Cognition's Aug 10 Cliff: 200 Windsurf Engineers, 5 Days to Intercept
Cognition gave 200 Windsurf engineers until Aug 10 to accept a 9-month buyout or work 80-hour weeks. Here is how to source the pool before it disperses.
On August 5, 2025, Cognition (the maker of Devin) told roughly 200 Windsurf employees they had until August 10 to accept nine months of severance or commit to six-day, 80+ hour weeks in the office. That is a five-day window to intercept engineers who, weeks earlier, were shipping an agentic IDE with $82M in ARR. If you are hiring AI coding agent engineers, this is the cleanest signal you will get all year.
What actually happened between Aug 4 and Aug 10
Cognition offered a nine-month severance buyout to about 200 Windsurf staff and gave them until August 10 to decide. Those who stay work six days a week, 80+ hours, in-office. This came less than a month after Cognition acquired Windsurf's remains for an estimated $250M, and after already laying off 30 employees in the first weeks.
The sequence, in order:
- Google's $2.4B reverse acqui-hire pulls CEO Varun Mohan, co-founders, and top researchers into DeepMind, leaving roughly 250 employees behind.
- Cognition acquires the remaining company for about $250M (roughly $1.19M per employee at acquisition).
- Within weeks, Cognition lays off 30 people.
- On August 5, CEO Scott Wu emails the remaining 200 with the ultimatum.
- August 10 is the accept-or-commit deadline.
Wu's email, as reported by TechCrunch, is not subtle: "We don't believe in work-life balance - building the future of software engineering is a mission we all care so deeply about that we couldn't possibly separate the two." Interim Windsurf CEO Jeff Wang, elevated during the 72-hour Google-to-Cognition scramble, called it "the wildest rollercoaster ride of my career." That is the mood inside the building this week.
Why this pool matters more than the headline count
The remaining 200 Windsurf engineers are the highest-signal batch of AI coding agent engineers to hit the market in 2025, because Google already skimmed the research layer and left the product, GTM, and infra people who actually shipped revenue. In Refolk's index of professional profiles, only about 1,399 US engineers publicly list the combination of AI Agents, LLM, and Code Generation skills. The Windsurf leavers represent roughly 14% of the entire disclosed US supply of that specific skill triangle.
Filter to Senior, Director, or VP seniority and the number collapses to 101 across the whole country. If even a third of the Windsurf 200 are senior, this single company's exit adds roughly two-thirds again to the visible senior pool overnight.
| Segment | Count | Source |
|---|---|---|
| US engineers with AI-Agents + LLM + Code-Generation skills | 1,399 | Refolk index |
| Same profile, Senior / Director / VP | 101 | Refolk index |
| Senior share of the AI coding agent pool | 7.2% | Derived |
| Windsurf engineers weighing the buyout | ~200 | TechCrunch, The Information |
| Windsurf leavers as % of US AI coding agent supply | ~14% | Derived |
| Windsurf leavers as % of senior US supply (if ~1/3 senior) | ~66% | Illustrative upper bound |
The point is not that 200 is a big number. The point is that 200 is a big number for this skill set, and it lands in one week, in two metros (San Francisco Bay Area and New York), from one former employer.
Why most of them will take the buyout
Nine months of Bay Area senior engineering comp is roughly $225K to $340K in the hand for doing nothing, versus 80-hour weeks under a founder who has explicitly told them he does not believe in work-life balance. That math tilts hard toward the door for anyone with savings, a partner, a kid, or a competing offer. Cognition's own ARR has more than doubled since the acquisition per Wu, which means the retention argument ("stay and ride the rocket") is weakening as the acquired IP does the work regardless of who is in the seat.
This is not a slow trickle. It is a cliff on August 11.
Three second-order effects follow:
- Compressed timing. Every hiring manager in AI infra will hear about the same 200 people in the same week. The winners will be the ones who had a shortlist and a warm intro path already built by August 8, not the ones who start Boolean stringing on August 12.
- Bidder concentration. Refolk's index shows the closest-match employers for this profile are Anthropic, Google, Microsoft AI, OpenAI, xAI, and Amazon, plus the direct IDE rivals Cursor (Anysphere) and Zed. Assume all of them have already opened a channel. Founders competing with them need a differentiated pitch, not a bigger comp band.
- A visible severance runway. Nine months of severance means candidates can afford to be picky. Expect the median offer decision to stretch to weeks, not days, once the initial rush settles.
This is not a slow trickle. It is a cliff on August 11.
What these engineers are actually worth
The Windsurf business generated $82M ARR across roughly 200 engineers, or about $410K of ARR per engineer, with 350+ enterprise customers and enterprise ARR doubling quarter-over-quarter at the time of acquisition. That is the pitch to a hiring manager who has seen too many résumés listing "worked on an AI agent": these people shipped an agentic IDE to paying enterprises at a revenue-per-head number most Copilot- and Cursor-adjacent teams would envy.
