing Scrum, but it is shrinking the team size from cross-functional squads to highly leveraged 1-3 person units. Here is how team-level descaling triggers a fractal collapse of organizational complexity.
Why the rise of Forward Deployed Engineering is the next agility problem — and how to scale "unreasonable agility" without it eating itself.
Spec-driven development looks like a step backward if you think of it as requirements theater. But the better frame is that the spec is becoming a higher-level programming language for human intent.
Claude's new /goal feature lets AI agents keep working until a completion condition is met. Every canonical example they show is about output: tests pass, code compiles, backlog empties. What's conspicuously absent is outcome. Here's why that gap matters and what it will take to close it.
AI coding tools can make engineering dramatically faster. That does not automatically make the business faster. The constraint often moves to deciding what is worth building, reviewing safely, getting adoption, and proving impact.
Most AI efforts are still stuck in personal productivity. The interesting shift starts when AI moves from something individuals use around the edges of the process to something that changes the process itself. This is what that shift actually looks like inside a Fortune 500 from the CTO's chair.
If your AI effort has plenty of pilots, training, tools, and output but still not enough business traction, the problem may not be AI. It may be that AI is improving the wrong part of the system.
Spec-driven development can be a smarter way to steer AI coding agents, or it can become waterfall with more tokens. The difference is whether the spec creates learning or pretends learning is no longer needed.
Moving from chatting with AI to unleashing agentic AI isn't just a productivity boost—it's a fundamental shift in leadership. Why the future of work isn't about human-in-the-loop, but humans moving up a level.
AI Vibe Coding is useless and frustrating when its bound by the constraints of the wrong ecosystem. It is exhilirating and high-impact when unleashed by an organizational operating system that's optimized for "building with AI"
A deep dive into the conversation with Dave West at Scrum.org about the 95% AI failure rate and why organizations must evolve their management operating system.
95% of AI projects fail — not because the models are wrong, but because organizations run high-uncertainty AI work through a project-mode operating system. Here is what changes when you treat AI as product work instead.
AI initiatives die in the proof-of-concept stage when they are run as fixed-scope projects. Clemens Adolphs and I unpack why AI work needs real product discovery, internal market validation, and outcome-based steering instead of a gantt chart with the labels filed off.
Why scattering AI initiatives without identifying bottlenecks wastes effort — and how a Theory of Constraints lens helps you aim GenAI at the right problems before you automate.
Adopting AI (or Agile, or OKRs) out of FOMO almost always turns into JOMO — the Joy of Missing Out. A five-question framework for connecting any initiative to your most expensive problems before you commit.
AI context — system prompts, knowledge bases, and tool configurations — is a product to be continuously discovered and improved. Applying product thinking to AI enablement.
AI tools promise to accelerate product discovery — but do they? A critical look at where AI actually helps in discovery, and where the bottleneck remains human judgment.
Practical patterns for using AI across the product development flywheel — from discovery and validation to faster build cycles and better feedback loops.
Every major tech wave has a cart-before-the-horse phase. What product-led AI looks like instead: starting with customer outcomes, embedding cross-functional ownership, and exploring problems before solutions.
GenAI transformation is a classic VUCA challenge — high uncertainty, fast-moving, multi-disciplinary. How organizations can use agile operating principles to navigate AI investments without drowning in hype.