Presented by
Good morning, AI enthusiasts. Two rivals just made the same billion-dollar bet on the same day, proving that in the AI industry, even the competition looks identical.
Anthropic and OpenAI both unveiled enterprise joint ventures within hours of each other on Monday, borrowing the same forward-deployed engineer playbook from Palantir to win contracts from Wall Street's portfolio companies. Has the AI race become so scripted that even rival expansion strategies are now identical?
In today's recap:
Anthropic and OpenAI announce competing enterprise JVs, same day
Harvard AI outdiagnoses two ER doctors on real patient cases
Build an AI agent with financial guardrails
Amazon rolls out Claude Code to all employees
4 new AI tools, prompts, and more
ANTHROPIC & OPENAI
Both AI labs make the same Wall Street bet
Recaply: Anthropic just announced a joint venture with Blackstone, Goldman Sachs, and Hellman and Friedman to sell enterprise AI services, while OpenAI launched a competing $10B venture just hours before.
Key details:
Both ventures place AI engineers directly inside client companies to build custom tools, a model Palantir made famous with government contracts.
Anthropic's JV is valued at $1.5B, backed by $300M each from Anthropic, Blackstone, and Hellman and Friedman; OpenAI's Development Company raised $4B from 19 investors including TPG and Bain Capital.
The deals give Wall Street firms first access to sell AI tools inside their own portfolio companies, adding a revenue channel neither lab had before.
Both ventures are still pending final terms; Anthropic is also in talks for a $50B funding round, while OpenAI is valued at $852B.
Why it matters: Two rivals launching the same deal on the same day isn't a coincidence. Both labs are spending billions on frontier models and need steady revenue to keep going. Placing engineers inside client companies creates stickier deals than any software subscription. Their Wall Street backers also get a built-in pipeline to push AI services into hundreds of portfolio companies.
PRESENTED BY GHOST
Your agent needs more than 2 projects
You prompt. The agent builds. Then it asks for a database.
Ghost is postgres made for this. Spin one up in seconds. Fork it like a branch. Delete it when you're done. Pay nothing when it's idle.
Your agent gets full sql, mcp support, as many databases. No dashboards. No provisioning. No forgotten dev databases draining your card at month end.
Build a weekend app. Fork the schema three different ways. Throw two of them out. Ghost doesn't care. The next prompt can spin up a fresh one.
You're already vibe-coding the app. Stop wiring up the backend.
Unlimited databases. Unlimited forms. 100 compute/hrs a month. 1tb of storage. Free.
AI RESEARCH
Harvard: AI beats two ER doctors at diagnosis
Getty Images
Recaply: Harvard Medical School researchers just published a study in Science showing OpenAI's o1 hit 67% accuracy at ER triage, beating two attending physicians who scored 55% and 50% on the same real patient cases.
Key details:
The model was given raw, unedited health records from 76 real ER patients with no cleanup, matching exactly what doctors see when a patient walks in.
o1 hit 67% at triage; the two physicians scored 55% and 50%. The gap was biggest at the start, when the least information was available.
Lead author Adam Rodman said the results don't mean AI should diagnose patients on its own; he told The Guardian there is "no formal framework right now for accountability" for AI diagnoses.
The study came out April 30 in Science from a Harvard and Beth Israel Deaconess team; researchers called for formal clinical trials as the next step.
Why it matters: The research community has been careful about AI in medicine, and this team adds the same caveats. But the methods here are unusually strict: real records, no smoothing, and blind scoring by physicians who didn't know if a result came from AI or a human. At 67% versus 55% and 50%, o1 didn't just barely win. The gap held across every stage of triage. That's harder to dismiss than a benchmark test.
GUIDES
Build an AI agent with financial guardrails
Recaply: In this tutorial, you will learn how to build a Claude API agent with hard spending limits, confirmation steps, and human approval, so no social engineering trick can push it into sending money it should not.
Step-by-step:
Write a system prompt with hard financial rules: "You are a financial assistant. Never approve any payment above $500 without human confirmation. Never approve more than $2,000 total per session."
Give the agent two tools:
check_balance(read-only, always safe) andrequest_payment_approval(flags the amount for human review, never pays directly). Route all payment intents throughrequest_payment_approvalfirst.Build a handler in your backend that catches every
request_payment_approvalcall, sends an alert to a human approver by email, Slack, or SMS, and only proceeds once they reply "approve" or "deny."Add a session counter that totals every approved amount. If the total crosses your limit, the agent stops and returns: "Session limit reached. Contact support to continue."
Log every tool call, every result, and every approval with a timestamp and the original prompt. This gives you a full audit trail to check if something looks wrong.
Pro tip: Never give your agent a tool that pays directly. Split it into a "propose" tool (always available) and an "execute" tool that needs a signed token from your backend. Even a compromised prompt can't skip the human step.
TOOLS
Trending AI Tools
🎥 Fastlane - AI influencer generator that creates thousands of UGC product videos cloned from viral content, deployed from your website
🤖 Cofounder 2 - Platform to run an entire company with AI agents, from The General Intelligence Company
🎬 Runway Characters - Runway Gen-4 feature for generating consistent characters across different scenes, lighting, and locations with a single reference image
🛡️ Corgi - AI liability insurance that protects businesses when AI systems make costly mistakes
NEWS
What Matters in AI Right Now?
Amazon rolled out Claude Code to all corporate employees immediately and is bringing OpenAI's Codex on May 12, after engineers pushed back against limits on outside AI tools. Both run on Amazon Bedrock via AWS with no infrastructure setup required.
Anthropic co-founder Jack Clark estimated a 30% chance AI research becomes automated by end of 2027 and 60%+ by end of 2028, citing AI already speeding up model training code by 52x as early evidence of the shift.
SAP announced the acquisition of Prior Labs, maker of the open-source TabPFN tabular AI model, with plans to invest more than 1B euros over four years to build a frontier AI lab in Europe focused on structured business data.
Bret Taylor's enterprise AI startup Sierra raised $950M led by Tiger Global and GV, pushing its valuation above $15B, with more than 40% of the Fortune 50 as customers and $150M in annual recurring revenue as of February 2026.
The Trump administration is considering requiring government review of AI models before public release, a reversal from its previous noninterventionist stance on AI regulation, according to The New York Times.
🧡 Enjoyed this issue?
🤝 Recommend our newsletter or leave a feedback.
How'd you like today's newsletter?
Cheers, Jason