Two less obvious skills sit under that number:
- Model-provider survival. Anthropic cut Windsurf's direct Claude access in June, with co-founder Jared Kaplan attributing the decision to rumors of an OpenAI acquisition. Whoever was still shipping in July knows how to reroute an agent product around a hostile model provider on short notice. Nobody puts that on a résumé; every AI founder currently needs it.
- Enterprise agent GTM. 350+ enterprise customers is not a research problem, it is a security review, SSO, on-prem, and procurement problem. The people who solved that at Windsurf are the ones Google did not take.
How to actually source the pool this week
Move on named individuals, not job posts. The buyout email went out August 5 and the deadline is August 10, so anyone waiting on inbound applications in the second half of August is picking through what Anthropic, xAI, and Cursor already passed on. A few concrete moves:
- Build the list before the deadline. Windsurf was formerly Codeium, so LinkedIn searches on both names will surface the population. Refolk's index resolves the Codeium/Windsurf history into a single continuous profile, which matters when the target renamed their employer twice in 18 months.
- Filter for the roles Google did not take. Skip anyone whose title sits in pure research; Varun Mohan and the top researchers are already at DeepMind. Prioritize IDE product engineers, agent infra, evals, and enterprise deployment.
- Ask what they signed. Non-competes are effectively unenforceable in California, so that is not the blocker. The buyout will almost certainly carry a release and NDA. A candidate who cannot discuss why they left is a candidate you cannot reference-check properly; know that before you make an offer.
- Reach into the ex-Codeium and ICPC-adjacent networks. Cognition's founding team (Scott Wu, Steven Hao, Walden Yan) came out of the competitive programming world, and much of the early Windsurf and Codeium engineering bench overlaps with that community. Warm intros through those networks land better than a cold InMail on a week when everyone is getting cold InMails.
Who the likely bidders are, and how to be the exception
Assume Anthropic, OpenAI, xAI, Microsoft AI, Google, and Amazon are already in the inboxes of any Windsurf engineer with a public GitHub. Cursor (Anysphere) and Zed are the direct product-competitor bids. That is the crowded lane.
The uncrowded lane is early-stage. A nine-month severance runway is close to an angel round for a founding engineer. Startups in the Founders Fund, Lux, and 8VC orbit (Cognition's own cap-table neighbors) have a natural cultural bridge and, more importantly, a pitch that isn't "join our 5,000-person AI org." For a Windsurf engineer who just watched a $3B OpenAI deal collapse into a $2.4B carve-out and a $250M scrap sale, the appeal of a smaller cap table with clear equity is not theoretical.
Two things to lead with in that pitch:
- Clear scope. Windsurf engineers just lived through 72 hours of ownership uncertainty. "You will own the agent runtime end to end" beats "you will collaborate across pods."
- Sane hours by contrast. You do not have to promise 40-hour weeks. You have to not be 80.
FAQ
How do I know an ex-Windsurf candidate is actually one of the 200, and not someone who left months ago? Check their last employer end date against early August 2025 and cross-reference the Codeium-to-Windsurf rename. Anyone who left before Google's $2.4B reverse acqui-hire in July is a different story - they missed both the buyout and the culture reset, and their reasons for leaving are their own.
Is the buyout release going to block them from talking about their work? Assume yes on specifics, no on generalities. Standard severance agreements in California can restrict disclosure of internal information and non-disparagement, but they cannot stop a candidate from describing what they built at a high level or from working for a competitor. Ask the candidate directly what the release covers before your reference calls.
Should I wait for the market to settle before making offers? No. The nine-month severance runway means the strongest candidates will still be picky in October, but the ones with competing offers will close in the first three weeks. If your role requires an ex-Windsurf hire specifically, the window to be in the first round of conversations closes around August 15.
What about the engineers who stay at Cognition? They are self-selecting into 80-hour weeks under a CEO who has publicly said he does not believe in work-life balance. That is a legitimate choice, and some of them will build extraordinary things. It also means the ones who leave are disproportionately the engineers with options, families, or a different definition of ambition - which is most of the people you actually want to hire.
Try it on your own search
Stop building boolean strings. Just describe the person.
Type one sentence and I plan the search, read GitHub, public LinkedIn and Crunchbase records, and the open web live, then hand back a ranked shortlist with the reasoning behind every name. No filters to learn, no export to clean up, no sales call to sit through.
- One sentence in, a ranked shortlist out. No boolean, no filters, no seat to buy.
- Read live at search time, not from a database that went stale last quarter.
- Watch every step as it runs, and see why each name made the list.
- Staff backend engineers in NYC who shipped Rust in production
- Series A fintechs in SF under 50 people, growing headcount this year
- Maintainers of fast-growing Rust web frameworks on GitHub
500 free credits on sign-up. No card, no demo call. See real searches.