Insights Topic
You can buy AI capability. You cannot buy AI impact. That gets decided by where you aim it, how you fund it, and what you are willing to stop.
This is the leadership side of my AI work: where the money is actually leaking, which bets are worth making, and how to steer them without drowning in pilots. If your question is what changes inside the delivery lifecycle once agents are writing the code, that is the companion topic below.
Start here · Cornerstone read
Most AI projects fail not because the models are wrong, but because the management operating system isn't built for high-uncertainty work. How agility principles change the odds.
Read the full articleThe usual explanation is that models are immature or data is not clean. In most organizations I work with, that is not the binding constraint. The work is being run in traditional project mode, with fixed scope and a business case written when nobody knew anything yet. That is the wrong control system for high-uncertainty work.
The second pattern is that speed landed in one isolated silo while the constraint remained elsewhere. Teams generate prototypes and copy faster, but deciding what is worth building, getting it reviewed, and getting somebody to change their behavior did not accelerate. Speed pushed into an unchanged constraint produces inventory, not impact.
AI theater has a tell: every number on the slide is an input. Licenses bought, workshops held, prompts run, hackathons completed, pilots launched. None of those say whether a customer, a cost line, or a cycle time moved.
The step almost nobody takes is moving past individual personal productivity into actual workflow redesign. When you orchestrate multi-step agentic workflows around real operational constraints (keeping humans in the loop for key reviews), the work itself operates fundamentally differently.
Stop brainstorming random use cases. Go find where work is already leaking value: the handoffs people quietly work around, the queue everyone apologizes for, the decisions that wait on one person, the shared services that tax the rest of the company.
The earliest enterprise value often hides in internal operations, like Legal, Procurement, Finance, and IT Ops. Forward-thinking organizations establish an **AI Use Case Factory**: a disciplined discovery mechanism that qualifies candidate bets by operational friction and business consequence before committing build budget.
Most AI portfolios do not have an idea shortage. They have a WIP problem. Forty proofs of concept in flight, nothing finished, nobody able to say which ones are working, and no kill criteria anywhere.
Applying **product thinking at the portfolio level** means treating AI bets as discovery options, not upfront capital projects. You fund small discovery batches to derisk value and feasibility, set strict WIP limits on active bets, and make decisive stop-or-scale decisions based on real evidence rather than executive enthusiasm.
As Shay Mandel highlighted, *"We are developers of the system, not the code."* Spec-Driven Development is not just for software developers; it is the operating system for knowledge work across the company.
When you write specifications down (for OKR drafting, use case qualification, content engines, or shared services), how your company operates becomes explicit, version-controlled, and executable. Leaders and teams stop being workers trapped in the system and become the designers developing the company as a product.
Insights Database
Browse the complete archive of articles and case studies related to ai activity to impact.
Next Insurance broke its whole product development lifecycle into agent skills and deliberately kept sprints, roadmaps, Jira and the role split. Why authorship is the first thing to change in an agentic rollout, and how you know when the process change is finally due.
Shay Mandel's teams turned every step of product development into a skill an agent runs, then told two hundred people to stop writing the artifacts and start fixing the thing that writes them. His account of what worked, what the agents get wrong, and what they deliberately did not change.
AI will flatten parts of the organization, but it won't remove accountability. The real shift is that more of the company's operating logic becomes encoded, observable, and continuously adjustable, and someone still has to own it.
Why I pivoted from Scaling with Agility to Scaling AI from Activity to Impact. Agility is the "Powered by Intel" inside the AI-Native organization, and the lessons from scaling it are what turn AI ambition into impact.
Every enterprise is arguing about the right per-employee AI spend cap. Tomer Elias and I compared notes on why that number is unanswerable until you can attribute the spend to something.
AI work pays off when it starts from a real business constraint, not a tool wishlist. Run a small test and check the problem is worth solving.
Most AI efforts start with tools and demos. Better to use product discovery to aim AI at a real business constraint and test the riskiest assumption first.
AI coding can make engineering dramatically faster. The business does not follow automatically: the constraint moves to deciding what is worth building.
Most AI efforts are still stuck in personal productivity. Zoetis CTO Kumar Venugopal on what it takes for AI to change the process itself.
Plenty of pilots, training, tools, and output but not enough business traction? The problem may not be AI. AI may be improving the wrong part of the system.
Moving from chatting with AI to unleashing agentic AI is not just a productivity boost. It changes leadership work: humans move up a level.
AI vibe coding gets frustrating inside the wrong operating system. It gets traction when the organization can turn experiments into business results.
Most AI portfolios suffer from the same disease: too many pilots, too many ideas, not enough kill criteria. Portfolio flow beats AI idea theater.
A conversation with Dave West at Scrum.org about the 95% AI failure rate and why organizations need a different management operating system for AI work.
95% of AI projects fail: not because the models are wrong, but because organizations run high-uncertainty AI work in project mode.
AI initiatives die in the proof-of-concept stage when run as fixed-scope projects. Clemens Adolphs and I unpack why AI work needs real product discovery.
Scattering AI initiatives without finding your bottleneck wastes effort. A Theory of Constraints lens helps you aim GenAI at the right problem first.
Adopting AI, Agile, or OKRs out of FOMO usually ends in JOMO. Five questions that tie any initiative to your most expensive problems first.
AI context (system prompts, knowledge bases, tool configs) is a product you keep discovering and improving. Product thinking applied to AI enablement.
Every tech wave has a cart-before-the-horse phase. What product-led AI looks like instead: start with customer outcomes and explore problems before solutions.
GenAI is a classic VUCA challenge: high uncertainty, fast-moving, multi-disciplinary. How to steer AI investments without drowning in the hype.
