One AI vendor or several? The lock-in question

When a company outgrows Claude Team or ChatGPT Business, the next step on the vendor’s website is an Enterprise plan. It is the obvious move and the hardest one to walk back. Here is what a single-vendor commitment costs, what you get for it, and how to decide.

What you are signing

An Enterprise plan is an annual contract for one vendor’s models, one vendor’s admin console, and one vendor’s view of what your company should be allowed to do. Claude Enterprise publishes its seat fee and starts at 20 seats. ChatGPT Enterprise publishes nothing and starts, by reputation, around 150. Either way, for the next year everyone in the company uses that vendor.

That is fine if that vendor stays ahead. The frontier has changed hands several times in the last eighteen months, and what matters is which model does your legal team’s contract review best. That answer moves.

The three costs of lock-in

Price

A company with no alternative has no leverage, so the renewal conversation is a formality. Usage-based billing makes it worse, because the metered part of the bill grows with adoption, and adoption is what you are trying to increase. The vendor is paid by the token. You are not.

Quality

Standardising puts every team on the same models, including the teams where another model is clearly better for the work. The finance team that would do fine on a fast model pays frontier prices. The research team that wants the other lab’s latest release waits for the contract to end. Nobody chose badly. The choice was removed.

Switching

By the end of year one, the company has skills, connectors, permissions and habits built around the vendor’s console. Moving them is a project, which is why companies that grumble at renewal almost always renew. The contract is the smaller part of the lock-in. Everything built on top of it is the larger part.

The case for one vendor

Standardising is right for some companies.

You need a compliance feature only that vendor offers. A specific data residency region, a particular key-management arrangement, a certification your regulator named. If that is a hard requirement, the vendor with it wins.

You are large enough to negotiate real terms. At a thousand seats, custom pricing and a dedicated account team are worth more than flexibility.

You have decided, deliberately, to bet on one lab. Some companies have. It should be a decision taken with the costs above in view, rather than the default because the upgrade button was there.

The alternative is one layer over every vendor

Running two Enterprise contracts side by side doubles the problem. Two consoles, two bills, two audit trails, and no way to set one policy across both.

Independence means one layer that sits in front of every provider. Your people get Claude, ChatGPT, Gemini and open models through one assistant on every device. IT decides which teams get which models, caps spend per team and per user, and gets one audit trail across all of it. When a lab ships something better, switching a team is a setting, not a migration.

That is what Harriet is. Keep Claude, keep your existing provider accounts if you have them, and keep the option to change your mind. Compare it against the Enterprise quote in front of you before you sign.

Common questions

Should a company standardise on one AI vendor?

Only if it is a deliberate choice rather than a default. One vendor means one console, one bill and one renewal, which is simpler. It also means that vendor's models for everyone, whatever happens to quality or price, for the length of the contract. In a market where the best model changes every few months, most mid-sized companies are better off keeping the choice open.

What does AI vendor lock-in actually cost?

Price, quality and flexibility. Price, because a company with no alternative has no leverage at renewal. Quality, because every team is on the same vendor even where another model is clearly better for the task. And flexibility, because switching means migrating skills, connectors, permissions and habits, which is why almost nobody does it mid-contract.

Can you use Claude and ChatGPT together in a company?

Yes. The catch is that neither vendor's admin console governs the other's models, so running both natively means two contracts, two sets of controls and two audit trails. A platform that sits in front of both lets teams use either model under one policy, one budget and one log.

When is a single AI vendor the right choice?

When you need that vendor's specific compliance surface, such as a particular residency region or key management feature. Or when you are large enough that custom terms matter more than flexibility and you have decided, on purpose, to bet on one lab. For everyone else the default should be independence.