I Drink Your Milkshake

$1.5 billion — what Anthropic, Blackstone, Goldman Sachs, and Hellman & Friedman committed Monday morning to a joint venture that will embed Anthropic engineers directly inside the operations of mid-sized companies. Apollo, General Atlantic, Sequoia, Leonard Green, and Singapore's GIC piled in alongside. OpenAI is reportedly running the same play with TPG and Bain. The labs were already selling picks and shovels. Today they started selling the miners — engineers trained on the model, paid for by Wall Street, loyal to the lab. Phase one was you firing your analysts. Phase two is them coming back on Anthropic's payroll. I drink your milkshake.

Decide which lab you are willing to be permanently dependent on. The implementation business locks the model in for the duration of the contract — three to five years minimum, longer in practice. Six forward-deployed engineers are not switching their stack mid-engagement, and you are not switching theirs. Pick the lab whose model, culture, and pricing trajectory you can live with for the long contract. The default answer is the answer. If your CIO does not have a strong default by end of week, you are already behind the operators who do.

Move on your existing Big 3 contract before summer. McKinsey, Bain, BCG, and Accenture know what landed this morning, and the leverage in the room just shifted. Their lab-aligned implementation businesses will land by Q3, but for the next sixty to ninety days the buyer holds the cards. Renegotiate. Compress duration. Insert a six-month off-ramp clause. Do not extend a multi-year transformation contract at last year's pricing. They will not raise rates this summer; they might lower them. Move now.

Start the data-readiness audit today. SAP just paid $1.1 billion to back the position that the model is no longer the bottleneck — "it stalls because the data isn't ready for AI agents," per CTO Philipp Herzig. The labs will sell you the implementation team. They will not sell you the cleaned, governed, agent-readable substrate the team needs to actually do work. That layer takes nine to eighteen months to build. Six forward-deployed engineers walking into a 2017-vintage data lake produces a 2017-vintage outcome. The clock starts now.

Today's actions touched directly on lab-partnership selection, data-readiness audits, and the non-PE mid-market squeeze — three of the most common conversations we are having with clients this quarter. If you are staring at "decide which lab to be permanently dependent on" and aren't sure how to run that decision, that is the conversation we are built for.

Outsider Labs Advisory →

Monday, May 4

1. Anthropic just bought McKinsey — and Wall Street wrote the check. Anthropic, Blackstone, Goldman Sachs, and Hellman & Friedman launched a $1.5 billion joint venture Monday morning to embed Anthropic engineers inside mid-market companies, starting with the founders' portfolios and expanding from there. Apollo, General Atlantic, Sequoia, Leonard Green, and Singapore's GIC came in behind. OpenAI is reportedly running the same play with TPG and Bain. The frontier labs always copy each other product-for-product. They will copy each other firm-for-firm now too. (Full analysis below.)

2. SAP put $1.1 billion on the line to say the model isn't the problem. SAP committed $1.1 billion-plus over four years to scale Prior Labs, the German startup pioneering Tabular Foundation Models for structured business data, and announced the acquisition of data lakehouse Dremio in the same press release. SAP CTO Philipp Herzig: "Enterprise AI doesn't stall because the models aren't good enough; it stalls because the data isn't ready for AI agents." That sentence is the thesis of the day, said by the company writing the biggest check on it.

3. A 2024 model still beats the ER doctor. Harvard published in Science this morning. OpenAI's o1-preview — released in 2024, two model generations behind the frontier — diagnosed 76 real ER cases at 67.1% accuracy versus 55.3% and 50.0% for two attending physicians. Blinded reviewers couldn't tell which diagnoses came from the AI and which came from the humans. In one case the model flagged a rare flesh-eating infection 12 to 24 hours before the treating doctor. If a 2024 model already wins, the operator A/B-testing 2026 models is asking the wrong question.

4. The patch wave is becoming a product. UK NCSC issued formal guidance this week that AI-driven vulnerability detection is producing critical updates faster than the global patching infrastructure was built to absorb. Theori's "CopyFail" exploit took an hour from discovery to a 732-byte script granting root on every major Linux distribution shipped since 2017. Anthropic's Mythos has reportedly found 2,000+ unknown flaws including a 27-year-old OpenBSD bug — 99% still unpatched. Anthropic Claude Security entered public beta the same week. Patching is becoming a feature, and the labs are going to sell it back to you.

