Martin Camacho

Too many startups assume: 𝙩𝙝𝙚𝙧𝙚’𝙨 𝙖 𝘾𝙋𝙏 𝙘𝙤𝙙𝙚 → 𝙩𝙝𝙚𝙧𝙚𝙛𝙤𝙧𝙚 𝙩𝙝𝙚𝙧𝙚’𝙨 𝙖 𝙗𝙪𝙨𝙞𝙣𝙚𝙨𝙨. Reality is rarely that linear. Viability tends to unfold in fits and starts, shaped by payer coverage, provider workflows, and the real operational burden behind reliable billing. ⚙️📉 Inspired by Matt Kamen's excellent post on CPT adoption curves & the recent ACCESS discussion, we’re sharing our internal framework: 𝐐𝐮𝐞𝐬𝐭𝐢𝐨𝐧𝐬 𝐭𝐨 𝐀𝐬𝐤 𝐁𝐞𝐟𝐨𝐫𝐞 𝐁𝐞𝐭𝐭𝐢𝐧𝐠 𝐨𝐧 𝐚 𝐍𝐞𝐰 𝐁𝐢𝐥𝐥𝐢𝐧𝐠 𝐂𝐨𝐝𝐞 It goes deeper than payment policy — covering clinical and documentation requirements, enrollment and service mechanics, operational readiness, and the maturity of the CPT pathway. 🧠📋 If you’re evaluating a CPT-driven business model — or working with someone who is — comment “𝘾𝙋𝙏” below + DM me & I’ll share the document. 📄➡️ #startups #HealthcarePolicy #HealthTech #CMMI #ACCESSModel #Medicare #DigitalHealth

42

64 Comments

20% confidence prediction I've had for a while: labs will stop offering API access to frontier models altogether to force people to use their proprietary products. Builders that aren't competing will be fine (OSS, previous-gen models, etc), but it'd very tough to compete directly in the niches they care about.

49

6 Comments

The $4 billion valuation just became the new benchmark for AI utility. Synthesia just raised $200M in a Series E led by GV (Google Ventures), with NVIDIA and Kleiner Perkins staying in the mix. But the dollar amount isn't actually the most interesting part of this news. Two things stand out to me: The Secondary Sale: They included a secondary transaction for employees. In a market where "paper money" is being questioned and IPOs are stalled, Synthesia is giving its team actual liquidity. That is a massive statement of confidence and a huge win for the people building the tech. The "Bubble" Argument: Everyone is looking for signs of an AI slowdown. But when GV (Google Ventures) and NVIDIA double down on a company that has moved past the "cool demo" phase into actual enterprise integration, it tells you that the capital isn't disappearing, it's just becoming more selective. This isn’t just hype; it’s a massive bet on synthetic media becoming the standard for how businesses communicate. Congrats to the teams at Synthesia and Latham & Watkins for getting this across the finish line. It’s a big day for the ecosystem.

22

2 Comments

We've been scanning health tech companies across every major LLM as part of what we're building at Tovio. The dataset keeps growing and some of the findings are pretty wild. Fabricated FDA clearances, invented clinical studies, certifications that don't exist. Across every model. We decided to open up part of it. (link in comments) Today we're making the Health Tech AI Representation Index as a free reporting side-product. Each company gets a profile showing how accurately LLMs represent them, scores, specific errors with sources, trend over time. Some things we found: → Average accuracy on regulatory claims is under 70% → 26% of errors have a regulatory exposure We're starting with 200 companies and 5 LLMs (3 unlocked soon). It's a small set but expect it to grow fast. We're adding companies and models continuously. This is a fraction of what we actually track. The full proprietary dataset powers the Tovio platform. Continuous monitoring, risk classification, audit trails for compliance teams. But we figured this subset is useful on its own. If you work in health tech, go look up your company. If you're listed, you can claim your profile and get access to the full picture and start mitigating the risk with us today.

26

2 Comments

We made a huge bet that engineers will not want to be forced into using python just to use LLMs. The most underrated thing about BAML isn't that we improve tool calling on every model (including gpt-5), or that we have the fastest iteration loop for testing prompts. Its the fact that as your team scales and inevitebly uses different languages, they can ALL still use BAML to write their AI pipelines. Python devs love BAML. Typescript devs love BAML, and now we're starting to see Go devs on it too!

