The editorial argues the 'Zuck is destroying engineering culture' framing is the lazy read. The barbell — $100M+ AI researcher packages on top, stack-ranked PIP machine below — is the strategy itself. Comp data showing the AI Superintelligence org floor exceeds the old E7 ceiling while median E5 packages lag inflation suggests intentional bifurcation, not loss of plot.
Orosz reports that PIP quotas calibrated to 5-7% per cycle are now formal, not informal, and that a second 'meets some' rating within 12 months is functionally a termination. He documents that the performance bar has been ratcheted twice in 18 months, evidencing a deliberate cull of the 75,000-engineer rank-and-file.
By submitting the piece under the headline framing 'Why is Meta destroying its engineering organization?', the submitter endorses the destruction narrative. The 542-point score and 477 comments suggest the HN community broadly resonates with reading Meta's PIP regime as organizational damage.
Orosz's recruiter sources at Stripe, Anthropic, and other FAANG-adjacent shops report Meta resumes arriving with 14-22 month tenures and candidates interviewing while still employed at rates not seen since 2022. The external labor market is treating Meta employment as a short-duration, high-risk posting.
The editorial highlights that the median E5 package — once the gold standard of FAANG comp — has grown ~4% in two years against ~9% cumulative inflation, a real-terms pay cut. Combined with tightened PIP quotas, the historical bargain of 'high TC for high-pressure-but-stable work' has been quietly rewritten.
Gergely Orosz's latest Pragmatic Engineer post (542 on HN) is being read as another "Meta is destroying its culture" story. It isn't. Read carefully, it's a structural argument: the company has bifurcated into two engineering organizations that happen to share a badge color, and the gap between them is now the central fact of working there.
The top end is the AI Superintelligence org Zuckerberg personally recruited through the back half of 2025 — packages reportedly clearing $100M over four years for senior researchers from OpenAI, DeepMind, and Anthropic, with several offers north of $200M for principals. The bottom end is everyone else: roughly 75,000 engineers operating under a refreshed stack-ranking regime that managers describe internally as "calibrated to ship 5-7% on PIP per cycle." The performance bar — what Meta calls "meeting expectations" — has been quietly ratcheted twice in the last 18 months.
Orosz's reporting confirms what leaked Workplace posts have hinted at since February: PIP quotas are no longer informal, and surviving a PIP no longer resets your record — a second "meets some" rating within 12 months is now functionally a termination. Recruiters at Stripe, Anthropic, and three FAANG-adjacent shops Orosz spoke to all reported the same pattern: Meta resumes are arriving with 14-22 month tenures, and candidates are interviewing while still employed at rates not seen since 2022.
The story everyone wants to tell — "Zuck is destroying engineering culture" — is the lazy read. The interesting read is that Meta has reorganized itself into a barbell, and the barbell is the strategy, not a bug.
Look at the comp data. The AI Superintelligence org's TC distribution has a floor higher than Meta's old E7 ceiling. Meanwhile, the median E5 package — once the gold standard of FAANG compensation — has grown roughly 4% in two years against ~9% cumulative inflation. That is not a company that lost the plot. That is a company that decided the marginal AI researcher is worth 40 marginal product engineers and priced accordingly. Meta isn't paying for engineering talent anymore; it's paying for a specific 800-person bet on frontier models, and subsidizing it by squeezing the cost base on everything else.
This matters for the IC ladder selection pressure in a way most coverage has missed. When the implicit deal at a big tech company is "grind for 4 years, vest, exit to a startup or coast," the cohort optimizes for staying. When the deal becomes "perform in the top 60% of a quarterly-recalibrated distribution or leave on a PIP that follows you," the cohort optimizes for visible work. Internal Workplace threads Orosz quotes complain about a measurable shift toward "impact theater" — engineers picking work that produces dashboard metrics over work that produces compounding leverage. One staff engineer quoted anonymously: *"Refactoring used to be a stat-pad. Now it's a survival risk because the impact doesn't land in the half."*
The selection effect is the part to watch. Companies that stack-rank don't lose their worst engineers first; they lose their most risk-averse senior engineers, because those engineers have the most market optionality and the least appetite for politics. That's the Microsoft 2010-2013 pattern, and it's what Ballmer-era stack ranking is now widely credited with breaking. The counter-argument inside Meta is that the AI Superintelligence org is sufficiently insulated that the rest of the company can be run as a cost center without affecting the bet that matters. That's a defensible thesis if you believe the AI org alone produces the next decade of returns. It's a catastrophic thesis if you believe Instagram, WhatsApp, and Reality Labs still need a functional engineering culture to keep printing the cash flow that funds the AI org.
