How's Cognition doing these days?

Is Cognition still growing fast now?

Cognition is still growing very fast today.

Let’s start with the funding. In May 2026, Cognition said it had raised more than $1 billion at a $26 billion post-money valuation. More interestingly, the company also said Devin had reached roughly $492 million in annualized revenue and that enterprise usage had grown more than 10x in 2026.

That is the kind of number we cannot treat like a normal AI demo story anymore.

What makes the growth feel more solid is the customer mix. Cognition now talks about Citi, Goldman Sachs, Mercedes-Benz, Dell, Santander, Elevance, the U.S. Army, the U.S. Navy, and NASA-JPL. That list is important because it is not just startup engineering teams playing with a new coding toy. It points to banks, automakers, government, and large IT-heavy organizations.

Cognition is currently getting pulled into places where engineering time is expensive and codebases are painful. That is a better demand signal than developer excitement alone.

The risk is that the valuation now assumes Cognition can turn that pull into a long-term platform, not just into a very fast year of agent adoption.

If you want more recent data on this point, please see our latest AI code assistant market report.

Is Devin actually doing useful enterprise work these days?

Yes, Devin looks most useful right now in boring enterprise work.

The Mercedes-Benz signal is a good example. In April 2026, Cognition said Mercedes-Benz was deploying Devin and Windsurf across global engineering teams. The concrete pilot was not sexy: COBOL modernization. But that is exactly why it matters. Devin reportedly analyzed more than 200,000 lines of code and cut a project estimated at eight months down to eight days during a four-week pilot.

Cognition has been pointing to similar work elsewhere. It says Itaú uses Devin to fix 70% of security vulnerabilities automatically. Infosys and Cognizant are meant to bring Devin into enterprise delivery work for their own clients. Cognition’s government launch also talks about Devin being used across the U.S. Army, U.S. Navy, Treasury, NASA-JPL, and defense partners.

Devin is winning where the work is repetitive, annoying, and expensive to staff. Legacy migrations, security fixes, code review, platform transitions, and backlog cleanup are the real Devin use cases right now.

Is Cognition becoming more than just Devin now?

Cognition is indeed trying to become the place where companies manage AI software work, not just the company that sells Devin.

You can see that in the product cadence. Devin 2.2 added desktop testing, self-verification, auto-fix, and faster startup in February 2026. Devin in Windsurf then connected local IDE work with cloud Devin sessions. By June 2026, Cognition introduced Devin Desktop, basically turning the Windsurf surface into an agent command center.

That matters because the market is moving quickly. If coding agents become a feature inside every IDE, Cognition needs a bigger job than “write code when asked.” The company is now building around review, codebase Q&A, DeepWiki, APIs, sessions, cost tracking, enterprise controls, and local-to-cloud handoff. That looks much more like an operating layer for agentic engineering.

The small but interesting signal is in the docs. Recent Devin release notes mention GitHub Enterprise Server and GitLab PR support, pre-approved testing, organization-member APIs, automations, and enterprise IP access list endpoints. Those are actual features you build when procurement, security, and engineering leadership are starting to care.

If you want more recent data on this point, please see our latest AI code assistant market report.

Is Cognition getting more serious about ROI now?

Cognition is getting much more serious about proving Devin pays for itself, and that is probably one of the best recent signals.

In June 2026, Cognition published a detailed method for estimating Devin’s “productive engineering hours.” The company said it trained the estimator on 258 Devin sessions from 126 users across enterprise customers, and that the model reached an rlog of 0.74 on held-out sessions. That is not a perfect productivity measure, but it shows Cognition understands the buyer conversation is changing.

The stronger move came one week later: Cognition introduced an AI Productivity Guarantee. The promise is simple. If Devin delivers less engineering value than the customer pays for, Cognition funds usage credits until it does, up to $10 million. This is not just marketing. It changes the sales conversation from “trust our agent” to “measure the work and hold us to it.”

Cognition knows the easy AI budget phase is ending. CFOs are going to ask what these agents actually save. A vendor willing to put credits behind that answer is either confident, aggressive, or both. For enterprise buyers, this is more reassuring than another SWE-bench number.

Is Devin getting cheaper now?

Yes, Devin is getting easier to adopt. Also, Cognition is becoming more careful about what it gives away.

The pricing change in April 2026 was a real signal. Cognition moved from the old self-serve entry point that started at $500 per month for teams to a broader setup with Free, Pro, Max, Teams, and Enterprise. Pro starts at $20 per month, Max at $200 per month, and Teams has an $80 monthly minimum. That is a big funnel change.

But Cognition also said products like Ask Devin, DeepWiki, and Devin Review would no longer be free because they consume real model capacity. That small sentence matters. It tells us Cognition is learning where the compute cost sits. Coding agents do not just answer once. Instead, they search, test, review, retry, and sometimes run long sessions. That can get expensive quickly.

So it looks like Cognition is widening the top of the funnel while protecting margins on heavy usage. That is exactly what we would expect from a product moving from early scarcity pricing into enterprise scale.

