Lukman Nuriakhmetov
ENRU
ПостИнженерное лидерство

The engineering job is moving from writing software to building the system that writes it.

Agentic code-review numbers show why a software factory is becoming urgent — and why the missing piece is the operator layer between stages, not the model.

Lukman Nuriakhmetov
Lukman Nuriakhmetov
1 мин чтения · 19 июня 2026 г.

I have been quietly building a software factory for my own work — and I think this is how most engineering will run in a few years.

The framing is having a moment: Factory 2.0 and others are pitching "software factories," and this week's agentic code-review numbers show why it is urgent — one report found time spent in code review up over 440%, and pull requests merged with no review at all up 31%, once agents started generating at volume. More generation without a system around it does not speed a team up. It buries it.

The loop I keep coming back to is concrete: operator sets context → an AI planner turns it into a plan and writes the docs → the plan becomes queued tasks → an AI producer writes the code → it lands in git → AI review and tests → human review → delivery. Then it runs again until the thing is done. Most of that is buildable today. What is missing is not the model — it is the operator layer that moves work between stages without a human copy-pasting between tools.

That layer is the part I have been building. A task system that is AI-facing as much as human-facing. Orchestration that triggers the producer and routes the output. Approval boundaries so the automated path cannot quietly exceed its authority.

The future of engineering is not writing more code. It is owning the line that produces and accepts it, and within controlled and structured boundaries.

Теги: engineering-leadership · ai-engineering · systems-thinking · software-factory

Связанные тексты

Пост3 августа 2026 г.

AI shortened how long it takes to build a feature — not how long it takes to learn from one.

37signals cut its cool-down along with its build time. Delivery compresses; customer absorption and operational recovery mostly do not, and the ratio between them shifts.

Engineering leadership
EN1 мин
Пост8 июня 2026 г.

AI made code cheaper. It made judgment more expensive.

As agents handle more of the implementation, domain expertise becomes the scarce skill that decides whether the output is actually correct.

Engineering leadership
EN1 мин
Заметка24 августа 2026 г.

Модель перестала быть интересной переменной

Неделя 17 августа 2026. Mistral утроила точность на финансовой отчётности — с 26,7% до 86% — изменив то, как модель ищет, а не какая модель ищет. Две разные модели из двух разных лабораторий получили примерно одинаковый трёхкратный прирост от одной и той же обвязки. Тот же узор прошёл через неделю: авторизация распалась на четыре отдельных вопроса, оптимизация токенов сократила токены и подняла суммарную стоимость в полтора раза, пропускная способность обогнала окружающую её ёмкость ревью, а исполнение переехало в общие командные каналы. Рычаг раз за разом оказывался вне модели. Скан-слой TL;DR, пять тем каждая с operator move, одна разобрана глубоко, три контр-сигнала и что отслеживать.

Разбор недели
12 мин