The AI platform you choose shapes the team you need to build

Five weeks ago I started asking a different question about AI. Not which tools to deploy, but what kind of organisation do you need to become to make AI really work?

Week by week the answer has gotten more specific. This week I want to go one level further out, because the decisions that will shape your AI future are not just internal, they start with a choice most organisations are already making, but without realising it.

Innovation used to live inside the firm. R&D teams behind closed doors. NDAs. Proprietary everything. Build it yourself, protect it, own it. But that model is largely gone and innovation now happens across boundaries. Through partnerships, acquisitions, open-source communities, startup ecosystems. The firms winning with AI are not necessarily the ones building the most. They are the ones orchestrating the best.

Which brings us to the decision you are probably already making. There are three types of AI platform, and they work very differently.

An integrator sells access to external developers' work under one roof. Microsoft Copilot. Salesforce Einstein. Most enterprise AI deployments land here. The platform provider decides what gets in, how it is priced and what data it sees.

A transaction platform connects parties to exchange something. Uber connects drivers and riders. AI tools that match your query to a supplier, candidate or service work the same way. The owner controls the rules. You play within them.

An innovation platform is a foundation others build on top of. Android is the classic example. Hugging Face is the AI equivalent: shared infrastructure where developers build, share and modify models.

Most organisations are integrators without knowing it. They have signed up for Copilot or a similar tool, built processes around it, and moved on. Reasonable. But it is a choice with consequences that rarely get discussed in the procurement meeting or when designing the operating model.

The platform you build on today shapes your switching costs tomorrow. If your skills, data pipelines and processes are optimised for one closed platform, moving away in three years will be expensive and slow.

More immediately: the platform type dictates the internal capabilities you need. An integrator platform requires governance, prompt skills and vendor management. An innovation platform requires engineering and data expertise. These are very different investments. There is no one-size-fits-all answer.

This is not an argument against any particular choice. It is an argument for making the choice consciously.

Do you understand what you are committing to, and what it will cost you to change?