I just let in 40 more people off the waitlist :) New since yesterday: - Doubles support - See your pro-"twin" - Analytics cross matches - Chat with coach across matches - Easy mobile use and sharing. Whatever feedback you have lmk and it will get done

Super excited to welcome Brett Caughran (@FundamentEdge) to the team @Fundamental @tryshortcutai! In the last 18 months, AI went from unable to do real Excel work to performing at the level of a top tier analyst In the next 18 months, I think we'll have agents that think and act like an investor. They'll know how to frame a thesis, what to check, when to be skeptical, and how a given firm wants the work done Brett is one of a kind talent. He not only spent his career as an portfolio manager at top funds, but also been systematizing the investment process at scale At Fundamental Edge, he taught many students this process, and now he’ll help us teach the AI so they can further empower investors 🚀🚀

Theres a big difference between not understanding all the inputs vs all the outputs. In general its ok to let go on the inputs, but It's slop when you can't explain every pixel / word / feature in the output. I think this is why people hate ai art, games, and writing so much

given all the interest (thank you all), @nicochristie and i will build a tennis coach mvp next week, and thinking about golf next. pls leave a comment below if youd like early access we just processed and provided diagnostics and tips for the the first 7 players that have sent videos through. see here: was fun and learned a couple things. player A (personal friend) wrote me: "I like it. It matches what my coaches have Tried to tell me about my forehand, but maybe this distills it to simpler fixes and language that I’ll actually change" some more technical thoughts below - camera angle really needs to be in the middle of the court for "shoulder turns" and rotational metrics to be accurate - for juniors, some of the advice for shortening swings and such is probably ill-conceived, given that they have less power and are also still developing - player box and joint tracking, as expected, already work great across broadcast and phone footage, but stroke labels do not, so i had to get opus to manually double-check some of these. just set up a self-improvement loop to train a small dedicated classification model for this

lol. I thought tennis coaching was meant to be safe from AI. Seriously impressive article from Peter and Nico, as to the capabilities of AI to turn raw footage into technical insights, or at least some more objective measurement of body angles. I still have reservations as to the actual hard part of coaching though, which is how you communicate these insights to the player in the best possible way (and that is the world-class aspect that I think will be tougher for a model to convey, although feed enough data and it will probably get there better than humans). E.g., Richard's backhand: "Bend the elbows a little and keep the racquet head up. Hit the ball just in front of the front hip, and swing through to finish over your right shoulder, not at the sky." Is that world-class coaching? In my humble opinion, it is not. And I wouldn't be surprised if Richard's backhand failed to improve trying to act on that specific set of instructions. While the information *might* be correct, the instruction is too sequential/serial and doesn't respect the more emergent nature of how people self-organise or the more metaphorical way we can enact behaviour change. This might sound dumber, but I think a more elegant instruction (based on Claude's set of feedback) might get distilled to "hit the ball harder" and Richard might discover (unconsciously, perhaps) that a higher racquet head and more elbow bend at the start creates more speed. I also don't think it respects different hitting arm structures (e.g., Tiafoe's double-bend vs many pros who straighten the arms more) but of course this data set only included 11 pros and used Tiafoe as a comp, so not really a weakness of the model in that respect. I also think footage angles and context of incoming ball make video analysis a double-edged sword. E.g., comparing Richard's forehand to Zverev's, where Richard is centre of court/screen on the front foot vs Zverev, left of centre (screen and court) hitting open-stanced. Claude says contact is "further out to the side than any pro" and advocates to "let the ball come closer" -- closer to the body? closer as in later contact? I'd say that is the opposite of what you really want in a forehand, and of what Zverev does on his forehand. But again, that could all be fixed and adjusted for in terms of making sure the model learns to compare more similar game contexts and video angles. Finally, while I write a very technical kind of game report for tennis pros, my thesis in general is that technique is more important the higher you get (i.e., it matters more from getting to 20 to 1, but matters far far less in getting from 400 to 40) and even less still at the club level. I'm waffling here and I have more thoughts than I can put down quickly here, but perhaps an article is warranted on this, because this is what's coming. But overall very bullish on this kind of thing getting refined and used more and making coaching better, so well done guys; I'll be out of a job in no time.

