Two Weeks to Pack: What You Really Lose When Your AI Coding Tool Is Sold
On Wednesday, 100,000 developers learned that the app they use to run AI coding agents had been bought by Meta, and that its cloud would close on 20 November. The code is safe. The way their teams work might not be.

On this page 5 sections
Wednesday morning
Picture Maya. She’s made up, but you probably know someone like her. She leads a small engineering team, and for the past six months her team has worked a little differently from most.
Every evening they hand a list of jobs to a set of AI coding agents. One writes missing tests, two chase bugs from the support queue, another tidies an old payments module nobody wants to touch.
By morning there’s a stack of proposed changes. The team reads them over coffee, keeps the good ones, and sends the rest back with notes.
The agents run in Conductor, an app that lets one developer run several agents side by side, each in its own copy of the code. On Wednesday Maya opened her laptop and read that Conductor had been bought by Meta.
The rest of the message was the part that mattered. Conductor Cloud, the hosted version her team’s overnight runs depend on, would go read-only on 13 November and shut on 20 November, with stored data deleted after that. The Mac app would keep working until early 2027.
Maya has two weeks to pack.
Her code is fine. That isn’t the problem.
Her first reaction is relief. The code lives in Git, like all code should. Every change the agents proposed and the team accepted is sitting safely in the company’s repository.
Then she starts listing what else her team built over those six months, and the relief fades.
In your repository
- Source code and tests
- Commit history
- Pull requests and review comments
- Build and deployment scripts
Often only in the tool
- The instructions each agent works from, tuned over months
- Task templates for recurring jobs
- Run history: what each agent tried, and why
- Which person approved which change
- Workspace and model settings
A typical set-up for teams running agents through a hosted tool. Your own split depends on how you’ve configured it.
None of those things is code, and all of them are the work.
The instructions are where the team wrote down what “good” looks like for the payments module. The run history is how they learned which jobs agents handle well. The approval trail is what they’d show an auditor.
That last point matters more each month. When Stack Overflow published its 2026 survey of more than 30,000 developers, its chief executive said this:
“As agents take on more work, developers need to know where an answer came from.”
Knowing where an answer came from is easy while the record exists. It gets much harder once the record has been deleted.
This will keep happening
It would be easy to blame Conductor, and unfair. Small, fast-moving companies build useful tools, and large companies buy them. That’s how this market works, and it’s speeding up.
The same day, OpenAI agreed to buy Pierre, which builds code storage for machine-written code. Its tools already help OpenAI’s coding agent, Codex, browse files and show what changed.
The biggest AI companies are racing to own the tools developers use every day. The fastest way to win that race is to buy them.
- Software delivery
Meta confirms it has bought Conductor. The same day, OpenAI agrees to acquire Pierre.
- World
Teams that run agents in Conductor Cloud start working out what they need to move.
- Deadline
Conductor Cloud becomes read-only. No new runs.
- Deadline
Conductor Cloud shuts, and stored data is deleted.
- Deadline
The Conductor Mac app stops being supported.
Sources: SiliconANGLE and Investing.com, 4 November 2026. Full links below.
So the useful question isn’t whether your favourite agent tool will be bought. It’s what happens to your team on the day it is.
Your code is in Git. Your way of working might not be.
Packing light
The teams that will barely notice the Conductor news have one thing in common. They treated the agent tool as something that runs the work, not as the place where the work is kept.
In practice that means a few unglamorous habits:
- Keep agent instructions in the repository. Write them as plain files next to the code they describe, reviewed like any other change. Many teams already do this with files such as AGENTS.md, which most coding agents can read.
- Treat settings as code. Model choices, task templates and limits belong in version control, not in a web dashboard.
- Copy the run history somewhere you own. A nightly export of what each agent did, and who approved it, is cheap insurance.
- Hold your own model accounts and keys, so changing the tool doesn’t mean changing the contract.
- Keep a second way to run the work, even a slower one, and try it once a quarter.
Ask one question in your next team meeting: if our agent tool closed in two weeks, what would we lose that isn’t in Git? Write the answer down. That list is your migration plan, and most of it can be fixed in an afternoon.
If you want a head start, we’ve turned this into a working pack: the inventory, a five-day plan, templates and the questions to ask your next vendor.
Free toolkit
Agent Tool Exit Plan
A five-day plan for moving your AI agent workflow when a tool is sold, shut down or replaced.
- The two-week test: six questions that show your exposure
- Inventory of 14 artefacts, with where they hide and how to export them
- Triage grid to decide what to move first
- Five-day migration plan with owners and “done when” checks
- AGENTS.md starter, run-log and approval-record templates
- 16 questions to ask before you choose the next tool
- Contract terms worth asking for
- Quarterly fire drill and sign-off sheet
Back to Wednesday
Imagine two versions of Maya’s team.
In the first, the instructions, templates and approval records live inside Conductor’s cloud. Her next two weeks go on exporting what she can, rebuilding what she can’t, and explaining to her manager why the overnight runs have stopped.
In the second, all of that has been sitting in the repository from the start. On Wednesday afternoon she points the same files at a different runner, checks that the tests still pass, and goes home on time.
Same team, same agents, same news. The difference was decided months earlier, by where they chose to write down how they work.
Key takeaways
- Meta’s purchase of Conductor gave 100,000 users two weeks before its cloud closes on 20 November.
- Code in Git is safe. Agent instructions, templates, run history and approval records often aren’t.
- Big AI companies are buying developer tools, so more of these deadlines are coming.
- Keep instructions and settings in the repository, export run history, and hold your own model keys.
- Run the two-week test: list what you’d lose that isn’t in Git, then fix it.
Case studies & examples
References & further reading
- Meta acquires AI coding startup Conductor, whose app runs several coding agents at onceSiliconANGLE, 4 November 2026
- OpenAI to acquire Pierre to expand Codex developer infrastructureInvesting.com via Yahoo Finance, 4 November 2026
- Stack Overflow’s 2026 Developer Survey: trust remains conditionalStack Overflow, 6 October 2026
- AGENTS.md: a simple, open format for guiding coding agentsagents.md
Links checked on 5 November 2026. External sources are listed for further reading; Delphilens is not affiliated with them.