Implementation, Not Models, Is AI's Next Big Bet.
Anthropic and Blackstone just put $1.5 billion behind the idea that wiring AI into a real business matters more than which model powers it.
On this page
AI implementation is the work of taking a general-purpose model and wiring it into one specific business's actual tools, data, and daily habits until it does real, repeatable work there. It isn't building the model. It's everything that has to happen after the model already exists for it to change how one particular company actually operates.
That distinction just got a $1.5 billion price tag. Anthropic, in a joint venture with Blackstone, Hellman & Friedman, and Goldman Sachs, launched Ode with Anthropic, a company built entirely around embedding engineers inside client businesses to make AI actually work there. OpenAI has its own version, called The Deployment Company. Two labs that could spend their money on almost anything chose to spend a meaningful chunk of it on people who sit inside a business and rewire how it runs, not on people who train the next model.
The model is the easy part now
Ode's chief technologist, Eddie Siegel, put it plainly in the same interview: "I think model selection matters, but it's not where the majority of calories are spent. It's one ingredient in a system that has to be engineered." His comparison: choosing a model is like choosing a programming language. Relevant, but "I would not define an enterprise transformation in terms of whether they choose Python or Java." The transformation is everything built with the language, not the language itself.
That's the whole shift in one line. For the last two years, the loudest AI conversation was which model is smartest this month. The people actually getting paid to make AI work inside real companies have quietly stopped having that conversation. Ode runs on Anthropic's models by default and uses others when it makes sense. The work that takes 100 engineers and commands a $1.5 billion valuation isn't picking the model. It's the unglamorous part: figuring out exactly where a specific business's process breaks, and building something that fixes it without breaking everything around it.
Who this bet is actually for, and who it isn't
Ode's own description of its ideal customer is specific: a CEO who's personally bought in, where the work is "the most important product feature that the company is going to build over the course of the next two years, or it's reworking the most important business process they have." That's Fortune 500 language. It assumes a company with a CEO several layers removed from the actual work, a budget that supports a dedicated engineering team, and a process big enough that fixing it moves a real number.
If demand for this work outstrips supply at the very top of the market, where the going rate can justify a hundred-person team, it's not close to met further down. A five-person accounting firm, a regional logistics company, a local healthcare practice, none of them will ever be Ode's customer, and none of them are getting The Deployment Company's attention either. They still have the exact same problem: a general-purpose model they've poked at in a browser tab, and a real business process that hasn't actually changed at all. That gap is where a solo builder fits, not by competing with Ode, by doing the identical job at a scale Ode was never built to serve.
| Backing | $1.5B joint venture, PE-backed | You, your time, and whatever tools you already pay for |
|---|---|---|
| Team | 100 engineers, elite generalists | You, maybe one collaborator |
| Client profile | Fortune 500, CEO personally bought in | One small business whose specific process you understand |
| What's sold | Custom-engineered transformation | A working, wired-in fix for one costly, specific problem |
How to actually bet on this at your scale
Pick one business's mess, not a general AI product idea
Don't build a horizontal tool that could theoretically help anyone. Pick one type of business you understand well enough to name its actual broken process, not a guess at one. The specificity is the entire value; a tool built for everyone ends up wired into no one's real workflow.
Find their top priority, not a nice-to-have
Ode's ideal-customer test scales down directly: sell into whatever's already keeping the owner up at night, not into something that would be pleasant to have. If you have to convince someone AI matters in the abstract first, you've picked the wrong entry point.
Sell the outcome, not access to a model
Nobody needs to be shown ChatGPT. They need the specific, annoying thing that currently eats their week to stop eating their week. Price and pitch the finished, working result, not the fact that AI is involved.
Treat model choice as the boring decision it actually is
Pick whatever model does the job reliably and cheaply, and spend the rest of your effort understanding the business. Your edge was never going to be which model you use; everyone has access to the same handful of models. Your edge is understanding one business's mess better than anyone else bothered to.
Ode's CEO Chris Taylor said the founding belief behind the whole venture is that non-AI companies will be among the biggest winners of this entire moment, if they adopt the technology properly. He's talking about companies that can afford a hundred-person team to do that adopting for them. The same belief holds for the business down the street that can't. Someone still has to do the wiring. At your scale, that someone can be you.
Sources and citable claims
"Model selection matters, but it's not where the majority of calories are spent. It's one ingredient in a system that has to be engineered."
Source: Eddie Siegel, Ode with Anthropic's chief technologist, quoted in TechCrunch, 2026-07-15.
The enterprise AI implementation gap that Ode and The Deployment Company are chasing is real, but it's aimed entirely at companies large enough to afford a dedicated forward-deployed engineering team. The identical gap, a specific business's AI use going unwired into its actual workflow, exists untouched below that market.
Source: Romy analysis.
Questions this answers
What does 'AI implementation' actually mean?
It's the work of taking a general-purpose AI model and wiring it into one specific business's real tools, data, and daily habits until it does repeatable, useful work there. It's not building or training a model. It's everything that has to happen after the model exists for it to actually change how a business runs.
Why are Anthropic and OpenAI both betting on implementation instead of just better models?
Because they've concluded that most companies can't turn a capable model into real business value on their own. Anthropic backed Ode (a $1.5 billion joint venture with Blackstone and others), and OpenAI launched The Deployment Company, both to embed engineers directly inside client businesses and do that wiring work for them.
Can a solo founder actually compete with something like Ode?
Not head-on. Ode is chasing Fortune 500-scale clients with a 100-person team and $1.5 billion in backing. But the exact same underlying work, understanding one business's mess well enough to wire AI into it properly, scales down. A solo founder can do that for one small business at a time, at a price and intimacy Ode will never offer.
Does this mean model choice doesn't matter?
It matters less than most people think. Ode's own chief technologist compared model choice to picking a programming language: relevant, but not what defines whether an implementation succeeds. The differentiation is in how well the system is wired into the business, not which model powers it.
What should I actually build if I want to bet on implementation?
Not a general-purpose AI tool. Pick one specific type of business, understand the one process that's actually costing them money or time, and sell them a working, wired-in fix for that one thing, priced on the outcome rather than on access to a model.
Find the first people most likely to want what you built.
Give Romy a URL or one line about your product. Get a first read on who to focus on, where to test the market, and the angle worth trying first.
