AIToday
Large Language ModelsAI Coding AssistantsZenn AI/MLPublished: Oct 10, 2026, 10:00 JST

dots + CompanyBrain: the developer merges the PR

dots + CompanyBrain: the developer merges the PR

A Zenn essay describes a workflow where you write rough specs into CompanyBrain, hand the goal to OpenAI's resident agent dots, and just review and merge the pull requests it raises.

Not sure about something? Ask the AI

Questions and answers are published on this page.

Summaries like this, in your inbox every morning.

Context & Analysis

The essay frames the change as a shift from prompting an AI to handing it a standing goal. Earlier AI worked as a single exchange; dots keeps running in the background and holds its own cloud environment, and specialist dots for roles like accounting, invoicing and customer support are starting to appear for companies.

The author ties this to a CompanyBrain concept from an earlier essay: write decisions, policies and common assumptions as Markdown in one place and link them to GitHub Issue and PR history so AI and people read the same context. For a resident agent, the biggest obstacle is said to be the cost of re-explaining context each time, and CompanyBrain is presented as the fix.

On automation, the author describes a loop where the agent opens a pull request, CI passes, and GA4 metrics — error rates, drop-offs, conversions after deployment — act as one safety valve for auto-merge, deployment and rollback. The stated limits are to automate only as far as tests protect, keep permissions narrow, avoid too much parallelism, and keep rollback available.

FAQ
What is dots?
It is a resident AI agent announced by OpenAI. Given a goal once, it keeps working in the background, has its own cloud environment per agent, reaches out to thousands of apps, and reports back at checkpoints.
When does the author allow agents to merge and deploy on their own?
In repositories where tests and monitoring can be trusted. The agent opens a PR, CI passes, then GA4 numbers decide whether Auto Merge and deployment proceed, with rollback if the metrics worsen.
How does the author keep agents from going out of control?
Automate only within what tests cover, limit the apps and data agents can touch, avoid running too many tasks at once, and always keep rollback possible. Being able to undo is what makes delegation safe.

AI news that matters for your work, delivered every morning.

Pick your industry and the AI tools you use, and get news related to your work every day.

Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →

Ask AI

Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.

Questions and answers are published on this page.

Related Articles

Next articleUkraine drones knock out Yandex AI data center