Mathieu Rihet

Super proud that our suite of AI Voice Agent products is leading the way to full healthcare automation. Doctors can finally do their jobs!

@pBermudezCanete Don't lie to your audience! 😆

For the record, I don’t do GLP‑1 dropshipping - this was an experiment to see if the traffic playbook works...

$60k ACV in 60 days vs $10k MRR from one meme reel -> here’s the math. Being in enterprise healthcare sales, I’m used to 2‑month cycles to close a $60k ACV deal. Then I saw MedVi: dropshipping GLP‑1s, >$400M in a year, built on one thing - stupidly large top of funnel → their landing page. So I pulled their Ahrefs. If they’re doing ~36k organic visits/month and even 30% convert, that’s ~10.8k new patients a month. At ~$1k–1.6k per patient per year, you’re looking at ~$11–17M in annualized revenue from just that one month’s cohort. I wanted to see if the “just juice the funnel” meta actually works. So I did the AI slop playbook: – threw up a landing page – posted 5 meme‑reels on Instagram – this one reel did ~187k views, ~26k interactions, ~10k likes – 99% from non‑followers That single reel drove 50+ signups. At $100/month that’s ~$5k MRR. At $200/month that’s ~$10k MRR. From one stupid meme and a mid landing page. Everyone in B2B talks about “pipeline” but sleeps on how insane consumer‑style top‑funnel math is in 2026. You can grind 60 days for one $60k enterprise logo… or engineer content that spins up $5–10k MRR in a weekend. Same brain. Different funnel.

@CanesDavid @WisprFlow Yes, I don't understand what the hell is going on with them but they need to fix it asap

@mathieu1865 @WisprFlow And they barely respond on support

.@WisprFlow was great but now the iphone app has a google SSO bug Anyone else having trouble logging in?

Honestly I think this study is somewhat semi-skewed especially when it comes to physician applications. AI hallucination is still a huge problem on the 5.5 model. In fact, I found that it hallucinates 7x more than Anthropic's latest model. Our company uses AI to automate outbound workflows for medical groups, like referrals and cancellation recovery. I am more comfortable pitching these workflows to physicians because they are 'safe'. If your AI messes up one out of 50 times (which is basically our current rate), you can establish guardrails such that the call either gets transferred or ends immediately whilst not endangering patient life. The same cannot be said for anything remotely related to clinical decision making support, which means the plateau has been reached there.

AI won't replace physicians. Until it does. That's been the comfortable consensus for a decade. And every few months, the evidence against it gets a little harder to dismiss. Last week OpenAI released GPT-5.5. Within 24 hours, an independent benchmark run by ETH Zurich researchers had already quantified the jump. On Matharena's BrokenArxiv test — which presents AI models with plausible but false mathematical proof statements from recent arXiv papers and measures whether they can identify the flaw — GPT-5.5 scored 73.66%. GPT-5.4, released less than two months earlier, scored 36.61%. That's a near-doubling of performance on novel reasoning problems in a single model generation. Not on training data. Not on memorized examples. On fresh, deliberately uncontaminated problems designed to test whether a model can look at something that appears correct and determine that it isn't. Physicians should pay attention to that specific capability. Looking at a clinical presentation that appears straightforward and identifying the subtle wrongness — the lab value that doesn't fit, the history element that contradicts the imaging, the diagnosis that's plausible but not right — is the core cognitive act of expert clinical reasoning. It's what separates a competent diagnostician from a great one. And it just doubled. In two months! The rate of model release: GPT-5.2 shipped in December 2025. GPT-5.4 in March 2026. GPT-5.5 in April. Three major releases in five months. Each measurably better than the last on the hardest available benchmarks. The pace isn't slowing. It's compressing. For context: the third-place model on this same benchmark, Kimi K2.6, scored 15.62%. The gap between GPT-5.5 and the rest of the field isn't incremental. It's a discontinuity. And two months ago, GPT-5.4 was at roughly the same level as where these competitors are now. Health systems planning AI strategy on a 3-5 year horizon need to internalize what a doubling curve actually means for clinical capability. If mathematical reasoning — the most abstract, rigorous form of structured problem-solving we can test — is improving at this rate, the downstream effects on clinical reasoning, diagnostic accuracy, and treatment planning are not speculative. They're directional. And every clinician with a $20/month subscription is already sitting on the latest capability the moment it ships. The profession keeps waiting for the plateau. The plateau keeps not arriving.