Is Cognition still ahead of Cursor, GitHub, Anthropic, and OpenAI now?

Difficult to say. Cognition still looks strong, but it is no longer alone in the “real coding agent” conversation.

Cursor is probably the clearest pressure point. In May 2026, Cursor said Gartner named it a Leader in the 2026 Magic Quadrant for Enterprise AI Coding Agents, with more than 70% of the Fortune 500 using Cursor. GitHub is also moving fast: recent Copilot coding-agent updates include model selection, self-review, security scanning, custom agents, and CLI handoff. OpenAI is now pushing Codex as an agent for planning, refactors, reviews, and releases. Anthropic said in June 2026 that more than 80% of code merged into Anthropic’s own codebase in May was authored by Claude.

That does not make Cognition weak. Rather, it means the moat is shifting. The raw ability to generate, edit, and test code is becoming less rare every month. Cognition’s better angle is that Devin was built from the start around long-running work, task ownership, review loops, and enterprise deployment.

A June 2026 AIDev paper found that 46.41% of fixes proposed by agents including Copilot, Devin, Cursor, and Claude were rejected. Another AIDev study found Devin had the only consistent positive trend in acceptance rate over 32 weeks, but also showed that no single agent won across all task types.

That is the right way to think about this market: it is not one winner magically solving software. Instead, it’s more like a messy workflow market where task selection, review quality, and integration matter a lot.

If you want more recent data on this point, please see our latest AI code assistant market report.

Is the Windsurf deal working out for Cognition now?

The Windsurf deal looks useful for Cognition’s product, even if the human side of the integration looked rough.

Yes, it’s a bit messy. After Cognition acquired Windsurf in July 2025, reports said it laid off 30 newly acquired employees and offered buyouts to roughly 200 remaining Windsurf staff. Several reports also described a much more intense work culture than many Windsurf employees expected.

For a normal company, that would be a major culture warning. For a frontier AI startup, it still matters, but investors may read it differently: Cognition wanted the product surface, the customer base, and the people who were willing to operate at its pace.

The product side looks much cleaner now. In April 2026, Cognition launched Devin in Windsurf, where a developer can plan locally, send work to Devin in the cloud, then review the PR back in the IDE. In June 2026, Cognition went further and introduced Devin Desktop, calling it the next generation of Windsurf. The center of the product is no longer just a code editor. It is an agent command center.

That tells us the acquisition was not just a rescue deal after the OpenAI and Google drama around Windsurf. Cognition needed a local developer surface. Devin alone could have stayed too separate from daily engineering work. Windsurf gives Cognition a much better chance of living where developers already work.

Is Cognition getting pulled into government work now?

Cognition is clearly trying to make Devin credible for government and regulated enterprise now.

In February 2026, Cognition launched Cognition for Government. The post was unusually specific: Windsurf is FedRAMP High, supports DoD IL4, IL5, and IL6 deployments, supports CUI and ITAR compliance, and can be deployed on-prem. Cognition also said Devin is available in AWS GovCloud, with a FedRAMP High Authorized version planned.

The government angle fits the product surprisingly well. A lot of government software work is not greenfield building but rather maintenance, migration, compliance, old systems, and contractor-heavy modernization. Cognition pointed to U.S. government IT spend of roughly $100 billion per year, with nearly 80% going to maintenance. It also said only 3 of 10 critical legacy systems flagged by the GAO in 2019 had been modernized.

The hiring page backs this up. Cognition is currently hiring federal account executives, federal technical engagement managers, federal deployed engineers, and federal DevOps roles. That is not random. It looks like a real go-to-market lane.

But the government can be slow, compliance-heavy, and support-heavy. If Cognition wins here, it will not be because Devin is cool but because Devin can survive procurement, security review, and messy legacy environments.

If you want more recent data on this point, please see our latest AI code assistant market report.

Is Cognition hiring like a software company or a services company these days?

Cognition is hiring like a software company that knows it needs a lot of hands-on deployment to make Devin work.

The careers page is one of the better weak signals. There are research and engineering roles, but the visible bulk is in customer engineering, sales, deployed engineering, partner roles, technical engagement, and applied AI transformation. Cognition is hiring deployed engineers in the U.S., ANZ, APAC, Europe, LATAM, and federal. It is also hiring account directors across APAC, Europe, DACH, MENA, ANZ, enterprise, and federal.

That tells us where the bottleneck probably is: “find the right customer use cases, connect to the right tools, get the agent through internal security, teach teams how to delegate work, and prove the value.” That is why the deployed-engineer role matters. It sits between product, sales, and implementation.

The risk is that this can become too services-heavy.

If every big customer needs a lot of hand-holding, Cognition’s revenue can grow fast while the business stays operationally complex. The upside is that these deployments can become repeatable playbooks. If Cognition learns which tasks Devin should own inside banks, automakers, defense teams, and systems integrators, the messy services layer becomes a moat rather than a drag.

Is Devin’s reliability still the thing to worry about now?

Yes, reliability is still the main thing to watch with Cognition today, but the problem is becoming more specific.

