Twitxy Saved

Sam De Brouwer

Verified account This account is verified. Learn more

@sam_debrouwer

Co-Founder and CEO at @xyailabs and curator at @tedaisf - Serial entrepreneur in tech, investor, board member, caregiver for life.
Palo Alto, CA linkedin.com/in/sam-de-brou… Joined September 2010
2.2K Following 2.9K Followers 9.8K Posts

Latest posts 5

Sam De Brouwer Verified account @sam_debrouwer
Aka @XYAILabs 👇

1. In healthcare, you have the providers, the payers, the pharma and other stakeholders, and none of their business models or incentives align.

2. AI fixes the messiness of the work. But AI doesn’t fix the messiness of an organization. That’s not an AI problem. It’s a leadership problem.
FARR Summit Verified account @Farrventures 4d
Why is healthcare messier than people think, and where can AI agents actually help? @sam_debrouwer , Co-founder & CEO of XY.AI Labs and Founder of TEDAI, took that question on at FARR with @bilawalsidhu , Co-founder & CEO of halfpixel.ai and Technology Curator of TED.

Sam:

1. In healthcare, you have the providers, the payers, the pharma and other stakeholders, and none of their business models or incentives align.

2. AI fixes the messiness of the work. But AI doesn’t fix the messiness of an organization. That’s not an AI problem. It’s a leadership problem.

FARRsummit.com
#FARR #FARRSummit
View quoted post on X
Sam De Brouwer Verified account @sam_debrouwer
There’s a term we don’t talk about enough: token maxing.
It’s a simple idea. The more an AI tool works for a team, the more tokens it uses. The more tokens it uses, the more someone pays for compute. Some AI-native teams track it closely because it can show how deeply AI is being adopted.
But for healthcare operations, token maxing is the wrong goal.
Our agents already handle a large share of repeatable eligibility and benefits work, but not because we try to maximize token usage. It’s because we design the workflow carefully, then let the agent run the repeatable parts with controls, validations, and escalation paths.
And I don’t bill my clients that way.
If you run a mid-market healthcare practice like a surgery center, per-token pricing doesn’t work. Your volume spikes when patients show up, when payers tighten rules, when schedules change, or when codes need review. A token meter can turn every spike into a higher invoice.
So I compile the workflow at build time. The team designs it once, with the right rules, data checks, prompts, integrations, and human review points. The agent runs from there.
No bad surprises, just predictable pricing within agreed operating volumes, whether you process hundreds of checks or many thousands.
It’s not job replacement. It’s task compilation.
Finance can budget without surprises. Ops can focus on throughput and exceptions, not on defending the AI bill.
Token maxing may be the right signal for some AI teams. Compiled AI workflows are the right model for ours.
xy.ai
Sam De Brouwer Verified account @sam_debrouwer
Let’s stop throwing tokens at the problems we’re solving with code generation becoming the universal interface for AI’s to accomplish any task. No reasoning forever, use it to write code once then let that code execute reliably, every time, aka @XYAILabs .
XY.AI Labs Verified account @XYAILabs Jan 7
Stop Building AI That Talks. Build AI That Works.
Most enterprise AI will fail.
Not because the models aren’t good enough
But because it's being built wrong. 🧵👇
xy.ai/tech-behind-xy…
View quoted post on X
1 replies 7 reposts 10 likes 1.3K views View on X