AI will not rescue a broken operating model. It will reveal it.
I have led responsible AI adoption inside a growing engineering organization.
I am excited about AI adoption and the possibilities it creates. Since the push really started in 2023, I have seen just about every reaction to it, from people who could not wait to jump in to those who were skeptical, hesitant, or resistant. What I have learned along the way is that how you approach adoption, and the leadership around it, can make a big difference in how teams respond and what they ultimately get from the technology.
The pattern I kept seeing was pretty simple. Teams that had time to learn, room to collaborate, clear guardrails, practical examples, visible work, fast feedback, clear ownership, and human review got better with AI over time. Teams without those things often just got faster at producing more work, sometimes work nobody had asked for and sometimes with more risk attached. Same tools. Same talented people. Very different outcomes.
That is also why I have spent so much time thinking about the ROI and actual impact of AI adoption. Adoption and utilization are interesting, but they are not the outcome. I want to understand whether AI is actually improving how work flows, helping teams solve the right problems, improving quality or experience, and ultimately creating measurable value for the business.
The leaders who create the most value with AI understand people, systems, flow, and change. They resist the temptation to confuse more activity with more value, and they invest in the smallest sustainable system rather than the smallest team.
AI is the capacity multiplier. It accelerates whatever system it is placed in, which means it amplifies shared understanding when the system is healthy and confusion when it is not. From Profitable Engineering (opens in new tab)
This thesis began as an article: AI Is a Multiplier.
How adoption worked
This point of view comes from operating experience, not observation. Engineering and Product leaders helped shape the practices required to make responsible use workable in daily delivery, from AI literacy programs built through real use to evaluation guardrails.
Adoption was measured at the system level, through flow and realized outcomes, not individual output. The result was improved delivery effectiveness while supporting responsible, scalable use across the organization.
How I approach AI adoption
My approach is practical, and it runs on a simple rule: match the level of AI autonomy to the level of consequence. Six steps, in order.
-
Start with the problem and the outcome
Map the need, workflow, dependencies, constraints, and economics, then establish measurable customer, business, delivery, or operational success.
-
Build AI literacy through real use
Give leaders and teams practical exposure and shared learning.
-
Decide where AI belongs and who is accountable
Allocate work among humans, deterministic systems, assistants, collaborators, and agents, with clear stewardship for outcome, data, risk, and escalation.
-
Match autonomy to consequence
Higher consequence requires stronger controls, clearer human accountability, and more evidence.
-
Design for verification
Build correctness, security, regression, architecture, policy, operational, and outcome checks into the workflow.
-
Measure the system and earn greater autonomy
Measure Flow, Realization, and cost per outcome. Increase autonomy only after evidence demonstrates quality, reliability, security, and realized value.
Own the assumptions, own the outcome
Accountability does not transfer to the tool. Everyone who uses AI owns what they create with it: the analysis, the report, the article, the code. That ownership starts before the prompt, with knowing the assumptions behind what you expect AI to produce, and it continues after the output, with standing behind the result as your own work. When the work is challenged, AI will not be called into court, the person who created it will.
AI changes who performs the work
The role of experienced engineers is expanding, not shrinking. AI reduces task effort, but it does not remove technical judgment, product context, quality thinking, operational ownership, or accountability. Experienced people become more important because they know how to guide, review, orchestrate, and validate what AI helps create. AI may shrink parts of the team, but it cannot shrink the work of judgment, learning, validation, and ownership.
I have written in depth about the work that remains as teams change and how AI reshapes development teams.
The real question
I am optimistic about AI. I have been using it since February 2023 and have seen significant gains in the recent wave of frontier models and agents. I can already see how it is reshaping how modern organizations work.
But I am also practical. The organizations that benefit most will be the ones with the strongest systems, clearest priorities, healthiest teams, and most disciplined path from idea to outcome.
Is your system strong enough for AI to make you better, or will it simply make your existing problems move faster?
Let’s talk
If you’re navigating AI adoption at the enterprise level, I’m happy to compare notes.