Cognition's 5-Day Windsurf Buyout: Sourcing 230 Engineers

Cognition acquired Windsurf three weeks ago. Last week it fired 30 people and offered the remaining ~200 a nine-month buyout with a five-day decision window. That is one of the most concentrated pools of production AI-coding engineers to hit the market this year, and if you wait for LinkedIn "Open to Work" flags you will be six weeks late.

What actually happened at Cognition and Windsurf

Cognition, the two-year-old San Francisco startup valued at $4B and known for the AI coding agent Devin, laid off 30 Windsurf employees last week and offered buyouts to the roughly 200 remaining, per The Information. The buyout is nine months of salary. Stayers face six days in office and 80+ hour weeks, framed by CEO Scott Wu as "extreme performance." Employees had until August 10 to decide.

The pool sits on top of an earlier disruption. Google DeepMind ran a $2.4B reverse-acquihire earlier this year that took Windsurf's CEO Varun Mohan, co-founder Douglas Chen, and the top research leads. Cognition then picked up what remained (an $82M ARR product and roughly 250 employees) for an industry-estimated $250M, a 92% discount on a business whose enterprise revenue was doubling QoQ.

The residual math looks like this:

The 230 who are leaving are the builders. Google already took the researchers. If you are hiring for a product that has to ship an agent to paying enterprise customers, this is the better cohort, not the worse one.

Why this pool matters more than "230 engineers" suggests

The ex-Windsurf 230 represent only about 3% of the US senior AI-agent talent universe, but they are the largest simultaneously-available cohort of shipped-agent engineers to appear in a single week all year. Almost every comparable engineer is locked inside OpenAI, Microsoft, Google, or Meta on retention packages you cannot compete with.

In Refolk's index of professional profiles, there are 7,749 US senior-and-above engineers who list both "LLM" and "AI Agents" as skills. That is the entire adjacent universe. Of the top ten current employers of those engineers, OpenAI and Microsoft tie for the lead, and the rest is fragmented across stealth startups. Five of the top ten hits cluster in the SF Bay Area and three in NYC, which happens to be exactly where the Windsurf cohort lives.

7,749

The mechanism matters. Talent inside OpenAI is not answering your InMail. The ex-Windsurf pool is the rare case where shipped-agent engineers are pushed onto the open market on a dated timeline. If your competitive story requires "engineers who have put a coding agent in front of paying customers," this week is the buying window.

The five-day window creates a two-wave market

There will be two distinct sourcing waves, and most recruiters will only work the second one. Wave one runs from roughly August 10 to August 15: decisions are locked, but nothing shows up on LinkedIn. Wave two hits in September and October: profile updates land, "Open to Work" badges appear, and every AI startup in the Bay throws money at the same names.

The edge is entirely in wave one. Here is why direct outreach this week beats waiting:

• LinkedIn signal lags by 4 to 8 weeks. People update titles slowly, especially when they hold nine months of severance and no urgency.
• The buyout forfeits unvested equity. Cognition granted four years of equity but coupled it to the 80-hour expectation. Buyout-takers walk with cash and no vested upside, which means they will start conversations early even if they do not sign fast.
• Bidding pressure compounds week over week. The same OpenAI, Anthropic, Cursor, Replit, and stealth-agent recruiters are working the list. Every day of delay costs you a warm reply.
• The comparable pools disappeared fast. Scale AI's 200-person cut after the Meta deal and Character.ai's post-Google buyouts both cleared quickly. This will too.

The edge is not in the list. The edge is in reaching the list before it exists on LinkedIn.

How to actually build the shortlist this week

Skip LinkedIn as your primary surface. The ex-Windsurf pool is discoverable right now through code, community, and product artifacts that predate any profile update. A working shortlist has four inputs:

1. The Codeium GitHub org and adjacent repos. Commit history reveals real Windsurf contributors, not marketing hires. Filter by contributions in the last 12 months to the IDE, autocomplete, and agent-runner code paths.
2. The Windsurf user Discord and the Cursor Community Slack. Windsurf staff engineers post there under real names. Cross-reference handles against GitHub and personal sites.
3. Conference and podcast appearances from 2024 to 2025. AI Engineer Summit, Latent Space, and similar venues surfaced the Windsurf builders who are not on the exec team Google took.
4. The "Codeium" alumni pattern. Windsurf was Codeium before the rename. Anyone with "Codeium" in their history and a 2024 to 2025 tenure is in-scope.

