Ona

The trait that decides it is absence. Work happens somewhere else, so your laptop is not warm, your checkout is not touched, and you find out how it went by reading a pull request like any other. For anyone who has watched an agent chew through a local repository for forty minutes, that is a better relationship than supervision.

The trade is that you cannot steer mid-flight, and you pay for the compute whether or not the result is useful. Pick it for well-specified tasks. Pick Devin if track record matters more.

The direction of travel is the problem. Documentation names one agent as the recommendation for new sessions and records the previous harness as deprecated for the largest customers. An organisation that built prompts, evaluations and habits around that earlier path is migrating on somebody else's schedule, and nothing published commits to a second option remaining.

What it does right is separation. Work happens in an environment created for the task, so a failed run damages nothing a developer was holding, and there is no local state to reconcile.

1. The architecture answers the question most autonomous products dodge, which is how the agent knows it succeeded. A full development environment means the build, the test suite and the dependencies are present, so correctness is demonstrated by running things rather than inferred from a diff.

2. That also makes the run reproducible for a reviewer, since the environment is defined rather than incidental. 3. No benchmark is published, so throughput and success rate remain unstated, and the architectural argument stands alone.

A development-environment company repositioning as an agent runner and then landing inside a model lab's platform is the cleanest exit narrative on this board. The asset being bought was never the agent; it was years of work on provisioning reproducible environments, which is precisely the infrastructure a lab needs and cannot rush.

Pricing power now belongs to the parent, and the parent has an obvious incentive to make this the default place its own agent runs. Likely path: deeper integration, thinner independence. Position: buy it, and expect the roadmap to serve the platform rather than your workflow.

The enterprise checklist is genuinely answered: deployment inside our own network boundary, policy enforced below the application layer, and audit trails that exist without me asking. That is more than most of this category offers and it is why this gets a meeting.

The cost is the meter. An entry seat is twenty dollars and the included allowance spans a twenty-seven-fold range, with top-ups sold in blocks, so sixty developers could be a modest bill or an alarming one depending on behaviour nobody can forecast. Approved with conditions: a contractual spend cap and per-team unit reporting from day one.

There is nothing to read and nothing to fork; the licence is proprietary and execution never touches my hardware, so this fails my first test before we start. Two things stop me writing it off entirely. It consumes MCP servers, so tools I already built are reachable from inside somebody else's runner, and a subscription I already pay for can be connected rather than duplicated.

That is a rental with a spare key, not ownership. Grudging respect for letting me bring the subscription.