The Model Is No Longer the Product
What OpenAI's and Anthropic's twin deployment bets mean for every B2B SaaS shipping AI features this year.
In a single 24-hour window earlier this year, OpenAI launched a $10B subsidiary called The Deployment Company. Anthropic launched a similar deployment arm. Two of the largest AI labs in the world made the same bet in the same week.
If you read that as “AI labs are getting into consulting,” you missed the actual signal.
The bet they made — and they made it with their own balance sheets, not their marketing budgets — is this: the frontier model is no longer the product. Deployment is the product.
Most B2B SaaS companies haven’t internalized that yet. Their procurement, their roadmaps, their internal hires are still being made as if the bottleneck were model intelligence. It isn’t. Hasn’t been for a while.
What’s actually true in 2026
Three things are simultaneously true right now, and they explain almost everything about how AI is moving through B2B SaaS.
One. Model intelligence is essentially solved for the workflows most B2B SaaS companies want to ship. The frontier models can summarize a support ticket, extract a structured field from a document, draft an investor update, classify an inbound lead, generate a customer-facing reply. None of that is the limiting factor.
Two. Every B2B SaaS company has board-level pressure to ship AI features. Their investors are asking. Their competitors are shipping. Their customers expect it. This pressure is not abstract — it shows up in quarterly business reviews and in funding conversations.
Three. Most internal teams don’t know how to scope or ship AI features. Their product managers haven’t internalized agent workflows. Their engineers haven’t used Claude Code seriously. Someone tried a prototype. Someone suggested a chatbot. Nothing useful has reached production.
The gap between those three things is the entire market. Pressure to ship, no internal know-how, model capability sitting unused.
That’s where deployment matters.
Why this is a structural shift, not a vendor opportunity
When a software category matures, the locus of differentiation moves. In the early years of the cloud, the differentiator was “can your service run reliably at scale?” After a while, AWS, Azure, and GCP were all good enough — the differentiator moved to architecture and integration. Same shape, different category.
AI is doing this faster than cloud did. Two years ago, picking the right model was a meaningful decision. Today, it’s roughly a coin toss between the top three for most workflows. The model is the cloud. Reliable. Roughly fungible. Not the place where customer outcomes are decided.
The place where customer outcomes are now decided is closer to the user. Specifically:
Which workflow are you shipping into the product?
Who is the user of that workflow?
What’s the input shape and where does it live in the customer’s stack?
What’s the output, and how does the user review it?
What does “good” actually look like, and how will the team measure it over time?
Those are not model questions. They’re product questions, integration questions, organizational questions. They require someone to walk into the customer’s operation and answer them — in the codebase, against the real data, with the real users in the room.
That’s what OpenAI and Anthropic just bet billions on. Not “we’ll sell you a better model” but “we’ll send people who will ship the workflow.”
Who can actually buy from them
Here’s the part that’s underappreciated.
OpenAI’s deployment company and Anthropic’s deployment arm are aimed at the Fortune 500. Minimum engagement sizes north of a quarter-million dollars. Procurement cycles measured in quarters. Security reviews that take longer than the actual work.
If you run a B2B SaaS company with 20 employees — post-product-market-fit, paying customers, a real product, and a board asking about AI — you cannot buy from them. The procurement process alone would consume more time than the work itself. You’d be six months into security review before code was committed.
You also can’t really hire your way out. A senior AI engineer good enough to do this work credibly costs around $300K all-in and takes six to nine months to find. You could pull two of your existing engineers off the roadmap for six weeks to figure it out internally, but the roadmap slips and the AI work probably still doesn’t ship.
So you do what most companies are doing: a prototype that doesn’t quite work, a chatbot the marketing team got excited about, an internal Notion page titled “AI strategy 2026” that nobody has looked at since February.
This is the gap that needs to close.
Forward Deployed AI Engineering
The term is borrowed deliberately. Palantir used it first — the engineers who embed inside a customer’s organization, work inside their stack, and ship inside their operation, rather than dropping a report and disappearing. OpenAI and Anthropic both used the same word — deployment — when they launched their subsidiaries.
The shape of the work is the same at every size. Embed. Ship. Code in the customer’s repo, not a Loom on a Notion page.
What’s different at the smaller end of the market is the constraint set. The 20-person B2B SaaS company can’t run a quarter-long procurement. They can’t fund a quarter-million-dollar minimum engagement. They can’t wait six months for a security review of a workflow that’s supposed to be live in eight weeks.
So the unit has to shrink. Same delivery shape, smaller container.
That’s what we built: ten business days, one narrow AI workflow, shipped into the customer’s product. Fixed scope, fixed price, code committed. Not a strategy. Not a chatbot the marketing team thought of. One workflow with a clear input, a clear output, and a person on the customer’s team who will actually use it.
It’s the same playbook the big labs just spent ten billion dollars validating — sized for the customer who can’t buy from them.
What this means for you, depending on which side of the table you’re on
If you’re a B2B SaaS founder under pressure to ship AI:
Stop buying as if the model were the bottleneck. It isn’t. Buy deployment.
Stop thinking in “AI strategy” units. Think in workflows. One specific process, one specific user, one specific input-and-output.
Stop pulling your own engineers off the roadmap to figure this out. The opportunity cost is higher than you think, and the probability of shipping is lower than you’d like.
If you’re an engineer or product person watching this category from the outside:
The frontier model is becoming infrastructure. The work is moving toward the user.
Embedded, in-the-codebase delivery is going to be a real category for the next 18–30 months, until larger consultancies catch up to the pricing.
The skill that compounds is judgment — knowing which workflow to ship, scoped how, with what success criteria. The code is the cheaper part.
The line worth remembering
The frontier labs didn’t bet on better models this year. They bet on shipping the model into customer operations. That’s a tell.
The companies that ship useful AI in the next 18 months will not be the ones that bought the best model. They’ll be the ones that bought the deployment.
The model is no longer the product. The deployment is.
If this topic is useful to you, I go deeper in my book, Forward Deployed AI Engineering: A Working Guide to the Hottest Job in Software.
It’s available now on Amazon: https://www.amazon.com/Forward-Deployed-AI-Engineering-Software/dp/B0H3VKKG38/