62

2 Comments

⚠️ Founders - consider the below reasons to not use Forward Deployed Engineers (FDEs) to acquire your *first* customers. This signals (1) lack of PMF; (2) miscalculation b/w development and sales timeframes; and (3) mischaracterization of Palantir Technologies timing when they deployed their FDE model. 💡 At the early-stage, it's crucial to demonstrate ability to repeatedly acquire customers with limited resources - capital, team, & time - that supports the narrative of 'more funding will amplify my traction.' Bespoke solutions for one company demonstrates zero repeatability. These custom solutions will not carry over to other customers within the same cohort. 💡 AI accelerates engineering but sales, system integration, and stakeholder buy-ins are perpetually slow. This imbalance results in concentrating all your engineering resources into one company, ignoring other opportunities, and still not going live on an accelerated timeline. 💡 Palantir didn't start using FDEs until year 3 when they got their first *gov't* customer, not their first customer. Gotham was already built and had strong PMF. Gotham allowed Palantir to take GTM risk by using FDEs b/c if it didn't work, Gotham was humming in the background. TLDR: At the early-stage, repeatable customer acquisition is the holy grail. Don't let AI deceive you that it can accelerate human behavior like it has with development cycles. It C.A.N.N.O.T.

33

1 Comment

Nudge & Structured Labs: Two of the hottest companies on Specter right now Nudge was founded by Fred Ehrsam and Quintin Frerichs in 2024, and is very HOT right now. It’s developing whole-brain interfaces for everyday use and has attracted interest from 30+ top-tier investors in the past month, including Benchmark and Sequoia Capital. Another company - Structured Labs, was founded in 2023 by Amrutha Gujjar & Shivam Singhal, and is a Y Combinator and General Catalyst backed startup that is creating smarter ways for people to explore and reason with data. It’s currently building Preswald, a workspace for analysts and engineers to explore data, test ideas and share learnings. The company has seen a HUGE uptick website visits over the last two months, having gone from a few hundred website visits a month to over 10k. Both of these companies are showing strong signals that they're getting traction and interest, and both featured on our Weekly Monitor newsletter, so check it out to stay in the loop!

29

6 Comments

There's a bigger hurdle to AI’s impact on biology than automation or compute. When Demis makes a bold claim or we hear about some 100x increase in a narrow bio problem, there’s usually a chorus of pushback that these aren’t really valuable additions. A lot of the pushback is that “biology is hard”, but this doesn’t explain why another 50 IQ points or 20h of task length can’t solve this kind of “hard”. I’ll try to be specific: What we call biology spans multiple physical 'layers of organization' with different behaviors and iteration speeds: Layer 1 = Molecular - e.g. drug molecules binding to proteins, which we need to make medicines. This is AlphaFold etc territory. For DNA, RNA, and proteins at least, we're good at simulating biophysics and experiments can happen as fast as arms can move things around. Layer 2 = Cellular - What's the response to a protein getting blocked? Some activity stops, the cell changes gene expression to compensate, and ends up in a new steady state (or maybe dies) minutes or hours later. New tech like CRISPR and single-cell sequencing apply here. Layer 3 = Organs - What happens to an organ when cells change? Often other cells get involved, and the new steady state might be destructive or ineffective. But this happens as gradual changes and responses over weeks to years. Many of the diseases we hope to cure live at this layer: Heart failure, dementia, aging… The key thing to understand, especially for non-biologists building AI for bio, is that layer 3 & beyond is emergent. The body is not the sum of all cells, thanks to feedback loops within and between cells. And with emergent complexity, scaling 1 and/or 2 alone does not mean 3 gets solved. AI thinking about biology is useful, for sure. And speeding up experiments in layers 1 and 2 helps disease research, for sure. Some diseases live at layer 2: infections, single-gene disorders, some cancers. And indeed the automation and scaling of layers 1 and 2 we’ve been doing for decades has given us decent answers to those already. But until we understand disease progression, any attempt to accelerate it (e.g. 'this protein is the problem, just put in a lot of it in the brain') risks creating a false goal. Neither more intelligence nor speeding up layers 1 and 2 can bypass the feedback loops of layer 3, which is where most diseases we care about actually live. This is the most important reason why we aren’t currently slated to double lifespan even if we get a ‘country worth of geniuses’.

