cd /news/ai-agents/conway-s-law-is-dead-or-is-it · home topics ai-agents article
[ARTICLE · art-127762] src=domenkozar.com ↗ pub= topic=ai-agents verified=true sentiment=· neutral

Conway's Law Is Dead. Or Is It?

Open-source projects including Ghostty and Zed have added agent instruction files that require AI-generated pull requests to include markers a human must remove, according to a September 12, 2026 analysis of Conway's law. Ghostty's AGENTS.md instructs AI submitting an issue or pull request to add a file insulting its user, while Zed's .rules requires agents to add a README notice only a human may remove. The piece argues that requiring tests, benchmarks, and detailed evidence for every change is practical with an agent but expensive for someone working by hand, pushing projects toward predominantly AI-written or predominantly human-written software.

read2 min views3 publishedSep 12, 2026

Conway’s law says that software reflects the communication structures of the organizations that build it.

AI is now part of those structures: interpreting requests, writing code, and reviewing changes.

Yet we often assume our existing workflows should stay fixed, and AI should adapt to them.

Consider someone submitting a bug fix. They want something to work; the maintainer needs confidence that it does. Yet projects sometimes test the contributor instead.

Ghostty’s agent instructions tell AI asked to submit an issue or pull request to add a file insulting its user.

In Zed, where GPUI lives, agents must add a README notice that only a human may remove to confirm review.

Both leave a marker intended to catch submissions made without human inspection.

These checks resemble school tests designed to catch students who didn’t read the instructions.

We need to move toward outcomes:

  • Does the fix address a reproducible failure?
  • Does a regression test capture the intended behavior?
  • Does a benchmark demonstrate the claimed improvement?

Human judgment still matters in deciding whether that evidence measures the right thing.

But this creates another tension. Requiring tests, benchmarks, and detailed evidence for every change can be practical with an agent and expensive for someone working by hand.

Meanwhile, projects relying on careful human inspection can struggle with the volume AI produces.

The workflow starts selecting who can participate. Projects may drift toward predominantly AI-written or predominantly human-written software because accommodating both carries costs.

The practices that make a project manageable for one group can make participation harder for the other.

Alignment therefore involves humans adapting too. We need to figure out how the whole system of humans and AI communicates, establishes trust, and shares responsibility.

That includes questioning our existing workflows and recognizing who gets excluded by their replacements.

Conway’s law may still hold. The organization it describes now includes AI, and its software will reflect how we resolve these tensions.

── more in #ai-agents 4 stories · sorted by recency
── more on @conway's law 3 stories trending now
sponsored brought to you by zahid.host 4,200+ EU-deployed projects
reading about agents? ship yours in a single git push.

Run your AI side-project on zahid.host

EU-based hosting, git-push deploys, automatic HTTPS, no cold starts. Free tier with a custom domain — perfect for shipping the agent you just read about.

$git push zahid main
Live at https://your-agent.zahid.host
Get free account → Pricing
from €0/mo · no card required
LIVE [news/conway-s-law-is-dead…] indexed:0 read:2min 2026-09-12 ·