Commoditize Everything in Healthcare

The basics are toast. Meeting notes? Scribe bots? Virtual receptionists? All headed to $0. They’ll be bundled into every SaaS suite, every EHR, every cloud console. Palantir, Google, Epic—they’ll ship “good enough” versions at scale, because scale is their religion.

But here’s the catch: good enough still costs them 200% margins to protect those $300M VP comp lines. And good enough costs health systems billions of dollars in inefficiencies. They can’t afford to be great. They can only afford to be good enough.

That’s the opportunity! 

Startups don’t win by competing on the commodity layer. We win by turning the 1% edge into a $100B moat.

The average Hospital uses 50-500 software tools, 

Zoom out:

• Average employee → $40K/year
• Elite world-class employee → $1B (Zuck’s FB offers)

Same delta in AI Agents.

• Commodity AI voice agent → that can handle patient calls, $0.
• Sully AI Receptionist → schedules + deferes >50% humans + detects+pro-active + wearable drift alerts → $1B+/year in outcomes

80% of clinical work is complex, time-sensitive, and multi-modal. That’s not commoditizable with one 7B model and a prompt. It needs:

• 20 API calls to 10 different systems, excluding the EHR.
• 10-20 different OpenSource LLM calls
• 3 wearables
• 1 EHR
• 1 genomic feed
• Real-time QA loop

Large players can’t move this fast. Their org charts won’t let them.

The Premium People Are Willing To Pay For

Ask a psychiatrist:

“Would you pay $500/month for an AI that texts your patient at 2am, detects flat affect, and books a crisis visit before you even wake up?”

Ask a daughter:

“Would you pay $99/month to know Dad’s Apple Watch + voice + med adherence predicts a fall risk 6 months out?”

The answer isn’t “maybe.” It’s “shut up and take my card.”

The basic features of agents will be commodized, and we will be the first to make them free, but we're building the state-of-the-art systems that can autonomously track the holistic picture of your health in commune with human doctors to provide the absolute best preventative and high-quality care.

These experiences don’t exist at scale today. When they do, the TAM isn’t “clinics.” It’s every single family.

Sully’s Play

We’re not building another agent. We’re building the autonomous OS (orchestrator).

• Plug in Oura, Apple, Epic, Cerner, Groq, Anthropic, your custom RAG
• Route every patient interaction to the right model at the right latency/cost
• Run evals on 100% of outputs (HIPAA-compliant, monitored by humans for safety)
• Predict 6-8-20 months out: “Mrs. Chen’s lithium levels + sleep fragmentation → 87% renal event risk by Q3 2026”

We’ll never do everything. But we’ll integrate enough to ship predictive care loops that no EHR dares to touch.

The Autonomous Operating System

Agents are coming. You’ll need a PhD agent to tune one for rare-disease diagnostics. You’ll need a GM agent to run 47 of them in parallel. And we will need a better operating system that can intelligently store, manage, and reason over all these concurrent operations. 

At Sully, we're building that AutonomousOS, and this isn't just a nice vision on the wall; we've already seen it from customers.

Feedback from a Sully.ai customer"

Commoditized AI will handle 20% of the edge cases. We will own (IA) the 80% where lives compound.

The race isn’t to build the model, agent, or solve for a use-case. It’s to build the system that autonomously operates systems.

Many of us watched Andrej Karpathy’s recent talk. 2-hours packed with how models are evolving and why “true AGI” is still a decade away. Karpathy’s AGI definition? A system you can ask to do a task, and it’ll crush it better than any human.

Haroon (Founder of RocketMoney) & I were driving down a few weeks ago to my place in Redwood City. Coincidentally, we were riffing on the same AGI framing—and how it might actually unlock faster through verticals we can build today. 

One idea that hit hard: His practical take on AGI is reactive brilliance. But what if the system doesn’t even wait for the ask? It’s already so embedded in the domain that it reads the room → infers intent from raw context → then executes the full loop end-to-end. That’s not just raw intelligence. That’s the spark of true autonomy.

Here’s the bet I placed that afternoon: Autonomy lands way before general smarts scattershot across everything. 

Feed models rich, vertical-specific context, and they don’t just play in the sandbox; they own the whole playground. 

Entire workflows, not fragmented tasks. 

We’re talking the next real leap: not marginally smarter tools, but outcome machines that run circles around humans in high-stakes verticals like healthcare. 

If AGI means outperforming us on command, autonomy means outpacing us on instinct, ingesting massive record sets in a blink, spotting signals as they flicker, reasoning over horizons we can’t even load into our heads. 

Sure, context windows are still the bottleneck today. 

But the research labs don’t sleep: Just look at those three researchers at The Chinese University of Hong Kong dropping “InfiniteICL: Breaking the Limit of Context Window Size via Long Short-term Memory Transformation. 

Infinite context isn’t sci-fi—it’s the quiet accelerator.

We’re shifting from “tell it what to do” to “let it see what needs doing.” That’s the daily grind at @Sully.ai; forging autonomous intelligence that doesn’t just assist, but anticipates, acts safely, and stacks wins for lives on the line. 

Patient outcomes don’t wait for shifts or sign-offs; they demand motion. 

Imagine a doctor grinding through the hospital workflows. Systems doesn’t just log data, it lives them. Spots the follow-up gaps before they widen, auto-books the slots, pings the patient with crystal-clear context, and seals the loop without a finger lifted. 

This is already happening today! 

feedback from a Sully.ai customer"

No more ghosts in the machine; just seamless care that flows.

That’s autonomous intelligence in its infancy: Deep domain roots turning faint signals—triggers, timelines, intents—into precision strikes that deliver every time.

30 years from now?

In 30 years from now, healthcare won’t recognize itself. The doctor’s office? A relic, like rotary phones in a touchscreen world.

Every human on the planet wakes to a personal AI guardian—autonomous, omniscient in their biology, woven into wearables, homes, and global data streams.

No more 26-day waits or midnight lab results lost in the void.

These autonomous agents doesn’t react to symptoms; it preempts them—modeling your genome against real-time vitals, environmental shifts, even your unspoken stresses pulled from voice patterns or sleep cycles.

A whisper of heartburn? It’s already coordinating with a nano-swarm of targeted therapies, routing virtual consults to human experts only when the edge case demands that rare spark of empathy machines can’t fake yet.

But here’s the fighter’s truth I learned tapping out opponents in Cairo boxing rings: The real revolution isn’t just the tech, it’s the mindset.

We’ll have AI collectives, swarms of specialized agents collaborating like a superhuman OR team, each owning a slice of the patient journey from preconception genomics to end-of-life dignity.

Payroll for humans? Slashed by 20x, freeing billions to chase meaning beyond charts.

Global health equity? Locked in, because autonomy scales without borders—rural clinics in Rajasthan get the same infinite-context brain as Mass General.

Because when autonomy hits escape velocity, it doesn’t just save lives. It redefines what thriving means for every soul breathing.

Let’s just build it.