AI agents are landing this year. No question. Every company is somewhere in the middle of an AI transformation.
But what's actually happening looks less like organizational transformation and more like individual experimentation. A horse-drawn train, no playbook.
Individual Wins
30x exists. People who have settled their own stack of agents and skills, running personalized automation.
But these wins are isolated. The reported "50% AI adoption rate" mostly comes from scattered usage — someone learned to use a coding agent for code review or auto-create, a PR description with an AI-generated paragraph here and there. All counted.
The train's motion is still horsepower. Individual, scattered, around the edges.
What the Org Is Missing
Estimation, scope, hand-off, decision rights, performance review — all still designed around the old cost model.
So you see a lot of misalignment.
- The fast ones have stable agents and skills adapted to their workflow, running heavy automation. The slow ones are still on auto-complete.
- Some still use the old productivity tooling to judge "this is 1.5 months as a stretch goal," while an AI-native engineer says "3 days" and gets pushed back as "you're underestimating complexity."
- Hiring has partially updated to evaluate AI efficiency, but only as a junior-level pilot.
- Performance review adds a line asking if you've been using AI well — asked, that's the end of it.
Org productivity is bound by the weak link. The more important the position of the weak link, the bigger the negative impact.
Transformation Needs Dedicated Time and Effort
Transformation takes time — for individuals and for organizations.
But many capable people are too busy, so a gap opens in their individual transformation. And these are exactly the people whose past excellence makes their drag larger (echoing my earlier busy paradox post).
A Company's AI Operating Model Transformation Is System Engineering
- Reset the cost model. Let AI-native speed anchor estimation. Retire the old baseline.
- Redistribute decision rights. Give the people who use AI well influence beyond their level and tenure. This violates the corporate default — but during AI-native transformation, cognition alignment matters 10x more than hierarchy alignment.
- Build surfacing mechanisms for real usage. Task-based benchmarks for actual usage, so placebo usage gets exposed.
- External force can help (e.g. FDE), but you need internal champions to sustain real change.
- No rush to lay off — when measurement is broken, how do you know you're not laying off the actual AI-native engineers?
CHUG CHUG CHUG CHUG CHUG * clop * CHUG CHUG CHUG CHUG CHUG CHUG