Skip to content

Blog

What faster code generation does to your delivery, with sources.

Definitions, research and field notes on agentic engineering, each with its sources and limits.

  • Every article and guide cites its sources

Breaking the Loop: Why agents never finish on their own

Agents in a software factory always find another problem, another ticket, another improvement. Only a second, much slower loop can say when the work is finished: the one that reaches the market.

Specification and review · 24 October 2026 · 22 min read

Start here

  1. 01 Coding agents are an organizational-design problem 6 min read
  2. 02 Why faster coding does not guarantee faster delivery 4 min read
  3. 03 Verification in a Software Factory 3 min read
  4. 04 Software Factory evidence and its limits 3 min read
  5. 05 What is a Software Factory? 3 min read

Factory skills

An 18-part series, one part for each automation in a running Software Factory, from writing the backlog to reporting on the work.

  1. 01 The Feature Planner writes its own backlog, behind a human gate 2 min read
  2. 02 Cross user feedback against usage before it becomes work 2 min read
  3. 03 Fix the incident and the rule in the same commit 2 min read
  4. 04 An improvement automation needs a cap on its ambition 2 min read
  5. 05 The Feature Builder checks for open bugs before it builds 2 min read
  6. 06 A bug-fix automation must never exit quietly 2 min read
  7. 07 Daily metrics that cannot drift: subtract yesterday's total 2 min read
  8. 08 An auto-merging PR reviewer is safe once it knows when to stop 2 min read
  9. 09 Green CI does not prove the live site works 2 min read
  10. 10 Verify UI components in the page they ship in 2 min read
  11. 11 The PR Shepherd resolves only the conflicts it understands 2 min read
  12. 12 Set the stale-work clock from your agents' cycle time 2 min read
  13. 13 Escalated work needs a return path back onto the line 2 min read
  14. 14 An automation that audits the factory must not rewrite its own record 2 min read
  15. 15 A metric without a red line is decoration 2 min read
  16. 16 An automated status report is only as honest as its prompt 2 min read
  17. 17 Report the daily delta, not the total that only goes up 2 min read
  18. 18 A label on an issue is the whole protocol between agents 2 min read

Guides

Each guide answers one question, drawing on several posts from practice and the research behind them.

  1. Measuring AI impact on engineering: what to count and what to ignore Code volume, pull requests, usage: AI pushes up every number that is easy to count. A Stanford talk and DX data put the measured gain at around 10 percent, and it swings widely from company to company. What to measure instead, where the returns come from, and why things get worse before they get better. 13 min read Measuring AI impact

Topics

Follow