This is the exact gap Refolk closes. Instead of running four separate boolean searches across GitHub, LinkedIn, and Discord exports, you describe the person in plain English and get a ranked shortlist back. Refolk pulls from GitHub, LinkedIn, and the open web in one pass, which is the only way to catch wave-one names before their profiles update.

What the ex-Windsurf 230 will actually accept

Nine months of Bay Area SWE salary is roughly $150K to $300K in cash runway, which resets the negotiation floor. These engineers will not accept lowball offers, and they will not accept "we move fast" as a substitute for a real story. Expect a 4 to 8 week decision arc, not a two-week close.

Three things move this cohort:

• Mission and product surface, not cash. They just left an 80-hour ultimatum. They will screen you on whether the product is real.
• Equity refresh that actually matters. They forfeited four years of unvested Cognition equity. A meaningful early-stage grant, not a top-up, changes the conversation.
• Founder or staff-plus scope. Many of these engineers led major surfaces at Windsurf. A senior IC slot on a legacy team is a downgrade they will refuse.

If your pitch is "come do the same job for slightly more money," you are wasting the intro. If your pitch is "own the agent surface end to end, here is the equity, here is the customer," you are competitive with Anthropic and Cursor for the same names.

Why this is a template, not a one-off

Reverse-acquihire buyouts are becoming a pattern. Scale AI cut 200 after Meta's $14.8B investment. Character.ai issued buyouts after Google's $2.7B deal took its founders. Cognition-Windsurf is the latest instance. Building a repeatable playbook for the next one is more valuable than sourcing this cohort well.

That playbook has three legs:

• Watch the deal announcement, not the layoff announcement. Reverse-acquihires telegraph the buyout wave weeks ahead of the actual cuts.
• Index the target company's GitHub and Discord footprint before the buyout hits. By the time layoffs are announced, you already have the shortlist warm.
• Reach out on day one of the decision window, not day one of the LinkedIn signal. The delta is 4 to 8 weeks of competitive advantage.

For sourcing AI coding engineers specifically, Refolk is built for exactly this shape of work: a plain-English description of the person, resolved across GitHub commit history, LinkedIn, and personal sites in a single query. When the next reverse-acquihire buyout drops, you should be able to run "ex-[company] senior engineers with shipped agent code, Bay Area or remote, in the last 24 months" and get a ranked list in the same afternoon.

5 days

FAQ

How many ex-Windsurf engineers are actually available?

Roughly 230. That breaks down as 30 laid off last week and about 200 offered a nine-month buyout with an August 10 deadline, per The Information. Not all 200 will take the buyout (staying means 80+ hour weeks and six days in office), but even a partial take rate stacks on top of the 30 already gone. Google DeepMind previously took the CEO, co-founder, and top research leads in a separate $2.4B licensing deal, so this pool is the product and infrastructure builders, not the research layer.

Why not just wait for their LinkedIn profiles to update?

Because you will be 4 to 8 weeks late, and every AI-coding recruiter in the Bay Area works the same LinkedIn signal. Buyout-takers hold nine months of severance, so they are not rushing to flip "Open to Work" on. The window where you can reach them before the bidding war is right now, through GitHub commit history to the Codeium and Windsurf repos, the Windsurf user Discord, and conference appearances. Refolk resolves those signals in one query so you do not have to run four searches.

What compensation should I lead with?

Not cash. These engineers just walked from an 80-hour ultimatum with nine months of runway, so a modest salary bump does nothing. Lead with product scope (founding or staff-plus, ownership of a real surface), equity refresh that replaces the four years of Cognition equity they forfeited, and a customer story that proves the product ships. Expect a 4 to 8 week decision cycle, not a two-week close.

Is this really a repeatable pattern or a one-off?

It is a pattern. Scale AI cut 200 after Meta's $14.8B investment. Character.ai issued buyouts after Google took its founders in a $2.7B deal. Cognition-Windsurf is the latest reverse-acquihire buyout wave. The playbook (index the target's GitHub and Discord footprint before the layoff hits, then reach out on day one of the decision window) will keep paying off. Build it into your standard operating procedure now.