@FierceHealth @pBermudezCanete

Stanford Health Care, Alameda Health System partner to support St. Rose Hospital bit.ly/4272wd8

@edorahg @evisdrenova @TrustVanta He did not contact customer success; he was just replying to the initial thread. All I'm saying is he could have multi-channeled, but definitely understand the frustration

@mathieu1865 @evisdrenova @TrustVanta Hmmm.. He did contact Customer Success. And this is not an edge case. They clearly are not taking into account the complete communication chain.

I will never use @TrustVanta again after how they just treated us. My company was acquired in August of 2025. We promptly emailed Vanta to inform them that we are shutting down our business and need to cancel. After 1 month, they finally replied with direction. "Can you send us acquisition documents?" I replied back and said "I can't legally share our acquisition documents with a vendor, what do you need?" They then ghosted us until January of 2026 (5 months later). During this time, they charged us for 2 quarterly payments (about $5K total). In January 2026, they finally said they need a certificate of dissolution, which I sent to them. They then refunded us $700. I asked for a refund dating back to August 26th when we initially sent the cancellation email. They responded with, "we cannot refund you because the Vanta software was still running until January of 2026. " It was running BECAUSE YOU GHOSTED US AND DIDN'T PROCESS OUR CANCELLATION. @christinacaci is this how you do business and treat your customers who trusted you for 2+ years?

@gabriel1 Yeah, that's so true.

hyper delusion of how good you are and hyper optimism is like 80% of making it

@Layton_Gott Build an MVP and ask for feedback.

Build a waitlist? Or Build a MVP? Which is better to test demand?

@KevinSzabo14 I don't think anybody is scared.

Why are people scared of AI?

@TTrimoreau Knowing how to prompt your AI agent

As a founder What is the single most profitable skill to have right now?

@t_blom About to already be the case for us 😭

By the end of 2026, I predict token spend will be greater than engineering salaries at early stage startups.

@garrytan I've been saying this: at this point, you should literally make it a company.

GStack still growing every day.

@paulg I found Superhuman's inbox to be pretty good at filtering those. Try it out!

I now get so many AI-generated replies that it's just too much work to report them for spam and block them, so I'm downshifting to muting.

@sohan_zhang No response is a response. I don't think this is disrespectful and have seen way worse horror stories. At Novoflow, one of our term sheets got pulled without warning. We thought we would never recover, but thankfully did.

one of the most frustrating experiences about the fundraising process is how decent respect gets thrown out the window one VC fund scheduled 3 meetings with myself and the team over 3 weeks, the last of which included an in person meetup that I took the courtesy of meeting them in the hotel they were staying at in SF. after a conversation, they mentioned that they would get back to me at the beginning of next week. cue 5 days later and i ask for an update. ghosted. i sent follow-up updates, as requested, over the span of a week, and still no response. i completely understand rejection, but it's very frustrating to take multiple meetings over weeks and get completely no response. future YCombinator founders, check the Bookface for investor reviews.

@HealthcareAIGuy @EvidenceOpen It's all fun and games until you ask it to integrate into a legacy electronic health record system!

NEW: OpenEvidence just launched “Coding Intelligence,” an AI medical coding feature. It auto-generates CPT, ICD-10, and E/M codes directly from visit notes, with built-in rationale and code sequencing to maximize reimbursement --applied automatically after each visit.

[1] for the last two years, the conversation has been about individual productivity. give everyone Copilot. ship faster. write more code. but the real shift is much bigger than that. it's not about making people faster. it's about redesigning the entire org around agentic systems — and the companies doing this are starting to look radically different. 🧵

@realleolu This is an amazing take, and you should make an article on this!! Post it on your blog.

@BryanOnel86 You are having way too much fun with this, Bryan 😭

You just can’t make this stuff up.

@SaiMandhan @HillValleyForum Cool picture!

live from the hill & valley forum so many great minds and a lot of optimism excited for the future, bullish on america 🇺🇸🫡