Sourcing Ex-Windsurf Engineers: 230 People, 5 Days, 76% of the Pool

Three weeks after Cognition closed its acquisition of Windsurf, CEO Scott Wu sent an internal email laying off 30 employees and offering buyouts to the remaining ~200, per The Information. The deadline to accept was August 10. If you run technical recruiting at an AI startup, you have five days to move on the single most concentrated pool of AI-coding talent that will exist this year.

The pool is 230 engineers, and it is 76% of the entire US market for this skill set

A defined cohort of roughly 230 ex-Windsurf engineers is now on the market, and per Refolk's index that cohort equals about 76% of the total US population of engineers with both LLM and code-generation skills on their profile. This is not a distressed sale. It is a majority share of a rare skill combination getting handed to whoever moves first.

The numbers, from The Information's reporting via TechCrunch and Refolk's own index:

The killer ratio is the top one. In most sourcing exercises the addressable market is a five or six-figure number and your job is to filter down. Here the addressable market is three digits and one company just released three quarters of it in a single email.

76%

Who these engineers actually are, and who is already gone

The remaining ~230 are Windsurf's senior and staff IC layer, not its research leadership. The top of the stack, including the CEO, co-founder, and research leads, already left for Google DeepMind in the $2.4 billion reverse-acquihire that preceded Cognition's purchase. Read that carefully before you build your target list.

That structure has three implications:

• The people still on the payroll are the engineers who shipped the actual Windsurf IDE product, agent loops, and completion models. Not the paper-writers.
• They watched their founders take a Google offer and leave. Loyalty to the Windsurf brand is not a variable in your outreach.
• They then watched Cognition, which called them "world-class people" three weeks ago, offer to pay them to leave. Mission-speak from a new employer will be read with a very specific kind of skepticism.

For a Series A or B founder, this is the sweet spot. Research-lead names rarely take Series A offers. Senior ICs who built a shipped AI-coding product absolutely do, at the right number.

The nine-month severance changes what "competitive offer" means

Cognition's buyout is nine months of salary, which is roughly triple the typical tech severance of four weeks plus one week per year of service reported by CNBC. Every ex-Windsurf engineer walks into your process with nine months of runway already banked, and your offer competes against the option of doing nothing until May.

Practical consequences for anyone doing Windsurf layoffs sourcing this week:

1. Signing bonuses matter less than they usually would. The candidate already has cash.
2. Equity math has to be transparent and defensible. These engineers just watched a $2.4B reverse-acquihire and a reported ~$250M sale close over their heads. They will ask for the cap table.
3. Start-date flexibility is a real lever. An engineer who can start in January but sign in August is worth more than one who wants to start Monday.
4. "Move fast" pressure tactics will backfire. They have nine months. You have five days. Only one of those is a real deadline.

Anthropic's public Claude Code Web Engineer bands sit at $320K to $560K. That is your comp anchor for total cash comp on this pool, not your ceiling.

The candidate has nine months of runway. You have five days. Only one of those is a real deadline.

Where to actually find them in five days

Start with GitHub commit history against the Windsurf and Codeium repos, cross-reference LinkedIn tenure at Codeium and Windsurf from 2022 to 2025, and layer on personal sites for signal on who is already talking. Then rank by whether their public work touches the specific problem you are hiring for.

The mechanical steps most teams get wrong:

• Do not rely on LinkedIn "Open to Work" filters. These engineers were offered buyouts on August 5. Nobody has updated their profile yet. The signal you want is tenure ending at Windsurf or Codeium, not a green ring.
• Do search on the Codeium legal name. Windsurf rebranded in late 2024. Older LinkedIn profiles and GitHub bios still say Codeium. Half your target list is hiding under the old string.
• Cross-reference GitHub org membership. The codeium and Exafunction GitHub orgs list contributors publicly. Merge with LinkedIn hits to deduplicate.
• Filter for IC seniority, not title inflation. Windsurf ran flat. "Member of Technical Staff" at Windsurf is closer to Staff at a normal company than Senior.

This is the exact gap where Refolk earns its keep on a five-day clock. Instead of building four different Boolean queries across LinkedIn, GitHub, and the open web and reconciling the results in a spreadsheet, you describe the person in plain English and get one ranked shortlist across all three surfaces.

