The agents left the pilot phase this week, and the fight moved to who governs them: Google folded all of Vertex AI into a single platform that sits over every agent you run, while a two-month-old OpenAI subsidiary went shopping for the people who actually deploy them. Neither move is about a smarter model; both are about who holds the ground the agents stand on. Follow the thread from the platform down to the price and the lock-in it quietly buys.
Your auditor just became your landlord
Google quietly collapsed all of Vertex AI into one Agent Platform, with identity, a registry, and a gateway sitting over every agent in your fleet. The part nobody flagged in the launch copy: the tool that governs your agents also owns the ground they stand on. See what changed →
A two-month-old company went shopping
Governing agents is one problem; someone still has to stand them up inside real companies. So OpenAI’s brand-new Deployment Company bought Palantir’s only elite partner, a shop that is roughly 90 percent ex-Palantir engineers. The question is why buy the boutique instead of just hiring. Read the logic →
The math on embedding 8,900 people
Talent that scarce gets expensive fast, so the outsourcers are industrializing it. The world’s largest IT outsourcer plans to embed up to 8,900 engineers inside client walls, Palantir-style, and the headcount is not the surprising part. See the play →
Sol, Terra, Luna, and one moving line item
All those forward-deployed engineers end up optimizing the same line item: the model bill. GPT-5.6 ships as a good-better-best ladder named Sol, Terra, and Luna, running from $5/$30 at the top down to $1/$6 at the bottom, which means the purchase decision just stopped being “which model” and became “which routing mix.” Do the math →
The cheapest yes is the expensive one
A price ladder that low makes agents easy to say yes to across finance, supply chain, and IT ops, and that is exactly where the trap is set. As the agents move into the systems that run the business, the next fight is not adoption but exit. Read the warning →
One number
8,900 engineers. That is one outsourcer’s plan to embed forward-deployed staff inside its clients, and it tells you the scarce resource is no longer the model but the people who can make it work in the building. See where they’re going →
One thing to use
The enterprise buyer’s guide to agentic AI lock-in. Before you sign for the platform, the deployers, and the cheap tokens, this is the checklist for keeping an exit. Open the guide →
Platform, people, price, then lock-in: the agents went to work this week, and every layer underneath them turned into a market. Hit reply and tell me which story deserves the full treatment next; I read everything. Forwarded this? Claim your own copy.
Dr. Joseph Joshua
P.S. Next issue: the AI that stopped reading screens and started reading gauges and brain scans.
AI and business tech news, verified by a physician who reads the filings. One email a week, no noise.
