14,500 Bay Area Layoffs Aren't 'Median.' Check Their Commits Since Exit

If you're hiring engineers in the Bay Area, the popular reading of this year's layoff numbers - that the median San Francisco engineer is being replaced - will send you looking in the wrong place. The laid-off pool isn't made of median engineers, and the people in it who have shipped LLM, eval, agent, or inference code since their exit date can be found. You just can't find them by searching titles.

That reading comes from Bloomberg's "Jobless tech workers are being left out of San Francisco's AI boom," by Francesca Maglione. It ran on September 17 and resurfaced on Hacker News around November 2, where commenters mostly agreed the median engineer is done.

Why the "median engineer" reading of Bloomberg is wrong

Bloomberg's numbers show the market for generic software roles shrinking. They don't show a pool of interchangeable engineers who are all obsolete. Some people in the laid-off group have retooled since they left, and the proof is in their code, not their headlines.

The figures behind the HN thread:

• Bloomberg's analysis of California EDD data found tech employers filed layoff notices covering more than 14,500 Bay Area workers in the 12 months ending in June. That's almost double the prior fiscal year.
• Lightcast data shows Bay Area software engineer demand down 42% since 2022, while AI-related positions rose 37%.

The "median" argument falls apart once you look at how layoffs work. Companies cut by budget and org chart, not by what each person built last year. A layoff notice can easily include the person who spent the past year writing vLLM patches at night. When generic demand falls 42% and AI demand rises 37%, engineers drift toward the growth. Laid-off engineers have the time to do it first.

42%

Where the 2026 Bay Area layoff pool actually sits

Most of it is in Santa Clara and San Mateo counties, not San Francisco. Through October 30, 2026 tech WARN notices in the four core counties named 11,890 workers. About 75% of them were in Santa Clara and San Mateo. San Francisco County had roughly 20%.

Source: golayoffs.com tech layoff tracker, calendar 2026 to Oct 30.

The more useful number is workers per notice. San Mateo averages about 285 workers per notice, against about 92 in Santa Clara and 83 in San Francisco. So a few large enterprise cuts at Oracle, Visa, and Meta make up most of San Mateo. Those cohorts lean toward infrastructure and backend veterans: people who ran systems at scale, not app developers working in the city. Inference serving and agent infrastructure need exactly those skills, so this part of the pool is worth more than the "SF engineer" framing suggests.

Two counts get mixed up here, so keep them separate:

• Bloomberg's 14,500+ covers WARN notices in the 12 months to June.
• The SF Chronicle's nearly 14,500 covers Bay Area tech cuts announced for calendar 2026 as of late October. That's already more than the roughly 10,170 in all of 2025.

The newest filings give you known exit dates. Front and Dropbox are cutting 110 San Francisco jobs combined, and Dropbox's 57 layoffs take effect November 18. Salesforce has filed three 2026 California WARN notices covering 211 San Francisco workers and hasn't publicly tied them to AI. Repeated small rounds like Salesforce's draw less attention than one big cut, so fewer recruiters notice them.

If you want to hire laid off software engineers in the Bay Area and you start your search in San Francisco, you're looking at about a fifth of the pool.

Why titles and "Open to Work" badges miss the retooled

Most engineers who have retooled haven't updated their headlines yet. In Refolk's index of professional profiles, searching headlines for availability or location phrasing alongside an LLM skill returned zero matches.

Here's what the index shows:

• 4,232 US professionals hold a Software Engineer, AI Engineer, ML Engineer, or Forward Deployed Engineer title and mention "LLM" in their headline.
• In a 25-profile sample from that group, "AI Engineer" (12) narrowly outnumbers "Software Engineer" (11). The San Francisco Bay Area is the most common region (3 of 25), ahead of New York and Dallas-Fort Worth (2 each). Meta is the only employer that appears twice. The sample is small, so treat it as directional.
• Searches for "open to work San Francisco Bay Area," or "San Francisco Bay Area" plus an LLM skill, returned 0 matches.

4,232

Put 4,232 LLM-headline engineers across the whole country next to 14,500 Bay Area WARN workers and you get about 0.29 per laid-off worker. The two numbers cover different scopes, so this is a rough comparison, not a conversion rate. It still makes the point: headlines lag the work. The engineer committing to sglang this month probably still reads "Software Engineer, ex-Oracle." And engineers don't put their location or availability in a headline, so filtering on those words finds nobody.

Why a green contribution graph proves less in 2026

Agents now inflate commit counts, so raw post-layoff activity is a weak signal. GitHub's Octoverse 2026 reports that more than 40% of pull requests now start with an agent, up from 17% in March. New repos with an agent config file logged 8x the commits of repos without one.

That changes how you read a profile. A solid green graph since someone's exit date might mean they shipped serious work. It might also mean they pointed a coding agent at a side project and let it run. Importing an LLM SDK proves little too. Octoverse 2025 counted more than 1.1 million public repos importing one, up 178% year over year. Calling an API is now table stakes.

The repos to check are the fast-growing AI infrastructure projects. Octoverse found vllm, ragflow, and ollama gaining contributors faster than vscode and flutter did historically. Add sglang and llama.cpp. A merge into one of these means someone outside the candidate's control reviewed the work and accepted it, which a self-owned repo can never show.

