I first met @rronak_ and @MichaelElabd when we were all freshman at Stanford. Today, 8 years later, we’re announcing that we’ve started Trajectory, a research lab and product company building the platform for continual learning. I believe that Continual Learning demands a fundamentally new interface for how we build products. That's a research challenge and a product challenge in equal measure, so we've assembled a team to meet both: researchers from DeepMind, OpenAI, Apple, Meta Superintelligence, Amazon AGI, and Scale AI, and product talent from Stripe and Figma. We’re also partnering with the best AI native companies @Clay, @Harvey, @Decagon, @Mercor, and @RogoAI to power their agentic experiences, and push the boundaries of what agents look like in the real world. Please reach out if you’re excited to build with us!

We partnered with @trajectorylabs to post-train NVIDIA Nemotron 3 Ultra for legal. Here’s what we found: 1) Open-weight models can reach frontier legal performance. On our Legal Agent Benchmark (LAB), Nemotron 3 Ultra started at a 0% all-pass rate. After post-training, it reached 5.8%, placing it between Sonnet 4.6 at 4.2% and Opus 4.6 at 6.6%. 2) Post-training dramatically improves reliability. Before training, many held-out tasks missed enough rubric dimensions to land around ~70% pass rates. After training, those tasks shifted toward ~95% pass rates. 3) Open-weight performance comes at much lower cost. Post-trained Nemotron 3 Ultra reached a similar quality band to leading closed models while running at roughly 1/8th to 1/50th the per-token price of Sonnet 4.6 and Opus 4.6. Most importantly: we post-trained this model on the @trajectorylabs platform less than 24 hours after Nemotron 3 Ultra launched, using the same harness, data, and recipe we used for Nemotron 3 Super. More to come as we continue to experiment with open-weight legal agents. Read more on post-training with Trajectory below:

The past few days made one thing clear: you have to own how your intelligence grows. We're thrilled to partner with Harvey to build that platform that turns models and harnesses from a fixed constraint into something you control

That feeling when you realize practitioners say “verifiable tasks” when they actually mean “easy tasks.”

🏹 5 Days of Trajectory. Day 4 - Why We’re Building Trajectory AI is the most capable software ever built. You correct it. You teach it what you want. However, the next session starts, and the learning is gone. This is deeply unnatural - nothing intelligent works this way. Today, we’re sharing the thesis behind Trajectory: - why continual learning is the next platform shift in AI - why the primitive governing that shift is the trajectory - our plan to move products from being shipped to being grown: first make the intelligence layer better, faster, and cheaper; then make it shapeable; finally, make it learn Read more below⬇️

Kinda surreal we got on Times Square even before we launched

We’re taking a quick break for the 5 days of Trajectory, but wanted to take this time to say that we’ve been named to @Redpoint’s 2026 Infrared 100 as one of the companies shaping the future of AI infrastructure. We're so grateful for the recognition so early in our journey, and want to congratulate the other awardees as well!

Building the platform for continual learning means doing both cutting edge research and working with customers to make the right interface - so grateful that we've assembled a team that can do both

Welcome to Day 2. Yesterday, we showed the broader work we're doing with the pioneers of continual learning. Today we'd like to deep dive on one: how we post-trained an open model for legal work, in partnership with @Harvey. We've built a platform where production data is the moat. Every correction, retry, and edit becomes signal you can post-train on, and the models are plug and play: customer's can drop in their model of choice, and improve from there. Fields like legal and finance make those demands absolute, with hard security, sovereignty, and provenance requirements. That's why we post-trained @nvidia 's open-weight Nemotron 3 Super, on Harvey's LAB benchmark. The results, in just hours: post-trained Nemotron 3 Super approaches the closed frontier, matches GPT 5.5, lifts rubric-pass criteria 25%, all while beating the performance-vs-cost frontier. That's the power of our platform. And this is just a glimpse towards what the future of intelligence will look like: continual learning, where products get smarter every time they're used. Thanks to @nikogrupen, @gabepereyra, @ItsJulioPereyra, and the whole Harvey team for their collaboration on this. Much more to come soon on continually learning legal agents

5 days. 5 announcements. 5 days of Trajectory starts today. Yesterday we launched Trajectory (@trajectorylabs). We are building the platform for Continual Learning. Our platform unlocks the signal already sitting in product usage, so companies can continuously post-train large agentic models that outperform the frontier. Today we are introducing the Pioneers of Continual Learning: some of the first companies building products that keep improving long after they ship. This is how products will be built in the future, and we are building it together. Here, we’d love to highlight a few of these companies: — @harvey is pushing the frontier of legal AI, in a domain that has little tolerance for mistakes. They’ve turned this standard into LAB, an open, expert-graded benchmark. Now, with Trajectory, they are building on that signal toward models that continually improve. @ClayRunHQ has built go-to-market to be AI-native from first principles, and now with Trajectory, they're A/B testing models live that are already cheaper, faster, and most importantly, continually learning. @DecagonAI runs most of their agents on models they trained themselves. Together with Trajectory, they are now exploring how to train models with special capabilties (steerability, interpretability) that owning your own intelligence unlocks, all with the goal of continually improving in production. @mercor_ai is the expert layer beneath frontier AI, turning the judgment of professionals across law, banking, and consulting into benchmarks that grade agents against real work. They are now working with Trajectory on how continual learning can unify model training and the data generation process. — We believe AI should compound, not stagnate. That's the future we're building with the Pioneers of Continual Learning. Read the full story in the blog post below.

Proud of you guys @QuantumArjun

🥺

Very bittersweet, but I'm leaving Apple. Anyone who knows me knows how much I admire Apple's story and ethos. The iPhone captured my imagination as a kid, and never let go. And getting to spend the early innings of my career here, working on brand new interfaces on Vision Pro, has been a gift I'll spend a long time trying to repay ❤️ A few things I’ll never forget: (1) Design around the magic moment: Building a good product is really about finding the one moment that does the convincing and building everything else around it. You'll know you've found it when someone smiles without meaning to. I'll never forget the first time a butterfly landed on my finger inside Vision Pro, and my body believed it before my brain did. (2) It takes research to will products into existence: Most people treat research and product like a handoff, where researchers figure out what's possible and the product team figures out what to do with it. The best work happens when both sides are in the same room arguing about the same thing. You don't know what the research is for until someone shapes how a person uses it, and you don't know what to shape until the research tells you what's possible. (3) The best products are arguments, not compromises. Every product is the output of thousands of decisions, and at most companies, each one gets averaged. The result is defensible in every meeting and exciting in none. Great products feel like someone meant them. The work isn't making good decisions, it's protecting the ones that matter from being negotiated into mush. Thank you to everyone who taught me, pushed me, and trusted me with hard problems. You know who you are (and by that I mean more of you should be on X haha) We're at a real shift in how products work, and in the interfaces we'll use to build and interact with them. These shifts only come around every couple of decades, and I couldn't imagine a more exciting time to be a builder. Excited to share what's next soon!!

ive always held that full immersion is sliiiightly overrated - any screen can feel immersive if you're locked in enough - sent from my iphone while laying in bed

The streets feel like GTA V SF edition

My hot take is that continual learning is as much a product problem as it is an algorithm problem You can't learn from what you can't see. And measuring "did this actually help?" requires long-horizon signals that most model calls never get The hard parts: - Thumbs up ≠ the answer was correct - "Looks good" ≠ it worked in production - Fast feedback ≠ valuable feedback Real signal comes later: - Did the user return? - Did the strategy work? - What happened downstream? The winner won't be the first to continual learning. It'll be the first to know what to actually learn.