The old critique was simple: Devin looked amazing in demos but struggled in real tasks. That critique is less useful now because the product and customer base have moved on. The better question is where Devin fails, how much review work it creates, and which tasks are worth delegating.

Recent research gives us a better frame. The June 2026 AIDev rejection paper found that 46.41% of agent-proposed fixes across Copilot, Devin, Cursor, and Claude were rejected. The reasons included incorrect implementation, failed CI, missing code, lost sessions, and low-priority fixes.

Another study of 7,156 agentic pull requests found that documentation tasks had an 82.1% acceptance rate versus 66.1% for new features. That gap is more useful than a generic “AI agents are good or bad” debate.

This actually supports Cognition’s recent direction.

Devin’s best commercial wedge is structured delegation: legacy cleanup, vulnerability fixing, tests, migrations, code review, and well-scoped backlog work. If Cognition keeps pushing customers toward the right tasks, reliability becomes manageable. If customers expect Devin to behave like a senior engineer across every ambiguous problem, disappointment comes back quickly.

So, how is Cognition doing these days?

Cognition is doing very well right now.

The company has recent revenue momentum, big enterprise logos, a government push, a serious ROI story, a broader product surface through Windsurf, and a hiring plan that matches real enterprise deployment. We can say pretty confidently that Cognition has moved beyond the launch-demo phase.

The bigger question now is whether Cognition becomes the control layer for AI software work before coding agents become a default feature inside every major developer tool. Cursor, GitHub, Anthropic, and OpenAI are all moving into the same zone. That makes Cognition’s current window valuable but not comfortable.

We are bullish, with one very clear condition. Cognition looks like one of the strongest companies in agentic software engineering today, but it has to win on workflow, trust, ROI, and enterprise adoption. Raw coding ability alone will not be enough for long.

If you want more recent data on this point, please see our latest AI code assistant market report.

Question	Answer	Recent signals behind the answer
Is Cognition still growing fast now?	Yes. Cognition’s growth still looks very strong and increasingly enterprise-led.	May 2026 $1B+ raise; $26B post-money valuation; reported $492M annualized revenue; 10x+ enterprise usage in 2026; banks, automakers, government, and IT-heavy customers.
Is Devin doing useful enterprise work?	Yes. Devin’s strongest use cases are boring, high-cost engineering tasks.	Mercedes-Benz COBOL pilot; 200,000+ lines analyzed; eight months to eight days claim; Itaú vulnerability automation; Infosys and Cognizant delivery partnerships.
Is Cognition more than Devin now?	Yes. Cognition is building an agentic engineering platform around Devin.	Devin 2.2; Devin in Windsurf; Devin Desktop; Review; DeepWiki; Ask Devin; GHES/GitLab support; enterprise API and IP-access features.
Is Cognition proving ROI now?	Yes. ROI has become a central part of Cognition’s enterprise pitch.	June 2026 productivity estimator; 258 sessions; 126 users; rlog 0.74; AI Productivity Guarantee up to $10M in credits.
Is Devin getting cheaper now?	Yes, but Cognition is also protecting compute-heavy usage.	April 2026 Free/Pro/Max/Teams rollout; Pro at $20; Max at $200; Teams minimum at $80; paid Ask Devin, DeepWiki, and Review.
Is Cognition still ahead of competitors?	Strong, but no longer comfortably alone.	Cursor Gartner Leader claim; 70%+ Fortune 500 usage claim; GitHub Copilot agent upgrades; OpenAI Codex push; Anthropic’s 80% Claude-authored internal code claim.
Is Windsurf working for Cognition now?	Product-wise yes, even though the integration looked rough.	July 2025 acquisition; reported layoffs and buyouts; April 2026 Devin in Windsurf; June 2026 Devin Desktop and Agent Command Center.
Is Cognition serious about government now?	Yes. Government looks like a real go-to-market lane.	Cognition for Government; FedRAMP High Windsurf; DoD IL4/5/6; AWS GovCloud Devin; planned FedRAMP High Devin; federal hiring cluster.
Is Cognition hiring like it can scale?	It is hiring for enterprise deployment more than pure research.	Deployed engineer roles across regions; federal roles; APAC, Europe, LATAM, ANZ sales roles; applied AI transformation managers; partner deployed engineers.
Is Devin reliability still a worry?	Yes, but the worry is now about task selection and review burden.	AIDev paper: 46.41% of agent fixes rejected; task acceptance gap between documentation and features; Devin positive acceptance trend but no agent wins every task type.
How is Cognition doing these days?	Very well, with platform risk replacing product risk.	Strong revenue claims; enterprise deployment signals; ROI guarantee; Windsurf product absorption; heavy competition from Cursor, GitHub, Anthropic, and OpenAI.

Who is the author of this content?

NEW MARKET PITCH TEAM

We track new markets so founders and investors can move faster

We build living "market pitch" documents for emerging markets: AI, synthetic biology, new proteins, and more. Instead of outdated PDFs or hallucinated LLM answers, our clients get a clean, visual, always-updated view of what's really happening: key players, deals, regulations, and signals that matter. Learn more about us.