Awell

There were nearly not enough sick people in the privacy committees that wrote HIPAA & GDPR. We're about to make the same mistake with AI. HIPAA & GDPR proved one thing: when patients aren’t in the room, the rules guarantee red tape, not results. Eight out of every ten patients are happy to share their data if it increases the chances of getting them cured. Those privacy rules were drafted by people who’ve never stared at a monitor praying their labs come back in time. Result? Patient data is locked in vaults while care teams still hunt data in silos like it’s 1999. AI committees are forming now. Same faces, same “safety” sermons, same bottlenecks already surfacing. And I bet in every single one of them lawyers outnumber patients when it comes to making the final call. Unless we stack the chairs with people who have a vested interest in getting cured, we're headed for decades of paperwork first, care second. If you personally have health issues or know someone who does, would you also like to see more sick people in AI committees? Have you seen any examples of healthy people policy stifling bedside results? Would love to hear your thoughts in the comments.

Your EHR should be a system of record, not the system of engagement. Healthcare orgs needs to be moving towards a composable care stack. Every part of your experience from patient intake to care management should be pluggable, swappable, and API-driven. This approach doesn’t just speed up innovation, it unlocks new ways to orchestrate care. That’s how AI agents can shine. Agents need to be able to reach across your EHR, CRM, comms tools (and more), to surface data, trigger tasks and connect care teams to the right next action. Move past your EHR bottlenecks and see what decentralized, agent-powered care orchestration can look like. DM Jonathan Belanger for more.

We call it “Hero in the Loop.” Every Awell agent is a sidekick taking care of the routine, the noise, and the repetitive admin, always deferring to the real care team heroes when it matters. The moment an agent hits uncertainty or an edge case, care teams get looped in where their judgment is needed. This isn’t about replacing clinical expertise. It’s about giving care teams superpowers.

3 things for any healthcare org wrestling with AI right now: 1. Orchestrate. Automation’s for checklists. Orchestration is how you keep care from slipping through the cracks. 2. Move to reasoning. Rigid rule-based flows break the moment real life gets messy. Agent-powered care flows that adapt and make judgment calls is what you're looking to build. 3. Test for chaos If your AI can’t handle bad data, escalate when lost, or show its work - reevaluate. Read more: https://lnkd.in/e28B5d_U

🟢 CareOps is going mainstream. Read below 👇🏽 Composable > Monolithic. The best teams are moving beyond the EHR by building modular care stacks stitched together by orchestration, not just interfaces. https://lnkd.in/gFifzxdN Automation isn’t enough. Automating steps ≠ coordinating care. Teams are realizing that context, handoffs, and uncertainty are the real operational challenges. https://lnkd.in/gHhh8YD7 Agentic systems are on the rise. Not chatbots. Not copilots. Background agents that can reason, act, and escalate as care needs to happen. The big unlock isn’t about AI per se. It’s embedding it where care happens. https://lnkd.in/e28B5d_U

Getting to a working initial implementation of a complex multi-agent system is fairly easy. Making them perform consistently in production with appropriate human-in-the-loop guardrails in place? That's where 80% of our engineering effort goes. Here are some lessons you can steal for free: 📚 Evaluation - We built a custom platform just for prompt tuning and model benchmarking. This unglamorous work matters more than the agent orchestration everyone talks about. 🐘 Memory design is critical - We use semantic & episodic memory to feed interaction patterns back into prompts. Procedural memory? Still "maybe later." 🚩 Google A2A - It provides a structured language to talk about agent to agent collaboration, but delegating tasks was a let down. Read more: https://lnkd.in/eRwcsmjQ