Cloudflare Cut 1,100 and Added 586 Engineers. Source the Inbound Bench.

Every WARN-watcher in your Slack has Cloudflare flagged as a May 2026 layoff story. They are sourcing the wrong people. The BNP Paribas LinkedIn pull, first reported by Business Insider in late June and since confirmed by Matthew Prince, shows engineering grew from 1,308 to 1,894 in the same weeks the company cut 1,100 jobs. The released cohort is finance, legal, and middle management. The bench worth sourcing is inbound, not outbound.

The category swap nobody priced in

Cloudflare announced its first mass layoff in 16 years on May 7, alongside a Q1 that hit $639.8 million in revenue, up 34% year over year. The cut was roughly 20% of the workforce, about 1,100 people, against a December 2025 base of 5,156 employees. Markets read "20% cut" and the stock fell more than 20% on the print.

Then the BNP Paribas team did the obvious thing nobody else did. They counted LinkedIn profiles. The engineering function did not shrink. It grew 45%, a net add of 586 builders, while the company shed measurers in marketing, finance, internal audit, and consolidated operational roles. CFO Thomas Seifert said the cut hit every team and geography except quota-carrying sales. He meant it. The sellers pond is fully stocked. The measurers pond is the one that emptied.

If you are sourcing Cloudflare engineers in 2026 on the assumption that 1,100 people just hit the market and a meaningful slice of them write Go and Rust, you are working off a model that does not match the underlying headcount data.

45%

What Prince actually meant by "builders, sellers, measurers"

Prince did not invent the taxonomy. He pulled it from Peter Drucker's 1954 The Practice of Management, which classified work into people who make the thing, people who sell the thing, and people who count the thing. Prince's claim is that the third bucket has gotten structurally cheaper to automate, and that Cloudflare's internal data agrees: 97% of R&D uses AI coding tools, 100% of production-code contributions are reviewed by autonomous agents, and internal AI usage rose more than 600% in the three months before the cut.

Whether you buy the framing or not (more on that below), it dictates the hiring funnel now. Every one of the 1,111 summer interns selected from nearly a million applicants was screened as builder or seller and explicitly "AI-native." Prince expects most of them to convert to full-time. Sourcing engineers who cannot demonstrate fluency with Workers, Claude Code, or Cursor-style agentic workflows will not clear Cloudflare's loop, and increasingly will not clear the loops at peers running the same playbook.

GitLab's Bill Staples ran the parallel restructure in May, cutting 7% and reorganizing engineering into 60 autonomous teams while flattening up to three layers of management. He called it the "agentic era." Same shape, smaller dataset. PayPal cut 4,760 the same week. Coinbase cut 700. Freshworks, Arctic Wolf, and Ticketmaster all moved with explicitly AI-justified reductions. The category swap is the pattern, not the exception.

The released measurers are not your sourcing edge

If you recruit engineers and you saw "Cloudflare layoff" in your feed, the temptation is to scrape LinkedIn for "Open to Work" badges on ex-Cloudflare profiles and run a pitch sequence. Do it, and you will mostly hit program managers, FP&A analysts, internal auditors, and middle managers from marketing, the function Prince called "chock full of Measurers."

There are good people in that pool. They are not, generally, the engineers you were briefed to find.

The released cohort also is not desperate. Severance runs base pay through end of 2026, healthcare through year-end, and equity vesting extends to August 15, 2026 with one-year cliffs waived. Call it seven months of runway. Competing offers need to clear a higher bar than the typical post-layoff sourcing call, and the candidates know it. The "we caught you on a bad week" pitch does not land here.

The real edge sits in the other 1,894.

How to source the inbound bench

The 586 net new engineers Cloudflare added between early May and late June are the interesting cohort. They accepted into a stock that had just dropped 20%, which means their grants are priced near the floor. They are still in onboarding. Many have not yet built switching cost. And BNP Paribas's geographic cut suggests they are not concentrated in the Bay. Cloudflare's engineering clusters outside SF include Lisbon, Greater Seattle, Greater Boston, and Austin. If your sourcing model assumes Cloudflare equals San Francisco, you are missing roughly half the surface area.

The stack to filter for is Go, Rust, C and C++, and the Workers edge platform. That is a narrow enough fingerprint that boolean strings get you part of the way. It is not narrow enough to identify the right 100 out of 1,894 without reading commits, blog posts, and conference talks. This is the gap where prompt-based search beats keyword search, which is why we built Refolk: you describe the engineer in plain English ("Cloudflare Workers contributors who joined in the last 90 days, Rust-heavy, based in Lisbon or Porto") and get a ranked shortlist drawn from GitHub, LinkedIn, and the open web in one pass.

The Andreessen counter-read

Hold both truths. Marc Andreessen on 20VC said the quiet part loud: "Essentially, every large company is overstaffed. I think a lot of them are overstaffed by 75%. Now they all have the silver bullet excuse: Ah, it's AI." He has a point. Cloudflare went from 3,682 employees at the end of 2023 to 5,156 at the end of 2025, a 40% jump in two years. Some portion of the May cut is a standard post-2021 correction wearing an agentic-era costume.

That does not change the sourcing call. Even if the framework is partly retrofitted, the net engineering add is real, the AI-native screen is real, and the categories of role that got cut are real. The framework is doing work whether or not it was the original cause.

The released cohort is finance and middle management. The cohort worth a recruiter's time is the 586 builders who just badged in.

What this means for your pipeline next quarter

Three concrete shifts.

1. Reweight your Cloudflare list from outbound to inbound

If you maintain a target-company sheet, the "ex-Cloudflare, May 2026" tag is low signal for engineering roles. The "joined Cloudflare May to August 2026" tag is high signal, especially if you are hiring for an infrastructure or edge-compute role and you are willing to make a 12-month play on equity dynamics. TrueUp data backs the broader pattern: open tech roles are up 14% year over year in 2026, hardware engineering is up 52%, and the gains are concentrated in technical and product roles while ops, HR, and GM openings have declined. The builder side of the swap is system-wide.

2. Add an AI-native filter that is real, not theatrical

Asking "do you use Copilot?" tells you nothing. Cloudflare's interns were screened against the bar that 100% of production-code contributions are agent-reviewed. The equivalent screen on your side is asking candidates to walk through a recent PR where they used an agent, what the agent got wrong, and what they overrode. Engineers who cannot answer that fluently are not going to land at the new Cloudflare or GitLab or Coinbase, which means they are also not the senior hires you should be sourcing for any company running the same playbook.

3. Stop confusing WARN feeds with hiring intelligence

WARN notices, 8-Ks, and "X company laid off Y people" headlines are lagging indicators. They tell you a category contracted. They do not tell you which category, or whether an adjacent category expanded inside the same company in the same quarter. The Cloudflare case is the cleanest example of the year: the headline number was -1,100, the underlying engineering number was +586, and the public reporting did not catch the delta for almost two months. The same dynamic is running at GitLab and at the smaller May cohort. Refolk's sourcing flows are built for exactly this gap, because describing the person you want in plain English does not care whether their employer made a layoff headline last week.

The honest bottom line

Cloudflare in mid-2026 is one of the largest active engineering hiring markets in infrastructure. It is also one of the most misread, because the headline number and the underlying number point in opposite directions. The recruiters who reframe fastest get a clean shot at 586 newly arrived builders on depressed-equity grants in cities most of their competitors are not looking in. The ones still scraping "Open to Work" badges on ex-Cloudflare measurers are sourcing yesterday's company.

The Drucker categories do not care which side of the AI debate you sit on. The builders got bigger. The measurers got smaller. Source accordingly.

FAQ

Is the 45% engineering growth figure verified beyond the BNP Paribas note?

The 45% figure originates with the BNP Paribas analysis of LinkedIn profile data and was first reported by Business Insider in late June 2026. Matthew Prince has confirmed the directional read in subsequent comments. It is a LinkedIn-derived count, so it inherits LinkedIn's normal noise (profile lag, self-reported titles, contractors classified as employees), but the magnitude is large enough that even a meaningful error bar leaves engineering as a net add against a 1,100-person total cut.

Should I still pitch released ex-Cloudflare people?

For non-engineering roles, yes, with realistic expectations. Severance runs through end of 2026 with healthcare and extended equity vesting to August 15, 2026, so candidates have roughly seven months of runway. They are not in a rush. For engineering roles specifically, the released cohort is small and not where Cloudflare concentrated cuts, so the pool is thin relative to the inbound 586.

What stack and locations should I filter on for the new Cloudflare engineers?

Go, Rust, C and C++, and the Workers edge platform are the load-bearing technologies. Geographies skew more distributed than peers, with notable clusters in Lisbon, Greater Seattle, Greater Boston, and Austin alongside the Bay Area. If your sourcing model assumes SF-only, you are missing a large share of the surface area, particularly in EMEA.

How do I screen for "AI-native" without it becoming a buzzword check?

Ask for a concrete recent example: a PR or a feature shipped using an agent, what the agent produced, what the candidate overrode, and how they reviewed the diff. Cloudflare's internal bar is that 100% of production contributions are reviewed by autonomous agents, so the candidates clearing their loop can talk specifically about the override pattern. Candidates who answer in generalities ("I use Copilot a lot") are not the bar Cloudflare or its peers are now hiring against.