The Loop Is the Moat: How Anthropic, OpenAI, and Sequoia Just Confirmed My Thesis
Where I revisit my Transformation Triangle thesis and explain why this past month has been one long act of public validation
I developed a thesis last August I call the Transformation Triangle. The thesis was straightforward and, at the time, not yet obvious. In a world where AI capabilities commoditize on a quarterly cadence, no single moat survives. Not software. Not expertise. Not credentials. The only defensible position is the integration of three elements at once: tools that scale, expertise, and education that transfers capability. Loops, not moats. I built the argument for professional services, but it applies to any business.
Five months later, in February, I returned to the framework to show that the legal AI leaders were already running it. Harvey at an eight-billion-dollar valuation while building Legal Engineers from Wachtell and Latham, then opening Harvey Academy and partnering with seventeen law schools. EvenUp at two billion with more than a hundred in-house experts reviewing every output. Luminance launching academic certifications across three continents. Spellbook in fifty-plus law schools. The market was converging on the thesis.
In April, I refined the framing one more time. What I had called a triangle, three pillars holding up a roof, turned out to be a dynamic system: three currents in a single flow, each leg feeding the next. Software, Services, Learning, closed into a loop. The loop is the moat. Anything less is a feature waiting to be absorbed.
If this sounds interesting to you, please read on...
This substack, LawDroid Manifesto, is here to keep you in the loop about the intersection of AI and the law. Please share this article with your friends and colleagues and remember to tell me what you think in the comments below.
Eight Days That Settled the Argument
On May 4, Anthropic announced a new enterprise AI services company. The founding partners were Blackstone, Hellman & Friedman, and Goldman Sachs, with backing from a consortium of alternative asset managers that included General Atlantic, Apollo, GIC, Leonard Green, and Sequoia. The company’s purpose, in the announcement’s own words, is to embed Applied AI engineers from Anthropic alongside the firm’s engineers, working inside mid-market customer organizations to build production Claude systems tailored to those organizations’ operations. The engagement begins by understanding where Claude can have the biggest impact and ends with systems wired into daily work.
Anthropic’s CFO, Krishna Rao, gave the rationale: “Enterprise demand for Claude is significantly outpacing any single delivery model.” Translation: selling the model is not enough. The customer needs the work delivered, not the tool licensed. Or as Richard Susskind might say, the customer needs to hang a picture frame and doesn’t care if it’s accomplished with a hammer and nail or a drill and screw.
On May 12, eight days later, OpenAI acquired the consulting firm Tomoro and launched the OpenAI Deployment Company. The investor consortium included Bain Capital, SoftBank, Warburg Pincus, Capgemini, McKinsey, and Goldman Sachs. The structure was identical in every meaningful respect. Forward Deployed Engineers, the model Palantir made famous, embedded inside customer organizations to design, build, test, and deploy production systems connecting OpenAI models to the customer’s data, tools, and workflows.
Earlier this spring, ahead of either lab’s move, Sequoia partner Julien Bek published an essay called “Services: The New Software.” It went viral, well past three million views on X, four hundred and fifty thousand impressions on LinkedIn. Bek argued that for every dollar a company spends on software, it spends six on services. The next trillion-dollar company, he wrote, will be a software company masquerading as a services firm. Selling the tool means racing the next version of the model. Selling the work means every model release makes the service faster and cheaper. The autopilot, in his vocabulary, captures the labor budget rather than the software budget, and the labor budget is six times larger.
Three of the most influential actors in the AI economy. Three different starting points. One structural answer. Embedded engineering inside the customer, outcome-priced delivery, model access bundled with services, financial sponsorship from alternative asset managers who understood, before anyone else, that the services profit pool dwarfs the software profit pool.
This is the Transformation Triangle.
How the Match Is Exact
Look at the Anthropic announcement again, but this time through the Transformation Triangle’s three legs.
Software is Claude itself, plus the Applied AI tooling that customizes it. That leg already runs.