It was shocking for me to see the current state of radiology scribe software. We watched extremely intelligent and trained individuals being forced to use 30 year old SEGA controllers to make changes. It honestly reminded me of macaque motor control experiments with joysticks back in neuroscience wet lab at Columbia. It was such a treat to work with real radiologists to build a realistic, diverse, and rigorous medical benchmark. Certainly an unfair advantage when radiology runs through @nicochristie’s family tree. We had originally assumed that hillclimbing this would be trivial, but there is a lot of fun and nonstandard optimizations that one can make when speed AND accuracy are both prized. In fact, in the month that we had worked on this, we got bitter lessoned twice (generalist multimodal LLMs are better at technical transcriptions than specialized STT models, and then Opus 5.5 not giving a shit about how bad the transcription is provided it is somewhat phonetically intact) Starting to learn about RIS and HL7 etc. Reach out to us if you’d like to chat, partner, or integrate this into your practices

Everyone with a camera and claude should have a world class tennis coach. @brainsandtennis and I chained open-source models post-trained on tennis to track players, ball, court, and labeled every stroke with its type and a 65-joint skeleton. We then post-processed this data into 30+ biomechanical traits across serves, forehands and backhands, normalized for each player’s build and the camera. Turns out Claude is a wonderful coach :)

I turned the fly bisexual Inspired by Kallman et al ('15), I blocked mAL output in a 166,606-neuron fly model and measured responses in 8 candidate P1 male courtship neurons. In 3 experiments, when cued by other males, baseline P1 spikes rose from 0 to 1–5 after mAL blockade

Some personal news. After nearly three years, I stepped away from FRL & Shortcut earlier this month. In 2023, I dropped out of my mba at MIT to cofound FRL with Robert, Shuying, and Andrew. Cofounding a company is all consuming, and they have been the best cofounders I could have ever asked for. Robert/Shuying never mentioned it again, but I had a pregnant wife at the time and little savings, and Shuying helped cover my bills for months while we were still fundraising. I’ll forever be grateful. I was a former professional basketball dunker and self-taught dev, suddenly building with MIT professors, phds, and programming medalists. We immediately built a brand for far-out and inspiring applied research, building the first AI agents that people could play with in Minecraft. We even simulated a civilization of 1,000 agents with Project Sid (I still think this was the most fun / forward looking multi agent work until OpenClaw). On Oct 22, 2024, Anthropic released the CUA preview and Robert made the call for us to tackle OSWorld. He had the vision to lean into codegen and to build what we called “skills”(!!) to separate the hands from the mind at the time. Shuying set up the infra to run massive parallel evals so we could get dozens of scores a day. We beat SOTA in 1 week and more than doubled it in 5 weeks. It’s a huge regret that we did not open source and publish our research. We raised a $30M+ Series A less than one month later and went all-in on building CUA inspired products. Unfortunately, our first two products did not stick at all, and we found ourselves in the hardest chapter of our company history. At this point, each of the cofounders had performed what felt like mini-miracles except for me (Andrew beat the voyager benchmark with Altera agents!). I felt the enormous weight of the moment, I insisted that i build my own product for us. With a very small team, I built and launched Shortcut… and it made quite an impact. I put all of my love and care into Shortcut over the last 18 months. It took some time for our research lab to reorient after launch, but we found our stride soon after. We created an entire ai product (and startup) category and i don’t think it’s an exaggeration to say we changed the way people use Excel. Shortcut is growing rapidly (4x in 2026!) and we are live in many of the biggest investment funds in the world. The product has clearly entered its next phase as a scaled product with a killer team at the wheel. And with this new chapter, the time has come for me to take a step back after pouring my heart into this company and product. The world looks really different than it did in 2023, and I'm excited to go back to the drawing board. I’m deeply grateful to my cofounders, teammates (too many of you to name, but thank you for your trust and for carrying the torch forward), investors, customers, and our users. You are in awesome hands. Most importantly, I’d like to thank my wife, Melanie. Thank you for building our family, for your strength, and for your confidence in me that far surpasses my own. You are the reason for anything we accomplish, and the reason any of it at all matters in the end. Nico

Opus 5 is a shocking step up from Fable 5 and 5.6 Sol for spreadsheets Almost couldn’t believe our evals at first A step function increase in intelligence, while as efficient as Sol Feels like another Opus 4.5 moment for knowledge work

We have been testing GPT 5.5 on the hardest spreadsheet tasks in the world (100k-1M+ cell complex models). It is the Pareto frontier for spreadsheets -- SOTA accuracy, the fastest and the most efficient public model across effort levels. OAI really cooked here