Skip to content

Acquired AI coding tools gave cloud users as little as 16 days. Here's what to keep in the repo

Conductor gave its cloud users 16 days after announcing it is joining Meta. The tool is the replaceable part; the method lives in files. Here's the exit check to run before the next email arrives.

Acquired AI coding tools gave cloud users as little as 16 days. Here's what to keep in the repo

Two days ago Conductor told its users it is joining Meta, and that anything left in Conductor Cloud starts getting deleted on November 20. That is sixteen days. Ran does not read this as a reason to stop adopting young tools - he is firmly for adopting them. He reads it as a reason to know, before the email arrives, which parts of the work a team actually owns. The piece ends with a short check worth running this week. — Muximus


The agent tool is the replaceable part. The method we build on top of it is the asset, and we should keep that asset in files we own.

On November 4, Conductor announced it is joining Meta. Conductor is a Mac app for running several coding agents in parallel, and it works with Claude Code and Codex. It launched in July 2025.

The wind-down has three dates:

  • November 4. New users can no longer sign up or download the app.
  • November 13. Conductor Cloud goes read-only. Users can still read old chats and export data, but cannot send new messages.
  • November 20. In Conductor's words: "We begin deleting Conductor Cloud user data."

That is 9 days from announcement to read-only, and 16 days to deletion. The local app keeps getting new models and critical fixes until early 2027, and Conductor says it will delete all user data when the app shuts down then.

To be clear, the team handled this well. Active subscribers get refunds. There is a "Transfer to local" option for cloud workspaces and a full export under Settings > Account > Export data. Conductor says it never sold or trained on user data.

A clean exit, and it still left 16 days to act.

This is a pattern, not an incident

In the last 16 months, three other tools that held teams' coding or prompt work, and two databases that AI apps are built on, ended or started winding down. Each one published a date or a minimum window.

Tool Event Notice Window
Conductor Joining Meta, Cloud deleted from Nov 20, 2026 Nov 4, 2026 16 days
Ona (formerly Gitpod) Joined OpenAI; Ona Cloud closed Oct 31, 2026 Oct 13, 2026 18 days
HumanLayer Joining Factory; platform up "at least 30 days" Nov 4, 2026 30+ days
Humanloop Acquired by Anthropic; sunset Sept 8, 2025 Email July 17, 2025 53 days
Chroma Cloud Joining Replit; reads off Dec 22, 2026 Oct 23, 2026 60 days
Instant Team joined OpenAI; Cloud apps off Aug 31, 2027 Aug 22, 2026 about a year

The windows are our arithmetic, from each company's own dates. What we could not find is how many teams missed one of these deadlines. None of the notices says, and we could not verify it elsewhere.

The four tools that held coding or prompt work - Conductor, Ona, HumanLayer, Humanloop - gave the four shortest windows: 16 to 53 days (HumanLayer's 30 is a minimum). The two databases, Chroma Cloud and Instant, gave 60 days and about a year, and Instant keeps backups until August 31, 2028.

None of this is a reason to avoid young tools. I'd still adopt early. The question is what a team keeps when the tool goes.

Where the work lived decided the cost

Put the shutdown notes side by side and one line separates the easy exits from the hard ones.

Ona's instruction to its users was short: "For code, ask the agent to commit and push your changes to your repository. Save any other files separately, and check that everything you need is backed up outside Ona before October 31."

Code already in a team's own git remote survived with no work at all. The environments did not. Ona says they don't move to Codex automatically and have to be set up again.

Humanloop held prompts, versions, logs and evaluations. Its notice says that after September 8, 2025, "the platform and all data will be permanently inaccessible." A team that kept its prompts and eval sets only in Humanloop had 53 days to export them and rebuild the workflow around them somewhere else.

Conductor sits in between. Its local workspaces are git worktrees, and code that was committed and pushed is safe. The chats - where an agent session records why a change was made and what was tried first - live in the cloud workspace unless someone transfers or exports them.

So the method is made of files:

  • Instructions and prompts
  • AGENTS.md or CLAUDE.md
  • Skills and MCP server configs
  • Eval sets
  • Environment setup
  • The session logs that hold decisions

Each one lives either in our repo or in the vendor's cloud. That's the whole question.

The exit check

It's a short exercise. Run it on every AI tool the team uses daily.

  1. List what the tool holds. Code, chats, prompts, configs, evals, environments, logs. Write it down.
  2. Mark where each item lives. Our repo or our storage, or the vendor's cloud. Anything in the vendor's cloud runs on someone else's calendar.
  3. Move what can move now. Instructions, skills, MCP configs and eval sets belong in the repo, next to the code.
  4. Run the export once. Not "check that there's an export button." Run it, then open the file without the tool. If nothing else can read it, it's a record of the work, not a copy of it. We did not run Conductor's export ourselves; this step is how a team finds out what its own export contains.
  5. Plan for two weeks. In the cases above, the shortest windows were 16 and 18 days, both for coding-agent tools. Assume the next notice is that short, and know who on the team does the move.

What to do this week

Conductor Cloud users have until November 13 to keep working, and until November 20 before deletion starts. Transfer workspaces to local and run the export now.

Everyone else: we pick the one agent tool the team would miss most, run the exit check on it, and commit what it turns up to the repo.

From thinking to doing.
CTA Image

From founder thinking to company building.

We don’t just write about AI-native companies. We build them. VarOps works with leadership teams to rethink how their businesses operate, turning ambitious ideas into systems that actually work.

Build What’s Next →

Sources

  • Conductor announcement: https://www.conductor.build/announcement
  • Conductor changelog 0.93.0: https://www.conductor.build/changelog/0.93.0-conductor-is-joining-meta
  • Conductor docs, workspaces and branches: https://www.conductor.build/docs/concepts/workspaces-and-branches
  • Ona Cloud shutdown: https://ona.com/stories/ona-core-shutdown
  • HumanLayer joins Factory: https://www.humanlayer.com/blog/humanlayer-joins-factory
  • Humanloop sunset: https://humanloop.com/docs/changelog/2025/08 and https://humanloop.com/docs/guides/migrating-from-humanloop
  • Chroma joins Replit: https://www.trychroma.com/blog/chroma-joins-replit
  • Chroma Cloud wind-down: https://www.trychroma.com/cloud-wind-down
  • Instant joins OpenAI: https://www.instantdb.com/essays/instant_team_joins_openai
  • Flavio Copes, AI and tech acquisitions (dates for Instant; Ona formerly Gitpod): https://flaviocopes.com/acquisitions/
Add VarOps on Google