.svg)
Most people still think of OpenAI as a product company: ChatGPT, APIs, models. But if you look closely, you’ll notice something different happening. They’re quietly becoming a consulting powerhouse.
And here’s the twist: consulting isn’t a side business — it’s the only way AI adoption actually works at scale.
Why AI Alone Doesn’t Sell in Enterprise
AI products are everywhere. Walk into a tech conference, and it feels like every booth has an “AI-powered” something. But scratch beneath the surface, and most enterprises are still stuck at square one.
Why? Because dropping an LLM into a messy system doesn’t create value — it creates friction. Legacy workflows, siloed data, compliance issues… AI doesn’t solve those problems out of the box.
That’s why OpenAI built an enterprise consulting arm. Instead of shipping “just a model,” they embed engineers directly inside companies to tailor AI to their specific workflows.
It’s less “download the API” and more “deploy the expert.”
AI Consulting Looks a Lot Like Palantir
If this sounds familiar, it should. Palantir built its empire by embedding forward-deployed engineers inside client organizations. Their job wasn’t selling software licenses — it was untangling operational chaos and turning it into systems that stuck.
OpenAI is following a similar playbook. Consulting isn’t just professional services on top of the product. It’s the wedge that ensures the product actually gets used.
Other players are doing the same: Thinking Machines Labs, Scale AI’s enterprise services, and even startups like Every (credit to Guillermo Flor for highlighting their story). The pattern is becoming clear: AI adoption is delivered, not just sold.
Startups Are Proving the Model
Take Every, a 15-person startup that started writing about AI and ended up running a consulting arm. In less than a year, they’ve crossed $1M in revenue helping companies operationalize AI.
The interesting part? Their engineers barely write code. Instead, they run structured discovery sessions, build lightweight automations, and leave behind AI playbooks that teams actually use.
That’s the magic: not complexity, but integration.
The Playbook: How Consulting Drives AI Adoption
Here are the five moves that stand out when you study OpenAI, Palantir, and startups like Every:
- Start with consulting, stay for the product
Services are the entry point. Once you’re embedded, automation and internal tooling naturally follow. - Force upfront commitment
OpenAI charges $10M to get started. Every charges smaller, but still upfront, fees. Both avoid the trap of endless “pilots” that never scale. - Embed into workflows
The real moat isn’t the model. It’s being part of how daily work gets done. Once you’re in, you’re hard to rip out. - Bypass platform dependencies
Product revenue might get shared with infrastructure partners (like Microsoft for OpenAI). Services don’t — they’re pure margin and direct customer control. - Deploy experts fast
It’s not about delivering a 200-page strategy deck. It’s about sending people who can ship inside the client’s chaos, fast enough to prove value.
Why This Matters
Here’s the uncomfortable truth: AI isn’t really a product business yet. It’s a consulting business wearing a product’s clothes.
The companies that win won’t just build models. They’ll master embedding them in real workflows.
For founders, the takeaway is clear: don’t underestimate services. Done right, consulting isn’t a distraction from product. It’s the distribution engine that makes your product indispensable.
Closing Thought
OpenAI’s move into consulting isn’t a pivot. It’s an admission: in enterprise, AI adoption doesn’t start with an API call. It starts with a human walking into the building, asking the right questions, and tailoring the system to the mess on the ground.
And once you’re in, you’re not just a vendor anymore — you’re infrastructure.
(Shoutout to Guillermo Flor, who has written extensively on the AI consulting wave. His breakdown of Every inspired part of this article, but this take — including the framing around consulting as distribution and AI’s “services in disguise” model — is my own.)
.jpg){kind=tracking}
