Devin's next commit

Bessemer is proud to announce our investment in Cognition, the company behind Devin. For the last three years, the story of AI has primarily been a story about assistance. Autocomplete got better, chat got better, the IDE got smarter. Every one of those advances made developers faster inside a workflow that barely changed.

Cognition built something radically different: self-driving software. Devin runs in its own cloud environment, takes a written spec, and works for minutes or hours, planning, writing, testing, and returning code ready for review. The unit of work for Devin isn’t a suggestion or a token. It’s a ticket. You hand Devin a task the way you would hand one to your team, and you come back later with the complete unit of work. This is the whole thesis, not a feature difference.

Why we’re backing Cognition now

Two years ago, autonomous coding agents demoed well and shipped poorly. Then, capability caught up. Frontier models got dramatically better at multi-step reasoning and tool use, open source models flourished, and Cognition had already spent two years on the unglamorous scaffolding that turns a capable model into a reliable worker: sandboxed cloud execution, multi-repo context, parallel child sessions, model routing, policy and audit controls.

Devin now does production work for the largest and the most regulated engineering organizations on earth. Mercedes-Benz has publicly described finishing a code modernization in eight days that would have taken eight months. Goldman Sachs, Santander, and Nubank are also known users of the platform.

Beyond the customer base, three factors convinced us to bet on Cognition:

1. Autonomy is an architecture. Running an agent in a managed cloud VM instead of on a laptop is not a convenience—it’s what makes delegation possible and what makes an agent deployable inside a bank. Isolation, sandboxing policies, admin and audit controls, and the ability to run without an engineer babysitting the session makes this production ready.
2. The enterprise is the harder path and the bigger prize. Cognition sells top-down to the world's largest enterprises, anchors on measurable projects, like a modernization, a migration, a large-scale refactor, a new product, and then it expands. Behind that motion sits an FDE organization that scopes the right work and clears the security and governance gates.
3. Model agnostic → best, optimized performance. Devin is built as a harness that runs on any frontier or open-weight model, with Cognition's own SWE model line as one option in the same router. For a CIO thinking about vendor concentration, cost control, and a model landscape that reorders itself every few months, that flexibility is not a nice-to-have. Cognition gets better every time the underlying models do.

Most importantly, Cognition is a team that rivals the avengers. It was founded by top technical leaders in the category, and two years and several hundred hires later, the bar has only been raised. The founders are all International Olympiad in Informatics (IOI) gold medalists, and the first ten people at the company held ten IOI golds between them. This is the kind of team most companies assemble once and then dilute on the way up. Cognition has been hiring S tier like none other.

What gets built when software reaches L5 autonomy

For as long as software has existed, what got built was governed by how many engineers you could hire. Every roadmap is a rationing exercise. Every CIO carries a backlog that was scoped, costed, and quietly shelved: modernizations deferred a decade, integrations that would have required a nine-figure services engagement, internal tools that never cleared the ROI bar at human labor rates. None of that work was unimportant. It was unaffordable.

If you lift the constraint, the question changes. Not, "What can we afford to build?" but "What should exist?" That’s the market—and it's not a better editor. It’s what Cognition calls "self-driving software development."

We’re thrilled to partner with Scott, Steven, Walden, Russell, Zach, and the entire Cognition team.