50

7 Comments

#AmTech Weekly Roundup: Ambience Healthcare raised $243 million in Series C funding from Oak HC/FT, Andreessen Horowitz, OpenAI, Kleiner Perkins, Optum Ventures, and Town Hall Ventures. Ambience is building AI documentation software to record patient appointments, generate comprehensive medical notes, and standardize coding/authorization workflows.   Michael Ng, Nikhil Buduma    Reveal Technology raised $30 million in Series B funding from Ballistic Ventures, Shield Capital, Booz Allen Hamilton, and others. Reveal’s two main offerings are 3D mappings generated from drone videos and mobile biometrics. They are deployed with the US Army, Special Operations Command, Marine Corps, and other NATO forces.   Garrett Smith    Prophet Security raised $30 million in Series A funding from Accel and Bain Capital Ventures. Prophet is applying agentic AI to Security Operations Centers (SOCs), or hubs for cybersecurity defenses within an organization.  Kamal Shah, Vibhav Sreekanti #VC #HealthTech #Healthcare #Defense #Cybersecurity

43

3 Comments

The Stacklok team has been spending a lot of time looking at ways to improve the security of code generation workflows. Using a modern frontier model helps, but that isn't for everyone or every task. For folks using mid-sized models for simple generation workflows, check out our findings comparing popular models here: https://lnkd.in/gC4PknJY tl;dr; * All models struggle with bad package detection (Llama 3.2-3B was better with 29% packages flagged, but most only came in detecting around 5%). * CVE knowledge is low (with peak awareness of 18% for Qwen2.5-Coder-3B-Instruct, but 5% being more typical) * Insecure code recognition is a mixed bag. * Size isn't everything (some smaller models out-performed larger models) * Newer isn't always better. Check out the linked post for details, and check out Codegate if you are using smaller models for agentic workflows and want extra security.

39

Announcing another very boring, but very important (and surprisingly differentiated) feature. HIPAA's principle of least privilege is clear: every user, API, and service should only access what's absolutely necessary for their role. Yet nearly every HIPAA-compliant platform today relies on blocky, pre-set roles like "Admin" or "Editor" that over- or under-permit. That's why we just released fine-grained permissions across our entire privacy platform. From our healthcare CDP to HIPAA-compliant video, translations, and maps; every feature is wrapped in precision-level controls.

143

22 Comments

AI is powerful, but human touch is irreplaceable. A few months ago, I was working with a HealthTech founder who had invested heavily in an AI-powered triage tool. The data was solid. The predictions looked great on paper. But something was off. Clinicians weren’t using it. The dashboards were technically brilliant - but clinically irrelevant. After sitting with two frontline staff members, one insight stood out: “I don’t need a score, I need to know what to do next.” That one sentence reshaped the product roadmap. No algorithm could have uncovered that nuance - at least not without being trained on years of deeply human experience. Here’s the thing: AI can accelerate insights, but it can’t replace the insight that comes from sitting across the table, reading the room, and asking the second question. What AI did give us, though, was time. Once we streamlined the backend process, the team had more space to have those human conversations - to listen better, design smarter, and build something people actually wanted to use. If you’re building in HealthTech, don’t just feed your models - feed your conversations. Your product depends on both.

14

5 Comments

Quite a sobering piece by the Washington Post, although quite expected and matches what we are seeing at Diadia Health. TLDR: models still hallucinate a lot in the medical domain and any recommendations should be taken with a large grain of salt. * Confidence without competence: ChatGPT graded the author's cardiac health an F. His actual doctor said he's at such low risk his insurance wouldn't cover additional testing. Models tend to over-index on the presented data with overconfidence and lack proper reasoning. The AI treated fuzzy Apple Watch estimates like clinical truth. VO2 max from a wrist sensor can run 13% low. Heart rate variability has significant noise. Yet the model built confident conclusions based on this information without any "critical" thinking. * Wild inconsistency / lack of determinism: scores oscillated between F and B. The model forgot the user's age, gender, and recent bloodwork across conversations. We see this as well in our evals. Without applying proper reasoning, rerunning the model on the same set of inputs might yield different results. * The real doctor spotted what AI couldn't: neither ChatGPT nor Claude suggested a lipoprotein(a) test. That's the difference between pattern-matching and actual medical reasoning. This is exactly what we're building at Diadia Health to solve. Our AI transparency engine addresses the core problem: models that sound authoritative while being wrong. We're seeing critical reductions in hallucinated and unsupported claims by grounding outputs in two layers: scientific evidence and mechanistic reasoning, leading to accurate personalized results (answering the "why" behind the "what"). The potential for AI in healthcare is real and we are on the cusp of massive breakthroughs in personalized medicine. But potential without epistemic humility is very dangerous.

17

1 Comment