The Nexairi Dispatch · Friday, September 11, 2026 · Issue #37
Why does self-service AI need 1,000 humans?
Self-service software that ships with a thousand consultants attached. What does that tell you?
Good morning, friends. Meta shipped a personal AI agent this week that books your travel and moves your money. A separate deal has Google and Accenture training a thousand engineers to sit inside customer companies and make enterprise AI work. The New York Fed looked at the same trend from the ground and found AI is reshaping jobs more than it is cutting them. Today we remember the lives lost and lives forever changed 25 years ago.
🏢 ENTERPRISE AI — Enterprise AI's real cost is people, not compute
What happened: Google Cloud and Accenture said this week they will train up to 1,000 engineers to work inside customer companies on AI projects. The job is not building models. It is rewiring workflows and getting teams to change how they work.
Why it matters: This is the quiet admission that enterprise AI does not install like software. If you are budgeting an AI rollout, the model license is the cheap part. The expensive part is the months of process work and the people who do it, on your payroll or a consultant's invoice.
What to watch: Watch whether firms start naming this labor cost up front in AI pitches, or keep it buried in a services line nobody reads.
Outside Nexairi
Meta launches Muse, a personal AI agent for money and errands — Bloomberg
Muse runs shopping, booking, health and finance tasks in an isolated Linux machine, with a free tier plus $20 and $100 monthly plans. It is US only and 18 and up for now.
A startup rebrands as cfo.ai and ships an AI agent it calls a CFO — PR Newswire
The agent, named Ari, connects to a company's financial systems, keeps a live model of the business and flags changes to the founder around the clock. It is aimed at startups and solo owners with no finance hire.
New York Fed says firms use AI to change work, not cut headcount — Liberty Street Economics
Among companies that have adopted AI, the median share of workers actually using it was 17% at service firms and 7% at manufacturers. Layoffs tied directly to AI stayed rare in the survey.
Cognition raises over $2 billion at a $48 billion valuation — Cognition
The maker of the Devin coding agent said its revenue run rate climbed to roughly $900 million and that it is moving into robotics. The round lands as coding agents become a real budget line at software firms.
OpenAI claims a proof of a $1 million math problem, then a fight breaks out — TechCrunch
OpenAI said an unreleased model plus about 10,000 agents cracked the Navier-Stokes problem in 88 hours. A NYU mathematician and an Anthropic researcher say they got there first, and the priority dispute is now public.
Tool Worth Knowing: mnemiq
mnemiq turns plain-English questions into SQL and checks each query against user permissions before it runs. A finance lead can pull numbers from the warehouse without waiting on an analyst. It runs on open 14-billion-parameter models, so a firm can host it in-house and keep the data local.
Deeper Read
Powering AI Is an Architecture Problem — MIT Technology Review
MIT Technology Review argues the grid strain from AI is not just a supply issue but a design one, and lays out how data centers and utilities would have to rebuild to cope.
AI Is Eroding Trust. Finance Teams Can Rebuild It — CFO.com
A finance-desk argument that accountants and controllers are well placed to become the people who vouch for AI outputs inside a company, with a look at what that role means day to day.
State of AI in Finance 2026 — CFO Connect
A survey report on where finance teams are actually putting AI to work, what is stalling, and a roadmap CFOs are using to sort pilots from production. Useful counterweight to the vendor stat above.
Quick Hits
- Google DeepMind's AlphaGenome predicts the effect of every single-letter DNA change
- Claude produced a 13-million-line, computer-checked proof of Fermat's Last Theorem in 11 days
- ChatGPT's new image editor adds sketch and circle edits and 50% faster generation
- Writer's 2026 report: 79% of firms hit AI adoption snags, up double digits from last year
- CFOs lean harder on AI forecasting while ROI stays hard to pin down
- Healthcare AI's next test is integration, not model quality
- Nexairi: the $100 billion US-Korea AI power plan is mostly unsigned
- Nexairi: NYC bans student AI through 8th grade and bets on AI literacy later