👋 We’re Cline, the first clinical harness where doctors can build, train, and deploy their own AI doctors, in production, for real patient care
Building in Houston, Texas
Joined January 2022
- Tweets161
- Following13
- Followers372
- Likes148
Software is going to be free. Here’s what that means for every doctor paying for an EHR:
- Software factories have collapsed the cost of building software
- Intelligence is becoming a utility, trending toward zero
- Durable software (your EHR, your data layer) will be priced at $0
- Disposable software is what actually moves patient care: real-time visualizations, patient-specific tools, AI-generated care aids built at the point of need
- What doctors should pay for: the intelligence that generates disposable software and the clinical harness that makes it safe to deploy
The vendors charging $800-$3,000/month are pricing against switching cost, not build cost. That math is running out.
Read the full article below.
We’re building it. AI doctors for doctors.
Malpractice should not only be determined by the care being delivered, but by the environment in which the care will be delivered from.
In the near future:
- Environment gets built in a software factory
- AI gets configured and tested for clinical use
- Malpractice is added prior to moving to production
withcline retweeted
Can’t build AI doctors for doctors without designing a new MGA for clinical AI malpractice. @withcline
withcline retweeted
Every doctor will offer a virtual DPC product and virtual direct primary care will be 100% dominated by AI doctors.
$150/mo or $300 for a family covered by the HSA/FSA.
Uncapped 24/7 access. Virtual visits, prescriptions, labs, etc with an option for in-person consultations.
Will AI push patient panel sizes up to 10,000? 🤔
Answer: Yes and no, but certainly not like how folks are describing it.
There are two different types of patient panels:
1. Panel per doctor or clinician
2. Panel per group or collective
AI will not increase #1 to 10,000 patients. That's just nonsense and doesn't follow any reality of clinical practice operations.
A 10,000 panel size will instead come from the group. option, but in this group it's just a single doctor and an AI doctor who is practicing in the capacity of an APP (under the doctors supervision).
In a traditional group practice (without AI doctors) a practice would need at minimum 3 full-time physicians with panel sizes of 2k, and between 3-5 APPs with panel sizes of 500-1k in order to get to 10,000 patients.
However, 1 doctor + 1 AI doctor will be able to take on that same panel size at a fraction of the cost while providing also providing 24/7, 365 coverage.
Now, do we think physicians should immediately rush to hit 10,000 from the start? Absolutely not.
But we do see groups reducing their clinical team sizes from 8 to half (if not more), and we expect each of those teams to practice with an AI doctor trained to the group and individual AI doctors trained to each group physician.
Times are changing: The doctors AI doctor to patient is the future of medicine.
withcline retweeted
System 1 models are the biggest threat to CDS tools like OpenEvidence, Doximity, and Abridge.
A clinical harness allows doctors to stack and route system 1 and 2 models for their AI doctors, generating a mix of structured outputs and reasoning that mirrors IRL clinical practice.
A new category of AI model just dropped. 👀
TypeSafe AI released Jev, a System 1 model that returns structured decisions in under 500ms with no hallucination possible. It does not generate text at all.
Here’s why this matters for doctors practicing with AI doctors ⤵️
- Physicians already operate in two speeds. Pattern recognition for routine cases (System 1), deliberate reasoning for complex ones (System 2). 10,000 hours of training is the process of expanding what a doctor can handle at System 1 speed.
- Almost every AI doctor product today routes everything through a frontier reasoning model. Same compute on a straightforward UTI as a treatment-resistant depression case. That is not how medicine works.
- The Cline clinical harness was designed around this distinction early on. System 1 layer runs continuously for structured clinical decisions. System 2 engages when the case earns it. Tasks graduate between the two under physician supervision, the same way a resident earns clinical autonomy.
We’re thrilled to see others arriving at the same conclusion, as it only validates our AI doctor framework and we can’t wait to see how fast this technology continues to scale for doctors and patients.
A new category of AI model just dropped. 👀
TypeSafe AI released Jev, a System 1 model that returns structured decisions in under 500ms with no hallucination possible. It does not generate text at all.
Here’s why this matters for doctors practicing with AI doctors ⤵️
- Physicians already operate in two speeds. Pattern recognition for routine cases (System 1), deliberate reasoning for complex ones (System 2). 10,000 hours of training is the process of expanding what a doctor can handle at System 1 speed.
- Almost every AI doctor product today routes everything through a frontier reasoning model. Same compute on a straightforward UTI as a treatment-resistant depression case. That is not how medicine works.
- The Cline clinical harness was designed around this distinction early on. System 1 layer runs continuously for structured clinical decisions. System 2 engages when the case earns it. Tasks graduate between the two under physician supervision, the same way a resident earns clinical autonomy.
