A Hand to Hold and a Throat to Choke

Why OpenAI is becoming a services company and what it means for the future of software and consulting

Any guesses for the largest enterprise AI company last year?
Accenture.

Yes, Accenture, the world’s most prestigious vendor of PowerPoint slides that explain why your ERP rollout is behind schedule.

In FY2024, they drove $900 million in generative AI revenue - 9x year-over-year - and ended with $3 billion in bookings, accelerating every quarter. Which is… a lot. Especially for a company that doesn’t build models. Or infrastructure. Or, really, anything.

But what it does do is show up at large companies and say, “Don’t worry, we’ll figure out what this AI thing means for you.” And then it charges many millions of dollars for figuring that out. Which, as it turns out, is a pretty good business model.

While model companies burn billions on compute to make models smarter, systems integrators like Accenture are cashing the checks - helping enterprises figure out how to actually use them.

Well, OpenAI is no longer content to sit back and let others monetize that gap between capability and application. It’s going full-stack.

They’ve hired a team of Forward Deployed Engineers - which is Palantir-speak for “very smart people who are technically consultants but don’t like being called that” - to help big companies fine-tune models on their own data, build custom apps, and generally do something with AI. Price tag? Min $10m spend. Competition? Everyone.

This move pits OpenAI directly against:

• Traditional SIs like Accenture and Deloitte

• The growing cottage industry of OpenAI-focused consultants and implementation partners

• And even Thinking Machines, the new venture from ex-CTO Mira Murati, aiming to help enterprises build AI-native systems on top of foundation models

OpenAI isn’t just selling models anymore. It’s selling transformation - and making a play to own the infrastructure, the interface, and the implementation.

What Becomes of Software and Services?

The AGI-pilled among us believe once AI can write your code, the end of traditional B2B software is near.

Why buy SaaS when you can spin up custom apps with natural language?
Why pay for a CRM when GPT-5 can generate one on the fly?

I get the vision. I’m optimistic enough to believe that AI will rewire industries - but jaded enough to know that implementation is where dreams go to die. Enterprises don’t just want models, they want guidance, governance, guarantees.

Enterprise adoption cycles don’t collapse just because the tech got smarter. They stretch. They stall. They involve compliance audits, quarterly roadmaps, and twenty-person meetings about data permissions. AI won’t fix that overnight.

The truth is: building may be easy now. Owning is not. Sure, you can build your own CRM now. But:

• Should you?

• Who maintains it?

• Who secures it?

• Who integrates it into the rest of your stack?

• Who audits it for compliance?

• Who explains to your sales ops team why the AI-generated CRM just emailed every customer named “Ben” a photo of a duck?

The marginal cost of creation may drop to zero. But the cost of maintenance, context, and ownership is alive and well. Just because a startup can build its own billing system with AI doesn’t mean it wants to. Or should.

This is why the idea that “AI kills software” is incomplete. It kills software as we knew it - but not the need to buy solutions.

The Arc of Abstraction

The entire arc of tech has been about lowering the cost of creation:

• From raw hardware to cloud infra

• From assembly code to Python

• From building servers to renting compute on demand

• From designing interfaces to prompting a model

But easier building has never killed specialization. It scaled it.

More abstraction didn’t mean everyone built their own tools. It meant more people could build more things and specialists - who built with taste, not just tools - got even more valuable.

AI will follow the same pattern. The easier it is to build, the more important it becomes to choose what not to build.

What AI Really Does: Collapse the Divide

What AI is really doing is collapsing the boundary between product and implementation.

If a model fine-tuned on your data can generate a tailored interface for your workflow, and a team shows up to do the tuning, and they charge you $10 million to do it - is that software or consulting? Yes.

And if a consultant builds you a system of agents that uses your knowledge base and automates your back office, but it’s wrapped in a slide deck and a retainer agreement - is that consulting or software? Also yes.

This is not the death of software or services. It’s their convergence.

In the AI-native enterprise:

• Software companies will look like consultants.

• Consultants will look like software vendors.

• And the thing they’re selling will be: a system that adapts itself to you.

Not off-the-shelf. Not fully bespoke. Just yours. Built for your org chart. Your data. Your process. Your dysfunction.

AI makes software personal. Not just in UX, but in structure, logic, and outcome. And the players who thrive won’t ask: “What are you building?” They’ll ask: “Who are you building it for?” AI-shaped software will conform to your company like a tailored suit.

OpenAI Gets It

So yes, OpenAI is still selling models. But its consulting business will be the bridge between intelligence and adoption - between what’s possible and what gets done. And if history is any guide, that’s where the real value gets captured.

In enterprise AI, the model is the means.The deployment is the moat.

---