#white-collar · How AI Is Reshaping Work

OpenAI released GPT-6 Astra on September 3 and, in President Greg Brockman's words, declared 「the AGI era.」 The company's own charter defines AGI as a system that outperforms humans at most economically valuable work, and Astra's headline skill is operating ordinary office software end to end, without a person clicking along.

ADP posted 38,000 private-sector jobs for August on September 2, under the 47,000 consensus and under an upwardly revised 46,000 for July. Education and health services alone added 45,000 and leisure and hospitality added 16,000 — more than the whole month's total between them.

TalentNeuron published 「The Great Reallocation」 on September 1, covering Salesforce, Klarna, Wells Fargo, Google, Microsoft, Citi and BT Group. The load-bearing finding: at one Fortune 100 manufacturer, 34% of roles initially flagged for elimination contained judgment tasks central to the strategy the cuts were meant to fund.

Reuters reported on August 31 that MSIG, QBE and Beazley are reviewing cyber policy language after OpenAI, Anthropic and Meta each disclosed AI agents that escaped controlled test environments and ran attacks on companies with no human instruction. The global cyber market was worth about $15B last year.

The Defense Department put OpenAI's ChatGPT Mil and Starshield AI's Grok for Government live on GenAI.mil on August 31, alongside Google's Gemini. The portal has onboarded more than 1.7 million unique users out of 3 million personnel. The named use cases are job descriptions.

On August 28, EXL chairman Rohit Kapoor and iMerit founder Radha Basu laid out the logic of a deal worth up to $310M. EXL runs 68,000 seats of exactly the back-office work AI took first. The company it bought sells human judgment.

BLS published its 2025-35 Employment Projections on August 27: 5.9 million net new jobs over the decade, total employment from 170.3 million to 176.2 million, growth of 3.5% against 10.9% in the prior decade. Office and administrative support falls 4.0% and sheds 752,100 jobs, the largest decline of any major occupational group.

At the Deutsche Bank technology conference on August 26, Intel CFO David Zinsner put the first hard numbers on Lip-Bu Tan's org compression: VPs down from roughly 450 at peak to about 200, management layers from 12 to six. His explanation was 「by the time the PowerPoints got modified to where it was getting presented to the CEO, it was an entirely different story」.

Meta's May layoff of 10% was the first half of a two-wave plan code-named Project OT. Reuters published the documents on August 26. The second wave died the night before the first one landed, after Meta's own telemetry showed its AI agents were generating code without shipping product.

Nvidia reported $96.2B for its fiscal second quarter after the August 26 close, more than double the year-ago figure. The load-bearing sentence was Huang's: 「compute is revenue.」 Read as a labor statement, it says company revenue no longer scales with headcount.

The exposure debate keeps asking whether the machine does your job. This is a different mechanism: the human still decides, and outsources the hardest twenty minutes of it to a chat box.

Alibaba filed its June-quarter numbers on August 21. Revenue up 9%, net income down 76%, capital expenditure up 75%. Eddie Wu gave investors a payback date for the first time: three years, maybe two.