5. The Pentagon picked seven and skipped Anthropic. The U.S. military signed classified AI agreements with SpaceX, OpenAI, Google, Microsoft, AWS, Nvidia, Reflection, and Oracle. Anthropic was excluded — the company has refused to participate in classified networks over autonomous-weapons and surveillance concerns. The White House is reportedly trying to access Anthropic's Mythos cybersecurity model anyway. Same week Anthropic joined hands with Wall Street and walked into the consulting industry, it walked away from a multi-billion-dollar federal contract pool. Pick which one of those tells you more about who they are.

The Lab Did Not Buy McKinsey. The Lab Hired Your Old Analysts.

The Wall Street Journal had it first this morning. Anthropic, Blackstone, Goldman Sachs, and Hellman & Friedman launched a $1.5 billion joint venture that will embed Anthropic engineers directly inside the operations of mid-sized companies — starting with the founders' portfolio firms and expanding outward. Apollo, General Atlantic, Sequoia, Leonard Green, and Singapore's GIC piled in alongside. OpenAI is reportedly running the same play with TPG and Bain. The structure mirrors Palantir's forward-deployment model. The pitch is a clean shot at McKinsey, Bain, BCG, and Accenture — combined with the lab's ownership of the model running underneath the work.

For three years the labs were selling picks and shovels. This morning they started selling the miners. The implementation team is the labor force. The labor force has been rebranded as a vendor relationship. The vendor reports to Anthropic. Anthropic reports to its cap table. You report to Goldman.

The brutal sequence is short. Phase one — eighteen months running — was the labs telling you AI would replace your knowledge workers. Klarna at 700 customer-service jobs displaced. Block at 4,000 roles cut. Atlassian, Duolingo, Oracle, Salesforce, Meta — every company on the list ran the same arithmetic. The math always looks good on the spreadsheet. Phase two arrived this morning. Six forward-deployed Anthropic engineers, embedded for a year, paid for by your PE sponsor, working on your operating P&L. The agent in the seat is loyal to the lab. The engineer next to the agent is loyal to the lab. You used to own the brain. Now you rent it. From the lab.

This trade works because of math the consulting industry has run on for forty years. For every dollar enterprises spend on software, they spend six on services. The senior partner at Kirkland still gets the call over Harvey. The Bridgewater PM still gets the call over an AI fundamental analyst. Years of judgment and decades of relationships do not get replaced. Every other rank in the firm is now competing with a model smarter than almost everyone on earth and a team that knows how to drive it. The bandwidth layer was always the cost line. The bandwidth layer is what gets eaten.

The mid-market is three markets in this trade. Fortune 500 will be fine — the Big 3 will launch their own lab partnerships by Q3 and the leverage stays with the buyer through summer. PE-owned firms will be served first and well; the venture has roughly 11,500 portfolio companies in pipeline before it has to sell elsewhere. The non-PE mid-market is the one to worry about. The implementation team Anthropic is staffing will be sized to serve the founders' portfolio first. The good independents will book solid by Q3. The model gap between the lab-aligned bench and the independent bench will widen every quarter the labs gate frontier capabilities to their own ventures. Where do you think Mythos and its successor models get deployed first? Not at the firm your CIO just hired.

The honest version of today's news is SAP CTO Philipp Herzig's, said while announcing a $1.1 billion bet that the model is no longer the bottleneck: "Enterprise AI doesn't stall because the models aren't good enough; it stalls because the data isn't ready for AI agents." Every plumber on the field said it differently on Monday. They all said the same thing. The labs heard the architects. Implementation businesses are how the labs monetize the next decade of operational catch-up. The clock starts now.

At COAI today: Full analysis at getcoai.com, including the Brockman diary as federal evidence in the Musk vs OpenAI trial — and what it means for your AI vendor's working journal. We also published a separate essay this morning, Borrowed Authority, on a sourcing error in last night's issue and the larger pattern it surfaced about whose voice is on the page in 2026.

— Harry and Anthony

Sources: