Ideas, observations and analysis without the fluff
Two years ago I could have told you exactly what my engineers were working on. Jira knew, because one person worked on one thing. That assumption broke, and the number I got for free went away without anything appearing to break.
The unit of work allocation is no longer the person. It is a thread of attention. Engineers now hold more situational awareness than management, so scheduling must be delegated down. Tooling becomes a cockpit, experience becomes a cliff, and steering attention toward strategy is the open problem.
My median approval-to-merge is 9 minutes. Open-to-first-review is 16 hours. The machine produces in hours; the human queue costs a day. Software factories failed for sixty years because there was nothing to instrument. Agents changed the physics. Andon, takt, and pull, running on a real line.
You read an argument that would have moved you, but the person making it is wrong about something else, and you find yourself unable to take it seriously. You haven't evaluated it. You've evaluated the source. This happens to you constantly, and you mostly do not notice.
The agent cannot be paged. You can. Agents unbundled authorship from accountability, and the part of authorship that mattered was never the keystrokes. It was the reading. It was the willingness to defend what shipped under your name. That part needs a name now. Call it self-review.
Practices that were nice-to-have for humans become load-bearing for agents. Most organizations won't fail at agentic development. They'll buy licenses, get some wins, and plateau. The playbook for the next plateau: four layers, a method, and the human work that decides whether any of it lands.
Six months ago I wrote that a flood was coming. It's here. We tripled deployment rate in one month. Stakeholders can't produce enough work for us. We built an unbelievably powerful engine that isn't connected to the wheels. Transform the rest of the business, or accept the plateau.
We run three backend languages. That was never a goal — it's what happens when pragmatic choices accumulate over time. Nobody sat down and decided we needed Scala, Python, and TypeScript. Each one arrived for defensible reasons, and each one stayed because the cost of leaving was higher
Most people building with AI are still the orchestrator — the thing holding their own tools together. The hoist is the moment you realize those tools aren't for you. They're for an agent. Stop being the agent. Become the arbiter.
The prevailing narrative goes something like this: agents are getting better, fast. Soon they'll plan, execute, test, and deploy. The engineer becomes an observer — maybe a light editor — and the system runs itself. Just give the agent the goal and get out of the way. It&
The economic protection of software was always reproduction cost. AI collapses that cost. Any business whose moat depended on software being expensive to reproduce is now structurally exposed.
Atlassian won’t vanish overnight—but inertia isn’t a moat. In an agentic world where software is cheap and specs replace tickets, workflow tools built for human task tracking face structural risk. The need for coordination remains. The shape of it doesn’t.
The taxonomy describes roles. Products of Consequence, Judgment, Counsel, Action. Four categories, four different jobs. But products don't exist in isolation. They compose into systems. A Product of Counsel consumes signals from Products of Judgment. A Product of Consequence orchestrates the whole stack, embedding Counsel, consuming Judgment, and
Agentic dev breaks planning. Work finishes before tickets get written. Estimates don't apply. But orgs still need dates. The fix: plan outcomes, not tasks. Artifact in the repo where agents read it. Ranges, not commitments. Most orgs aren't ready. Figuring it out live.
Self-awareness is a form of self-knowledge that emerges when you know your strengths as well as your weaknesses. A self-aware leader knows when he can succeed as well as when he's licked. Such a leader has the strength of character to step aside in favor of a better alternative.