In partnership with
Attio is the AI CRM for high-growth teams.
Connect your email, calls, product data and more, and Attio instantly builds your CRM with enriched data and complete context. Whether you’re running product-led growth or enterprise sales, Attio adapts to your unique GTM motion.
Then Ask Attio to plan your next move.
Run deep web research on prospects. Update your pipeline as you work. Find customers and draft outreach emails. Powered by Universal Context, Attio's intelligence layer, Attio searches, updates, and creates across your data to accelerate your workflow.
Ask more from your CRM.
Anthropic refused a Pentagon contract on ethics grounds in March. In May they signed a compute deal with Elon's Colossus to keep Claude Managed Agents running. I'll come back to that one at the end.
This week
Two weeks apart, Anthropic and Google shipped the same product under different names. Anthropic calls it Claude Managed Agents. Google calls it Project Antigravity + Managed Agents API. Strip the branding and the architecture is literally identical: the model "brain" runs on the vendor's cloud, the tool-execution "hands" run in an ephemeral Linux sandbox the vendor also owns, sessions are stateful and survive for days, MCP is the wire protocol, and credentials live in a vendor vault you never touch directly. Both billed at runtime-seconds + tokens + tool-calls. Both charge $0.08/hr of active container time. Both ship a multi-agent orchestrator, an evaluator-LLM ("Outcomes" / lifecycle hooks), and a long-running memory layer ("Dreaming" / context compaction). Same blueprint, two logos.
What's actually being sold here is not intelligence (it's the control plane). The model is the easy part to swap; you can route Haiku, Sonnet, Gemini Flash, GPT depending on the task and nobody notices. But once your agent runtime, your MCP servers, your vault, your audit logs and your 30-day checkpoints all sit on Anthropic's or Google's infra, switching costs look like a cloud migration. That's the lock-in they're racing for, and it's why both rolled it out the same month. The middleware market (LangGraph, CrewAI, LlamaIndex) just got cannibalized from above. The model vendor now owns orchestration, memory, evals and the sandbox.
Test for tomorrow morning: open whatever agent you're running in production and count the things the vendor now provides natively that you used to glue together yourself. Sandbox, file system, MCP routing, credential vault, multi-agent loop, eval rubric, long-term memory. If three or more are now "managed", you no longer own your runtime (you rent it). Price your exit before that number hits five.
In the news
Anthropic & Colossus (the part nobody wants to write about). Anthropic spent two years building the "ethical" lane of frontier AI, refused Pentagon work in March 2026 on safety grounds, then in May quietly signed for inference capacity on Elon's Colossus cluster. The same compute that trains Grok. The official line is "diversified supply". The real line is: 80x revenue growth in Q1, Managed Agents sessions burning container-hours by the millions, and not enough clean GPUs on the planet to keep the lights on. The "we're not OK with this, but it's hard to do otherwise" energy is impossible to miss 😬. Full breakdown.
This week on the blog
Anthropic Refused the Pentagon on Ethics. Two Months Later They Bought Elon's Dirty Compute. — The compute math that broke the ethics narrative.
Microsoft Just Killed Claude Code Internally. Their Own Devs Loved It. — Strategy tax in action: when the org chart beats the better tool.
I Deleted My Last n8n Workflow. Convex Ran the 30-Minute Job in 80 Lines of TypeScript. — The opposite move: pulling runtime back in-house.
Revenue Stack (May 2026)
Most "AI agent" side-projects die at the same place: the bill. Not because the model is too expensive... because nobody measured what the agent actually consumes per task before scaling it. The fix is boring: a unit-economics sheet before the second prompt. Token in, token out, container-seconds, tool calls, expected runs per user per month. Three numbers and a calculator beat any pricing page.
A weekend read if you're shipping AI features without measuring what they cost.
Where does your agent runtime live today?
Phil
PS: hit reply with the one piece of your stack you'd refuse to hand over to the model vendor, no matter how cheap the managed version got. I want to see where the line sits for people who actually ship.
Back to Anthropic and Colossus: the ethics statement and the compute invoice were never going to coexist forever. Managed Agents at 80x growth needs GPUs the way a city needs water. At some point you stop asking the supplier where the river starts. That's the trade. Nobody at Anthropic is celebrating it; nobody is walking away from it either. The cage is comfortable, and we all helped build it.
You're receiving this because you signed up at rentierdigital.beehiiv.com.