Build On the Platform, or Build Your Own? Healthcare Roadmaps in the Agent Era

A platform crossing a billion users changes the math for everyone building on it, and in 2026 that stopped being abstract. ChatGPT reached a billion weekly users and shipped Work, an agent that connects to your tools and produces finished output, not just chat. As one Latent Space breakdown put it, it is the agent for a billion users. When the default tool in every knowledge worker's hand can already run multi-step workflows, the question for a product team is no longer whether to build an agent. It is what to build yourself, and what to build on top of the platform.

Here is how I think about it, and it holds even in healthcare, where the instinct is to build everything custom. Build on the platform for the generic plumbing, the parts that are becoming a commodity. Build your own only for the regulated, differentiating core that the platform will not own for you. The twist in healthcare is that compliance and clinical safety move that line, so I decide capability by capability, never once for the whole roadmap.

The ground just moved

The scale of these platforms is the whole point: they amortize agent infrastructure across a billion users, and you cannot outspend that on plumbing. The numbers are worth sitting with. ChatGPT crossed a billion weekly active users in 2026, and its Work agent reportedly reached ten million users within three weeks of launch. OpenAI describes these as Codex-powered agents that automate complex workflows, run in the cloud, and connect to tools like Slack and email. That is an enormous amount of generic agent capability, memory, tool use, browser automation, being built once and handed to everyone. For a healthcare product team, trying to rebuild that substrate from scratch is a losing race. The interesting question is what sits on top.

The new default: build on top

For most capabilities, building on the platform is now the right default, because the platform does the undifferentiated work better than you will. A few years ago, building an agent meant building the whole stack. Now the platform hands you the loop, the tool integrations, the memory, the browser. Building that yourself, for a healthcare product, is spending your scarcest engineering on the least differentiated layer. The teams that win treat the platform as the substrate and put their effort into the thin layer that is actually theirs: the clinical logic, the workflow fit, the trust. Build on top by default, and reserve from-scratch builds for the places where on-top genuinely will not work.

Where healthcare breaks the default

Healthcare bends the build-on-top rule in one direction: the more regulated or clinically risky a capability, the more you have to own it, even when the platform could technically do it. In a normal software product, you build on the platform almost everywhere. In healthcare, two forces push back. One is compliance: PHI, business associate agreements, data residency, audit. If a capability touches protected health information, you cannot casually route it through whatever the platform does by default, you need contracts and controls, and sometimes that means owning it. The other is clinical safety and liability: if a capability can affect a patient, you own the accountability no matter whose model is underneath. So I map every capability on two axes, how much it differentiates me and how regulated or risky it is, and let that decide.

How I decide, capability by capability

The roadmap decision is never build or buy for the whole product. It is a per-capability call, with a healthcare gate on top. Generic agent plumbing: build on the platform, always. Domain workflow and clinical logic: build your own, that is your product. Anything touching PHI: build on the platform only if you have the contracts and controls in place, otherwise own it. And anything that can affect a patient: you own the accountability regardless of whose model runs underneath. The mistake I see is teams answering this once, at the whole-product level, and either rebuilding the world or routing PHI through something they never vetted. Answer it per capability and the roadmap gets both cheaper and safer.

The agent platforms are not a threat to healthcare product teams, they are leverage, if you use them for the right layer. Let the billion-user platform carry the plumbing. Put your team on the regulated, clinical, differentiating core that is actually yours, and treat every PHI and patient-safety boundary as a line you own even when the model underneath is someone else's. Build on top where you can, build your own where you must, and decide it capability by capability. That is the roadmap that survives the agent era.