Who you are competing with, and how they will try to win

The competition is Anthropic, Anysphere (Cursor), Lovable, CodeRabbit, and the frontier labs that already dominate this profile's destination market. Refolk's index shows the current employers of US engineers with LLM + code-gen skills skew heavily to incumbents:

If the pool disappears into any of those four, it disappears behind 12-month NDAs and internal-transfer walls. Your competitors for the Cognition Windsurf buyout cohort, in rough order of threat:

• Google DeepMind. Windsurf's founders and research leads are already there. Ex-teammate referral is the highest-conversion channel in tech recruiting, and Google has a five-day head start via Slack DMs you cannot see.
• Anthropic. In early July, Boris Cherny and Cat Wu left for Anysphere and Anthropic re-hired them within two weeks. Anthropic will pay whatever it takes to keep a Claude Code IC. Extrapolate.
• Anysphere (Cursor). After acquiring Graphite in December 2025, Cursor's BugBot now reviews 2 million-plus PRs monthly. Code-review headcount is scaling and ex-Windsurf ICs fit the profile exactly.
• Lovable. Doubled annual revenue to $200 million in four months as of November 2025, over 2 million users, more than $200M raised. European base makes them a wildcard bidder US recruiters underestimate.
• CodeRabbit. Shipped a greater-than-10% recall improvement on code review with Opus 4.7. Smaller, hungrier, could realistically win 5 to 10 ex-Windsurf hires on mission fit alone.

If you are a Series A founder reading this and thinking you cannot out-cash Anthropic, you are right, and it does not matter. Five to ten hires out of a 230-person pool is a company-defining outcome, and the frontier labs do not want all 230 either.

The message that works on this specific cohort

Trust-first, not mission-first. Scott Wu's "we don't believe in work-life balance" line, combined with a mandated six days in office and more than 80-hour weeks for anyone who stays, taught this exact cohort that "world-class people" language is negotiating theater. Your first message needs to prove you know that.

What to put in the first outreach:

• Name the situation. "Saw the Cognition news" beats "excited to connect." They know what happened. Pretending otherwise is worse than useless.
• Show the equity math. Strike price, current 409A, last preferred, dilution assumption. Two sentences. Ex-Windsurf engineers now assume opacity means a bad number.
• Be honest about hours. If you run 55-hour weeks, say 55. If you run 40, say 40. The pool has calibrated on lies this month.
• Skip the mission paragraph. It reads as Scott Wu cosplay right now. Come back to mission on the second call.

The engineers you want to hire ex-Windsurf are the ones who will fact-check you on your first message. Write for that reader.

5 days

What to do in the next 72 hours

Build the list, send the messages, book the calls, in that order, and do not wait for the buyout deadline to pass. The playbook, compressed:

1. Hours 0 to 12: Pull the full ex-Windsurf and Codeium tenure list. Aim for 230 names with LinkedIn URL, GitHub handle, and one piece of public technical signal each.
2. Hours 12 to 24: Score the list against your specific role. Cut ruthlessly. You want 40 to 60 names you would genuinely hire, not 230 you would interview.
3. Hours 24 to 72: Send personalized first messages, one per candidate, referencing the specific technical work you saw. Include equity math and honest hours in message one.
4. Days 4 to 7: Book calls. Move fast on offers for the top decile. Frontier labs are running the same clock.

If you are staffed for one recruiter and one hiring manager, sourcing after acquisition layoffs at this speed is where a plain-English search tool collapses the timeline. Refolk builds the 230-name list from a sentence, which turns hour zero to hour twelve of that plan into hour zero to hour one. The rest is a human job.

The pool the entire sector wants to hire ex-Windsurf engineers from is small, named, and dated. In 12 months it will be a footnote in someone's Series B announcement. The only question is whose.

FAQ

How many ex-Windsurf engineers are actually available right now?

Roughly 230, comprising 30 who were laid off outright and about 200 who were offered a nine-month-salary buyout with an August 10 decision deadline, per The Information's reporting. Some percentage will accept Cognition's terms and stay, some will take the buyout and go to Google DeepMind through existing founder relationships, and the remainder are your realistic addressable pool. A reasonable working number for open-market sourcing is 120 to 180 engineers, most of them senior or staff IC, with the top research leadership already gone.

Why is the 302-person US pool number so small?

Refolk's index counts engineers whose public profile shows both LLM work and code-generation work as named skills, which is a narrow intersection compared to the broader 21,630-person pool of senior and staff engineers with TypeScript plus LLM skills. Coverage is not total, so 302 is a floor, not a ceiling. The point is not the exact denominator, it is that the ratio structure holds: the ex-Windsurf cohort is a majority share of the specifically-qualified US pool, which is why frontier labs will move on it hard.

What is a competitive offer for this cohort?

Anthropic's public Claude Code Web Engineer bands run $320K to $560K in total annual comp, which is the anchor for what a senior ex-Windsurf IC can command at a well-funded lab. Startups will not match that on cash and should not try. Compete on equity percentage, technical scope, honest hours, and a start date that respects the nine months of runway the buyout gave the candidate. Signing bonuses are less useful than usual because the candidate is not cash-constrained.

Should I skip the ones who decline the buyout and stay at Cognition?

No. The internal email framed six-day, 80-plus-hour weeks and "we don't believe in work-life balance" as the price of staying, which means fence-sitters who declined the buyout out of inertia are still gettable in the following 60 to 90 days. Add them to your CRM with an October follow-up date. The first re-outreach after someone has actually worked a few 80-hour weeks under a new regime tends to convert well.