The Hard Part Isn't AI — It's Trust

Still early. Still learning. But one thing is clear while building Tulu Health: the hard part isn't AI. It's earning trust inside workflows that have evolved over years, sometimes decades.

When you first walk into a clinic — not as a patient, but as a technology builder — you notice something quickly: the gap between how operations look on a diagram and how they actually run in the real world.

On paper, there's a structure. There are SOPs, reporting lines, handoff protocols. But in reality? There are missed calls that nobody has time to return. Follow-ups that slip because the coordinator is managing fifteen other things at once. Systems that technically "integrate" but share data in ways that require a human to make sense of them.

This is the world we build in. And it took us a while to understand that the challenge was never the AI.

What "Messy" Really Means

Clinic operations are messy — not because the people running them are incompetent. The opposite is true. They are messy because they evolved organically, under pressure, over years. Every workaround exists because someone once ran out of time and found a faster path. Every informal process was once someone's clever solution to a rigid system.

When you introduce AI into this environment, you're not walking into a blank slate. You're walking into a living system with memory. Staff have seen technology "transformations" before. They've watched software projects fail, replaced by more manual processes than they started with. They've been trained on tools that disappeared after the first budget cycle.

So when we show up and say "AI can handle your patient intake, follow-ups, and no-show prevention" — the question isn't whether the technology works. The question is: why should they believe us?

What We've Learned About Trust in Healthcare AI

• Trust is earned in sprints, not proposals. No clinic owner trusts a demo. They trust a result. Showing one tangible outcome — a recovered no-show slot, a member who came back — builds more confidence than any slide deck.
• The staff who resist are often the ones who care most. When a care coordinator pushes back on AI-driven scheduling, it's usually because she's seen patients fall through cracks when systems didn't account for context. That pushback is signal, not noise. It makes the product better.
• Ownership changes everything. When a practice manager feels like the AI is theirs — tuned to their patients, their terminology, their protocols — adoption rates go up dramatically. Tulu Health deploys forward-embedded engineers for exactly this reason.

Start Small. Fix Ten Painful Things.

The insight that's shaped how we build came from repeated conversations with clinic owners and operators asking the same question: when does AI actually get adopted in a practice?

The answer wasn't "when the technology is good enough." The answer was: when it quietly fixes ten small, painful workflows that staff have given up trying to solve.

Not a revolution. A quiet accumulation of small wins. Nobody picks up a call? AI does it and logs it. Patient didn't respond to a follow-up reminder? AI sends it at the right time in the right channel. A member's renewal slipped through the cracks? AI flags it 60 days out, not after they've already left.

Each of these fixes is almost invisible. But together, they change the culture. Staff stop working around the system and start working with it. And that's when real adoption happens.

AI in clinics will not start as a revolution. It will start by quietly fixing ten small, painful workflows.

Healthcare Isn't a One-Size-Fits-All Problem

One thing we've had to unlearn: the assumption that what works at one practice scales directly to the next. It doesn't. A boutique longevity clinic has completely different patient communication norms than a multi-location medspa group. A practice with an in-house lab has different workflow needs than one that sends everything to Quest or LabCorp.

Building for this forces you to rethink a lot of assumptions around scale, ownership, and reliability. You can't just deploy once. You have to re-earn trust in every context.

That's harder to build. It's slower to scale. But it's the only way that works.

Where We're Headed

We're still early. The honest version of Tulu Health right now is a team that has figured out how to earn trust in complex clinic environments, build AI that integrates without disrupting, and show results quickly enough to survive the inevitable skepticism.

The technology — the AI agents, the LLM orchestration, the Patient Twin that keeps every profile current — that part keeps getting better and will continue to. But the capability that's actually hard to replicate is the understanding of how clinics work: the handoffs, the moments when a coordinator makes a judgment call that no algorithm has ever been trained on.

That's the moat. Not the models. The trust.