3 agent harnesses sold this year. Only 1 price is public.

So a sale hands your team's work to a new owner. Each of these three ended differently for the people inside.

One runs on a clock. One goes open source. One reprices toward its owner's models.

If your team runs coding agents, here is what each exit did, and a test for your own. The fastest clock is Conductor's.

Three tools, three endings

Conductor: sixteen days

Conductor's announcement is blunt: "New users can no longer sign up for Conductor or download the app."

Then comes the clock. On Friday, November 13, "Conductor Cloud workspaces go read-only. You can still read old chats and export data, but you can’t send new messages."

On Friday, November 20, "Conductor Cloud shuts down." The same page: "We begin deleting Conductor Cloud user data. Please export your data before this date."

November 4 to November 20 is sixteen days.

Check it yourself. Open conductor.build/announcement. Find the line that says "Please export your data before this date."

The app on your machine lasts longer, with "new models and critical bug fixes until early 2027," Conductor says. Then: "We’ll delete all Conductor user data when the app shuts down in early 2027."

Paying users get something back. "Active subscribers will get a refund," and "pro users can keep using pro features for free." Free users "can’t upgrade to a pro plan."

Meta confirmed the deal to TechCrunch but "wouldn’t comment further." So no price.

What did Meta get? Conductor's agents "come from major labs like Anthropic and OpenAI," TechCrunch wrote. And Meta already had its own: "In August, the company launched Muse Code, its own coding agent."

Meta closed Conductor to new users on day one. You don't do that to a revenue line you want to grow. Hypothesis: Meta bought the team and the way developers work, not the sales.

HumanLayer: open source and an invitation

Factory's post gives the plan in one line: "we will be open sourcing HumanLayer and inviting all HumanLayer users to Factory."

It gave no price and no exact user count. It praises "a devoted developer community with hundreds of daily active users."

It also talks up work "especially for larger teams with hundreds or thousands of engineers."

So HumanLayer's users get the code, once Factory opens it, and an invitation to move. Our read: Factory wants the users more than the code.

Cursor: sold, then repriced

The deal is done. "Cursor has officially been acquired by SpaceX," Cursor's blog says.

Then the menu changed. SpaceXSI announced The X Subscription, with "the benefits of SuperGrok, Cursor Pro, and X Premium+, previously $90/month combined."

Who paid, and who pays now

Cursor is the only exit with a price. SpaceX's filing pays Cursor's owners in SpaceX Class A stock, "based on an implied equity value of Cursor of $60.0 billion."

SpaceX's own shareholders paid in dilution. The stock "represented a 3.4% dilution" at its IPO valuation, CNBC reported.

The only revenue figure beside the price is Cursor's own. CNBC: "In November, Cursor said it crossed $1 billion in annualized revenue."

SpaceX "has not provided its investors with details on Cursor's customer list, momentum or revenue."

And Cursor sold while it was losing ground. Its share "declined from 41% in June 2025 to about 26% in May," CNBC said, citing Ramp. It added: "Anthropic now controls half of that category."

The included usage leans toward the owner's models. Each plan "includes at least twice as much coding usage with our models as the comparable Cursor plan."

Everyone else's models sit outside the plan. Cursor's pricing docs: "Third-party models aren't included; on Premium and above you can use them at API list price, paid from prepaid credit."

On top of that, "eligible third-party model requests include a Cursor Token Rate of $0.25 per million tokens," the same page says.

One public price: $60.0 billion, beside a revenue number Cursor gave out itself. The other two buyers paid a price nobody has published. We will not guess one.

So what?

Put the three side by side and a pattern shows.

Who is buying. Every buyer has something of its own to sell developers. Meta has Muse Code. SpaceXSI sells "our models." Factory is inviting HumanLayer's users to Factory.

Our read: the buyers did not need another agent. They wanted the place where developers already work.

What an exit does to you. The tool stops taking users and runs on a clock. That was Conductor's sixteen days.

Or it goes open source and you move it yourself. That is HumanLayer.

Or it stays and reprices toward its owner's models. That is Cursor.

This year has one of each. The new owner picks which one you get.

Your way out

A hypothesis, for whoever picks your team's tools. The cost of a sale isn't the seat. It's the plans and history your agents depend on, and whether you can take them with you.

For founders, the same gap is a product: a way out of any harness that a buyer cannot close. We have no data yet on how many teams would pay for that.

The catch

The buyers may be generous. Conductor is refunding its subscribers. Factory is opening HumanLayer's code instead of locking it up.

And the owner's models may be the better deal. If they do the job, "at least twice as much coding usage" leaves users ahead.

Three deals are not a trend, either. Two of them have no price and no public terms.

So the claim is narrow. The buyer may be kind. The buyer still picks the default.

Try this

List every harness your team's coding agents run in.

For each one, write three answers. Whose models does it run? What does it charge for the others?

Then the hard one: how would you get your work out if it stopped taking users?

If your team is on Conductor Cloud, that answer has two dates. New messages stop November 13. Deletion starts November 20, so export before then.

If the third answer is "we can't," start there. Move the plans and specs your agents use into your own repo, where a new owner cannot close them. Which leaves one rule:

The tool is the store. Whoever owns it stocks the shelf.