# I used to run a 150-person engineering org. Now agents open most of the PRs on my project. What carried over and what didn't — retained record

> Retained reddit record from reddit:AI_Agents.

Canonical: https://highsignal.app/data/records/8a9bf47c373263e4

[← sources](https://highsignal.app/data)


source: [reddit](https://highsignal.app/data/reddit) (reddit:AI_Agents)  published 2026-10-26 13:31 ingested 2026-10-27 02:35

[original source ↗](https://www.reddit.com/r/AI_Agents/comments/1xfwwpw/i_used_to_run_a_150person_engineering_org_now/)

## retained text



```
For context: I co-built an engineering org at a services company from 30 people to 150. This year I've been running a client project where coding agents do most of the work. Ten weeks in: 1,575 PRs, 1,363 of them opened by agents.

I expected the management part to shrink. It mostly moved.

What carried over:

- Tickets. A vague ticket used to get sorted out in a conversation. Now a separate step compares the diff against the acceptance criteria, so the ticket is the spec and the test at the same time. Writing them well is where most of my time goes.
- Nobody approves their own work. Same rule for agents: the one that wrote the code never marks it done. One step reads CI, one compares the diff with the criteria, one clicks through staging and takes a screenshot per criterion. Any of them can send the card back.
- WIP limits. We run up to 15 stories at once. Same as with people, the limit is how much the humans can follow, not how many workers you have.
- Blockers have to be visible. Anything that needs a person goes to a Blocked column instead of waiting quietly. Basically the standup, without the standup.

What didn't:

- Budget per head. A year ago one developer on my project spent $1,000 on agents in a week, and at the time it looked like an outlier. Now we look at cost per finished story: $36 at the median, but one in ten goes over $147, and that tail is what I actually watch.
- Code review as the main gate. Review is about 11% of our agent spend, and people don't read every diff anymore. What people own is the requirements and the final sign-off with the client.
- Growing people. I grew a few seniors into team leads and that was the best part of the job. An agent doesn't get better from feedback, only its instructions do. I haven't figured out what replaces that for the humans on the team.

For those of you running agents on real projects: what did you stop doing, and what did you have to start?
```



document fetched 2026-10-27 00:19 · canonical https://reddit.com/r/AI_Agents/comments/1xfwwpw/i_used_to_run_a_150person_engineering_org_now