Services is the embedded engineering team inside the customer, sitting with clinicians and IT staff, building the workflows. That is the leg the new company creates.
Learning is what makes the whole structure compound. Every engagement generates signal: corrections, edge cases, workflow patterns, vocabulary specific to a physician practice or a community bank. That signal flows back into Anthropic, into the Applied AI engineers, into the next deployment, and eventually into the next training run. The loop closes.
The software gets smarter because the services generate signal that the learning leg captures and returns to the software. The services get faster because the software does more of the work and the learning sharpens the judgment behind it. The learning gets richer because both legs are running.
Now look at OpenAI’s structure. Models plus FDEs plus customer deployments plus signal flowing back. Same three legs. Same loop.
Now look at Bek’s essay. Software companies “masquerading” as services firms. The argument is that the system the labs are now building is the only one that works, and the entry point to building it is wherever a category is already outsourcing services that can be re-delivered with AI underneath.
The vocabulary is different in each case. Anthropic says “enterprise AI services.” OpenAI says “deployment.” Sequoia says “autopilot.” I have been calling the underlying structure the Transformation Triangle. The labels differ because the lineages differ, but the structure is the same.
The Model Alone Is No Longer The Product
May 21st, Greg Brockman, President & Co-founder of OpenAI tweeted, “the model alone is no longer the product.”
The companies that built the foundation models, the most successful tools ever shipped in the history of computing, are the same companies that just announced they cannot survive on tool revenue alone. Anthropic and OpenAI created the technology that commoditizes everything built on top of it.
They watched their own customers, the wrapper companies, get eaten by their own product roadmaps. They have now reached the same conclusion every downstream company is reaching: if you sell the tool, you race the next version of the model. If you sell the work, every model release works for you, because every model release makes the work cheaper and faster to deliver.
The model is not the product. The loop is the product. The labs that invented the commoditization technology are not exempt from its consequences. They are simply earlier to the conclusion.
Eight months ago, in a piece called “Every Law Firm Is Now a Software Company,” I described how this would play out across professional services. Three months ago, I showed how it was playing out across legal AI. One month ago, I named the loop as the only structure that survives. This month, the labs themselves confirmed it.
What This Means If You Are Building Anything
The convergence is the signal. If Anthropic and OpenAI cannot survive on tool revenue alone, no one downstream of them can either. The companies that read the signal early and start running the full loop will compound. The companies that wait for further confirmation will discover that they missed the boat.
Three questions require your attention:
Does your Software get smarter from your Services? If your engineering team ships the same product whether or not your delivery team is engaged with customers, the loop is broken on the Software side. You are not building a loop. You are building a feature factory.
Do your Services get faster and more accurate because of your Software? If your delivery team works the same way they worked before AI, you have adopted a tool, not transformed a business. The loop is broken on the Services side. Your delivery economics will not survive a competitor running the full loop.
Does your Learning go anywhere? If the corrections, the patterns, the edge cases your team encounters stay in their heads, in Slack threads, in the memory of a single senior partner, you have brain drain. The loop doesn’t compound. Every engagement starts the same as the last one. Your competitor, the one running a feedback loop, gets one engagement smarter every week.
The answer to any one of those questions for most businesses today, including most law firms, most consulting practices, most software companies, and most professional service firms of every kind, is “No.”
That is the gap. And the gap is what the new lab-services companies, backed by Blackstone and Bain and Goldman and Sequoia, are about to exploit.
Closing Thoughts
I called the Triangle in August. I tracked its validation in February. I submitted my book manuscript to my publisher wherein I refined it in April. This month, the foundation model labs themselves started building it in public. The thesis is not a thesis anymore.
The model is not the product.
The loop is the moat.
Tom Martin is CEO & Founder of LawDroid, Adjunct Professor at Suffolk University Law School, and Author of the forthcoming AI with Purpose: A Strategic Blueprint for Legal Transformation (Globe Law and Business). He is “The AI Law Professor” and writes his eponymous column for the Thomson Reuters Institute.