Leadership1 distinct publisher3 min readUpdated
The Pragmatic Engineer reports staff and principal engineers leaving for Anthropic, OpenAI and Google despite one-off grants of $400K to $1M+. The cost saving ended. The retention damage did not.
The Board Room · Leadership desk
Compiled by The Board RoomSomething wrong?How this is made
Meta cut 10% of staff while revenue and profits were at an all-time high, then reassigned roughly 20-30% of its software engineers into data labeling with basically no notice [1][2]. Two months on, according to Gergely Orosz's Pragmatic Engineer newsletter, the company is writing one-off equity grants of $400,000 to more than $1m, vesting over three years, to keep the senior people who read those two moves and resigned [9][21].
The sequence matters more than the sums. Orosz reports that engineers who were not reassigned also started interviewing, which is the ordinary consequence of a reorganisation that tells everyone their work can be swapped for annotation duty overnight [3]. In the UK, where mass redundancies require prior notification, Orosz confirmed that a good chunk of notified staff were later "un-notified" [5]. By then they had already begun job-hunting [6]. That is the whole lesson in one detail: the notice period does the damage, and withdrawing the notice does not undo it.
Meta's response has been to abandon a long-standing policy. Long-tenured engineers told Orosz that the company simply did not counteroffer or negotiate when someone resigned; it now does, specifically for people leaving for Google, Anthropic and OpenAI [7][8]. Orosz spoke with seven recipients, all at IC6 (staff) or IC7 (principal); none were aware of IC4 or IC5 engineers being offered anything [10][11]. The largest grants, above $1m, all went to engineers holding Anthropic or OpenAI offers, while confirmed $400K and $600K grants went to engineers with offers from smaller AI startups [12]. Two recipients were not asked to produce an offer letter at all: their director and HR offered discretionary retainer equity once resignation was handed in [18].
Retainers priced against a competing offer become ammunition for that offer. In one case Orosz confirmed, an engineer force-reassigned to a data labeling team interviewed out, got a Google offer below their current pay, and decided to leave anyway [13]. Their new manager produced a large three-year retainer grant; the engineer took the number back to Google, Google outbid it, and they left [14]. The pattern holds in aggregate. Three engineers with Anthropic offers were countered with $1m-plus grants vesting over four years; two declined immediately and the third accepted, then left a month later and forfeited the grant [15][16]. Nobody in that group stayed [17].
Anthropic and OpenAI are now the main landing spots, and Orosz notes both can match Meta's total compensation, run secondary equity sales while private, and have plausible IPO paths [19]. That neutralises the one argument a public company usually wins on.
Watch whether retainers spread below IC6, which would signal the attrition is no longer confined to the top two levels [11]. Watch how many accepted grants are forfeited mid-vest, as the third Anthropic hire's was, because money does not fix an assignment problem [16]. And watch whether the data labeling reassignments are quietly reversed [2].
Follow any of these and your For You feed starts watching them — no settings page required.
Ranked by verification strength, evidence, and original report placement.
A good chunk of software engineers at Meta, including those not reassigned, started interviewing elsewhere.
Meta carried out 10% layoffs at a time when its revenue and profits hit an all-time high.
Meta reassigned about 20-30% of its software engineers to data labeling with basically no notice.
In the UK, mass layoffs require a notification to those potentially affected before cuts can happen; a few weeks after this notification, a good chunk of Meta people were 'un-notified', which Orosz confirmed.
Those who had been put on notice in the UK had already started to look for jobs.
Long-time Meta engineers said Meta simply did not make counteroffers or negotiate when an engineer handed in a resignation, but this practice has been abandoned.
Evidence-backed comparisons of source perspectives and observed adoption signals. Read the methodology
Which Builder, Operator, and Investor concerns the observed source mix emphasized—not a truth score.
Evidence, demonstrated adoption, hype gap, incentives, and confidence are assessed independently, each on its own current evidence. How these are measured.
Single-outlet insider sourcing, no company response
Everything rests on one newsletter issue in which the author says he confirmed specific cases with named-level but anonymous engineers. The individual anecdotes are detailed and internally consistent, and one engineer's written self-assessment is quoted at length, but there are no documents, no Meta statement, no independent corroboration, and the aggregate figures (10% layoffs, 20-30% reassigned) are restated from the author's own earlier reporting rather than externally verified.
Practice observed in a handful of confirmed cases
There is concrete evidence the retainer practice is actually in use rather than proposed: seven interviewed IC6/IC7 recipients, individually confirmed $400K and $600K grants, three $1M+ Anthropic counter-offers, and one case where the grant was used as leverage with Google. But no company-level counts, spend totals or attrition rates are disclosed, and the claim that retainers go to 'most IC6 engineers who resigned' remains an inference from a small sample.
Specific cases generalised into an organisation-wide wave
The headline figure of $1M retainers is directly supported by the reporting, and the 'zero out of three retained' outcome is stated with its sample size. Overstatement comes from scale language: 'bleeding top talent', 'hiring spree' and 'most IC6 engineers who resigned' extend roughly ten individual confirmed cases into a company-wide trend, with no denominators, no Meta response, and an unverified expectation of AI-lab IPOs within 3-12 months. The source also includes a counter-example, an engineer who chose to stay after a reorg gave them a supportive manager, which sits against the sweeping framing.
Scoop-driven subscription newsletter plus self-interested sources
The article is an explicitly promotional free issue of a paid newsletter ('Full subscribers received the article below seven days ago... you can subscribe here'), so exclusive insider detail about Big Tech dysfunction directly serves conversion. The author also has a stake in validating his own two-month-old thesis that Meta made a 'self-inflicted error'. Sources are anonymous engineers negotiating or having just negotiated compensation, who benefit from high reported retainer benchmarks and from an unflattering account of Meta management; Meta is given no voice.
Credible individual cases, weak basis for scale
Confidence is moderate for the specific mechanics, that Meta now issues discretionary seven-figure retention equity to staff and principal engineers, sometimes without an offer letter, and that this has repeatedly failed against Anthropic and Google. Confidence is low for the magnitude, direction and financial consequence claims, given one publisher, anonymous small samples, self-referential baseline figures, an unrebutted subject company and a clear commercial incentive for the framing.
build
Meta is now paying $400K-plus retainers to keep engineers it reassigned to data labeling1 distinct publisher
product
Washington's secret AI test is coming for open weights, and release dates go with it2 distinct publishers
security
The nationalization argument is really a vendor-continuity memo1 distinct publisher
invest
Google Ships Flash Instead of Pro While OpenAI Loses Its Two Best Operators1 distinct publisher
Distinct publishers with included, body-backed reporting in this cluster.
1 article · August 21, 2026