A PR merged into someone else's repo is the one commit nobody can approve for themselves.

How to measure the post-exit commit window

Match the candidate's last employer to its WARN filing's effective date, then count only the work that came after it. Exit dates are public, so you can measure the retooling window instead of guessing at it.

1. Pull the filing. Dropbox's effective date is November 18. Salesforce's three San Francisco filings took effect May 1, August 7, and October 5.
2. Confirm the employer and team. Use the profile to check that the person was in the affected org, not just at the company.
3. Resolve the GitHub handle. Link the professional profile to the code identity through a personal site, a commit email domain, or a talk bio.
4. Filter activity to after the effective date. Look only at contributions to LLM, eval, agent, or inference repos.
5. Weight outside merges over self-owned commits. Use the signal table above.
6. Note how long the window is. Most of the calendar-2026 cohort left recently, so expect weeks to a few months, not years.

One catch: until November 18, Dropbox's cohort hasn't left yet. What you can see for them now is pre-exit side work. That's still a strong lead, because these engineers hit the market on a known date with a clear story.

Step 3 is where manual sourcing stalls. Tying a LinkedIn profile to a GitHub handle is slow detective work, and you have to repeat it for every candidate. Refolk does that cross-referencing for you and returns people with their code and profile already linked, so you can spend your time judging the work instead of hunting for it.

Community signals: un(PTO) and where retooling shows up socially

Communities of laid-off workers are where you can watch retooling happen in public. Treat them as places to show up, not lists to scrape. The clearest Bay Area example is un(PTO), founded in December by former Google employee Basem Istanbouli. It has more than 1,700 members, and up to 80 people attend its weekly outings.

What works in spaces like this:

• Give something. Host a session, pay for an outing, or offer mock interviews for AI-adjacent roles.
• Ask for referrals, not résumés. Members know who has been building with agents since leaving.
• Read the public trail. Hack nights, demo days, and HN threads (the resurfaced Bloomberg thread included) show who is building and who is only talking about it.
• Don't mass-DM from a member list. A 1,700-person community remembers the recruiter who did that.

Community activity is supporting evidence. Use it to find the code, not to stand in for it.

Why to move before AI-lab IPO money reprices them

Because they're cheaper now than they will be. A laid-off engineer with three months of shipped agent work costs less today than after they land at a lab, and the OpenAI and Anthropic IPOs that Bloomberg discusses would add liquid equity on top of pay that's already high.

The benchmarks are already public:

Forward-deployed engineering is the AI-lab role that laid-off backend and infrastructure veterans map to most naturally. If you're a founder or a mid-size company hiring for that profile, those are the ranges you're competing with. With general SWE demand down 42%, the engineers with real post-exit evidence are a small slice of a very large pool. Sorting that pool by hand takes weeks you don't have, and the people you actually want get picked up while you're still working through it.

When you reach out, write the pitch around the work, not the layoff:

• Open with the artifact. Name the merged PR or the eval repo you read.
• Acknowledge the exit date without dwelling on it. "Since leaving in August, you've shipped..." works. "Saw you were laid off" doesn't.
• Map their past to your role. An ex-Oracle or ex-Visa infra engineer who has patched vllm is the inference hire you need. Tell them that directly.
• Be concrete about comp early. The FDE ranges are public. Pretending otherwise wastes everyone's time.

Sourcing laid off tech workers in 2026 rewards whoever reads the evidence first. The headline counts are public. The post-exit commits are too, for anyone who checks.

FAQ

Are most laid-off Bay Area tech workers in San Francisco?

No. Of the 11,890 tech workers named in 2026 WARN notices across Santa Clara, San Mateo, San Francisco, and Alameda through October 30, about 75% were in Santa Clara and San Mateo. San Francisco County had roughly 20%. San Mateo's large enterprise cuts, averaging about 285 workers per notice, skew toward infrastructure and backend veterans.

Is a busy GitHub contribution graph after a layoff a good sign?

On its own, not much. Octoverse 2026 reports more than 40% of pull requests now start with an agent, and repos with agent config files log 8x the commits. Look for PRs merged by outside maintainers into projects like vllm, sglang, llama.cpp, or ragflow, and for eval or benchmark repos that strangers have starred or forked.

How do I find engineers with LLM experience on GitHub if their headline doesn't say so?

Start from the exit date, not the headline. In Refolk's index, only 4,232 US engineers mention "LLM" in their headline, and searches for availability or location phrasing returned zero. Match the candidate's last employer to its WARN effective date, link their professional profile to their GitHub handle, and check what they've contributed to AI infrastructure repos since.

When should I contact laid-off engineers with AI skills?

Early. The calendar-2026 cohort left recently, so their retooling windows are weeks to a few months long, and some filings, like Dropbox's November 18 cohort, haven't taken effect yet. Forward-deployed engineer pay already runs $185K to $320K at OpenAI and Anthropic, and lab IPO liquidity would push that further, so engineers with real post-exit work cost less now than they will after those IPOs.