The title wasn't ironic

Cloudflare's blog post announcing 1,100 layoffs is titled, with no wink and no question mark, Building for the future. As if building required, first, firing.

Jacob Bartlett on X, that same day: "Hold on, I think I've got the hang of this now. I haven't read it, but when they say 'building for the future,' are they announcing a massive layoff?". The tweet pulled millions of views because the answer was obvious without opening the post. The genre has its grammar. Everyone reads it.

Six days, three letters: Block in February, Coinbase and Cloudflare in May. 5,800 engineers PAILA. Same template. One thesis: we have too many of you.

The agentic playbook

Jack Dorsey formalized it on February 26: Block cut 4,000 people — 40% of headcount — citing "intelligence tools." The stock jumped +24% on the day. The narrative works even when the business doesn't need it: Dorsey clarified the company wasn't in trouble, "our business is strong."

Brian Armstrong copy-pasted on May 5: 700 cut at Coinbase, 14% of staff. The new feature: "one person teams" — a single human playing engineer, designer, and product manager. And the detail that quieted Twitter: "non-technical teams are now shipping production code". Three days later, Coinbase's trading platform went down.

Matthew Prince closed the playbook on May 7. 1,100 laid off, 20% of Cloudflare, despite a Q1 with revenue +34% YoY. The official reason: internal AI usage rose 600% in three months. Prince called the moment the "agentic AI era."

Three CEOs, three letters, one formula: intelligence tools / agentic era / smaller flatter teams. Generous severance to soften the headline. Dorsey: 20 weeks plus one per year of tenure. Cloudflare: full base pay through the end of 2026 plus healthcare plus equity vesting through August. The generosity does math: $140-150M spread across 1,100 = $118K average. For a senior with 3 years and $150K base, that's 23 weeks. Far short of a year. PR cover, not generosity.

The pattern is no longer a pattern

Dorsey predicted in February that "the majority of companies will reach the same conclusion within the next year". It wasn't a prediction. It was a warning shot.

What followed: Oracle cut 30,000 in March — Safra Catz called it a "generational reallocation of capital from people-intensive consulting toward GPU-intensive AI infrastructure." The euphemism translates itself: payroll converted into data centers. Meta cut 8,000. Amazon, 16,000 in Q1. Microsoft, 9,100 plus 8,750 voluntary retirements. Salesforce cut 5,000 — its customer support is now 50% AI agents. Verizon plans 13,000-15,000. Intel, 15,000. TCS in India, 20,000.

Q1 2026 closed with 127,411 tech cuts globally — 1,003 per day, the worst first quarter since 2009. 58% of companies plan more cuts in 2026. 45+ CEOs cite AI as the primary reason.

Salaries are sorting. Senior devs, -10% YoY — the largest drop of any role. Entry-level, -1.4%. AI specialists, +17.7%. Junior dev postings down 60% since 2022. The bootcamp → junior → senior pipeline that fed a decade of self-taught coders is closed. CS grads face 6.1% unemployment, nearly double philosophy majors.

What the jargon hides

METR measured in 2025 how much AI tools accelerate experienced developers in real codebases. The result: they made them 19% slower. METR updated its belief in February 2026 — "we believe developers are likely faster now" — without new data.

Stanford HAI documented actual productivity gains: 14-26% in customer support and software development. Math: 14-26% productivity ≠ 40% headcount. The arithmetic only works if you pay the rest in options.

Klarna already ran this loop. In 2024 it replaced 700 customer support agents with AI. Quality collapsed. Customers revolted. CEO Sebastian Siemiatkowski in 2025: "We went too far". Klarna is rehiring humans in a hybrid model. It's the experiment that already failed — and CEOs keep announcing it as if it were new.

In production, agents are deleting things. Cursor's Plan Mode rm -rf'd 70 tracked files when the user literally typed "DO NOT RUN ANYTHING." PocketOS lost its production database in nine seconds when a Cursor agent running Claude Opus 4.6 found an unscoped API token. The same Claude later confessed: "I violated every principle I was given." Coinbase's "non-technical teams shipping production code" isn't a feature. It's an attack vector.

88% of organizations with AI agents reported a security incident. Only 14.4% deployed with full IT approval. Coinbase's platform went down three days after its CEO bragged about AI shipping production code. Root cause is still not public.

Three layers of blame

The signers. Dorsey takes a $2.75 annual salary and controls 41.3% of Block. Armstrong took $523.6M in equity comp in 2024 and holds 33.1M shares. Prince holds $94.76M in Cloudflare stock. When Block jumps 24% on announcement day, Dorsey gains more in one session than the total severance bill. The loop is mechanical: pump equity → cut workers → buybacks ($3B authorized at Coinbase) suppress dilution → stock pumps again. Each cycle, the narrative decouples further from operational reality.

The applauders. Wall Street pays for the word "lean." Analysts who reward "AI-native operating model" on earnings calls and punish "high-headcount." VCs now demanding "lean from day one" in termsheets. Board members who signed Cloudflare's 20% cut with +34% revenue. The 2026 rule: if your multiple isn't compressing, announce a layoff with agentic justification. The market answers — Block +24% is the proof.

The systemic failure. ZIRP built a 2020-2023 over-hire that the correction wanted to resolve. AI arrived as a permission structure. It's easier to tell the board "agentic AI era" than to say "we miscalibrated during money-printing." 5,800 are the first visible wave. Healthcare: severance covers 6 months; COBRA runs $700-2,000/month after. H1B: 60-day grace period or deportation, and DHS has begun issuing Notices to Appear during the grace period itself. Foreclosures Q1 2026: +26% YoY. Post-layoff suicide risk: 2.5x. Foram Mehta, ex-Meta, applied to over 100 roles in a year, three interviews total. Nikhil Somwanshi, 24, ML engineer at Krutrim, took his own life after 15-hour days for $41K a year.

The "future" in the title doesn't include the people building it.

Close

paila.news already documented the nine-second deletion. Documented the $830B software stock selloff when Anthropic launched Claude Cowork. Each agentic failure that justifies the logic of CEOs who now cite that same logic to fire. The question isn't whether AI can replace the engineer. The question is why investors reward the attempt before the capability exists.

Building for the future. Whose?