We’re thrilled to see others arriving at the same conclusion, as it only validates our AI doctor framework and we can’t wait to see how fast this technology continues to scale for doctors and patients.
withcline retweeted
It all starts with @withcline's clinical harness
- Select and route your models (end to end HIPAA compliant)
- Add your context via the PracticeBrain
- Design, plan, and build your custom EHR in the harness software factory
- Train your AI doctor and AI practice manager to run it
They’re finally figuring it out. 😉
AGENT HARNESSES ARE THE NEW GPT WRAPPERS
What exactly is an agent harness? A harness does 4 things:
1. Runs the model in a loop so it keeps working step after step instead of answering once and stopping.
2. Gives it hands to read files, call tools, open portals, and run code.
3. Manages its memory so hour three of a job still knows what happened on hour one.
4. Enforces the rules about what it can touch and when it has to stop and ask a human.
WHY IT'S INTERESTING
- The market is MEGA. I think about it like a wrapper let you sell software ($800B market) but a harness lets you sell the work ($5T+ market)!
- It doesn't depend on any one company's model. The knowledge about the job lives in the harness, so you can run GPT today, Claude next month, an open model like Google Gemma, Qwen, Deepseek etc on your own machine after that, and it keeps working. A wrapper was one model doing everything and a harness is a router.
- It gets better the more you use it. See, every correction a human makes becomes a rule the harness keeps, so what you have after 500 jobs is a completely different product than what you had after 5. Wrappers only got better when OpenAI got better.
- You can charge for finished work! The harness knows when a job is done, so you can price per claim, per filing, per review, per resolved exception, or per closed month. This helps compete with SaaS!
- Most jobs are just the same handful of decisions, repeated, using the same few tools. That's true for most of the 800+ occupations out there, which is why almost all of them could have a harness built for them. Lots of space for startups to be building.
-The frontier labs won't come for these. OpenAI is not going to learn how a freight claim gets denied in Rotterdam or which prior auth your specific payer rejects. Those markets are too small for them and the knowledge only exists inside the job. They'll keep making the models better, which just makes your harness better which is cool.
- OH, AND really important, you can ACTUALLY build one now. OpenAI's Agents API, Claude's managed agents, Vercel's eve, LangChain harness framework etc. They all make it way easier to build agent harnesses.
Agent harnesses truly are the new GPT wrappers.
This is just the beginning.
Webside manner (the COVID-era pivot from bedside care) was always a bridge, not a destination, and what it was bridging to is arriving now.
TLDR on our article below:
- The staff-augmentation market (OpenLoop, Wheel, SteadyMD, CareValidate) was built on friendly PCs and group NPIs: clinicians rent in, don't own the relationship, and can be repriced or cut off anytime.
- That model already fueled a pill-mill problem, prison sentences, and ongoing DOJ/FTC enforcement. The bill is coming due.
- Unit economics settles the AI doctor question: a human physician runs $250k-$400k a year for 20-30 patients a day. That's the same math that just drove Stripe's $7B+ acquisition of OpenRoute for routing intelligence: don't spend your most expensive resource on repetitive work and tasks.
- Virtual care won't shrink, but the market question changes completely: not "who can source the most doctors" but "who owns the AI doctor, and who's accountable when it's wrong."
- The real shift for doctors: they stop being the rented asset and become the business that rents. Own the AI doctor, own the NPI (we've proposed NPI-3), own the liability, on your terms instead of someone else's.
- Liability isn't the problem. Liability without control is, whether that's a rented staffing arrangement or a consumer AI product that only calls a doctor back in when something breaks.
- Two paths open up: run your own AI doctor to extend a local practice, or keep your day job and rent yours out as a 1099 side business.
This is exactly the model Cline is built for.
TLDR: This isn't a technology problem. It's a labor crisis that's been building for 30 years.
- The 90s residency caps and IMG restrictions choked physician supply at the root. Managed care and credentialing rules then cut private practice off from hospital access, and hospitalists took that revenue for good.
- M&A, EHR mandates, and private equity finished it off. Physician ownership of practices went from ~76% to ~35%.
- APPs were pitched as the fix. NP panels run a fraction of a physician's, supervision still costs $10K-$30K a year in most states, and the NP pipeline pulls straight from an already short-staffed RN pool. We didn't solve the shortage, we just relocated it.
- None of that is a knock on APPs. It's that virtual care was never the right place to spend that labor. A human has fixed hours no matter how you deploy them.
- This is where AI doctors fit: virtual care, physician-supervised, AI operated, available around the clock, scaling with demand instead of headcount.
- Text your AI doctor first. If you need a human clinician, it routes you to them directly with full context already in hand.
That's the continuity we lost 20 years ago that we finally get back with AI.