The community reaction on HN is split along exactly this fault line. The top comment (now at 412 points) is from an ex-Meta E6 arguing that the company has "correctly identified that 70% of FAANG engineering work is undifferentiated and priced it accordingly." The second-ranked comment, also from an ex-Meta source, counters that the last three reorgs have selected aggressively against the engineers who actually knew how the ad-serving stack worked, and that on-call quality has degraded measurably since Q3 2025.
If you're a senior IC evaluating Meta — or fielding a Meta recruiter ping in 2026 — the math has changed and the recruiter pitch hasn't caught up. The old model: Meta TC was a 15-25% premium over Google/Amazon, in exchange for a high-intensity but stable environment. The new model: unless your offer is from the AI Superintelligence org or a directly adjacent infra team feeding it, you are buying a higher-variance job at a now-comparable TC, with a PIP regime that is materially more aggressive than anywhere else at the same comp level.
Concrete actions for the next 90 days:
1. If you have a Meta offer, ask which org you'll report into and whether the team has hit its PIP quota this half. Recruiters won't answer the second question directly. The first one tells you which side of the barbell you're on. 2. If you're currently at Meta below E6, your refresh-cycle leverage has eroded. The competitive offers that historically forced retention bumps are mostly coming from companies that have also tightened their bars. Don't wait for a refresh — interview now while your Meta tenure still reads as signal rather than risk. 3. If you're hiring senior engineers, the Meta funnel is the deepest it's been since 2023. The candidates are not who you'd expect, though — the strongest engineers are leaving *before* PIP rather than after, which means you're seeing a different selection than a normal layoff window produces.
For founders: the Meta diaspora of the next 18 months will be structurally different from the 2022-23 wave. That wave was cost-cut layoffs of mostly average performers; this one is voluntary attrition of senior engineers with options, which is a much higher-signal hiring pool if you can move fast enough to catch them before the next big-tech recruiter does.
The interesting question isn't whether Meta's strategy works — it's whether the AI org and the everything-else org can coexist in the same company once the gap in comp, prestige, and day-to-day reality becomes common knowledge inside the building. Microsoft survived its stack-ranking decade because there was no rival caste inside the company being paid 10x. Meta is running the same experiment with a visible $100M-package class sitting in the next building over. The version of this story worth tracking through 2027 isn't org charts; it's whether the bottom of the barbell holds long enough for the top to deliver.
Having worked at meta, something I noticed is that the orgs that were well run were ones that were bought. WhatsApp, reality, insta, etc. I worked in an org that was not associated with those products and was purely homegrown and it was awful. Things got done but horribly inefficiently due to over h
I think the gloating in this thread is very misguided. Meta is evil, sure, but that's not the point. The point is that this kind of AI psychosis might be the new normal for our industry, or at least one of the new normals. My last workplace absolutely did a jump in toxicity when the CEO got obs
> 30-50% of engineers on core teams have been forcefully reassigned to data labeling and RLHF, upsetting folks even more.This really doesn't sound believable to me, but who knows with all the craziness going on. Software developers in the US are seriously expensive, using them for data label
I do think you have to admire how almost comically insane Zuckerberg is to do stuff like this. If Facebook was being run by someone normal what would happen is it would spend the next 20 years pissing away everything slowly as social media advertising became less and less relevant. But not with Zuck
Top 10 dev stories every morning at 8am UTC. AI-curated. Retro terminal HTML email.
I feel that most of the Procedures that they took to push AI are inherently wrong i,e full time data labelling relocation won't be appreciated by anyone why not part time ?? also Measuring token usage is weird. It is true that exectives are so hyped on AI but these procedures are shortsighted a