8VC

AI is getting more capable and harder to measure. Benchmarks test contrived tasks. Test sets leak. Leaderboards get optimized against. Labs grade their own exams. As models, harnesses, and applications multiply, one question becomes increasingly important: what can these systems actually do? Vals AI is building the trusted answer. Vals builds benchmarks the way work is actually done: evaluations authored alongside practitioners on real professional tasks, with private test sets that can’t leak into training data. What started as an academic exercise is quickly evolving into critical infrastructure for AI adoption. We led Vals’ seed, and are proud to continue partnering with Rayan K. Krishnan, Langston Nashold, and the team as Andreessen Horowitz leads their Series A. To go deeper, 8VC’s Vivek Gopalan sat down with Rayan to discuss the evolution of benchmarks, measuring increasingly agentic systems, the weird and wonderful ways we test intelligence, the global AI race, and why evals sit on the critical path to ASI. Watch the full conversation here: https://lnkd.in/gWAv6vV5 Our thesis on why independent measurement will become a critical layer of the AI stack: https://lnkd.in/gPqPNpFu

Nick Perry, Doug Proctor, and the intrepid Candid Health team are tackling one of the most glaring inefficiencies in healthcare and bringing a new standard of financial infrastructure to the industry. We're proud to have backed them since their seed, and to renew our support today.

We're excited to announce that 8VC led Sable’s $45M raise alongside Sequoia Capital. Explaining what your product or service does is one of the few parts of business that never got cheaper. Marketing creates the assets. Sales tells the story. Experts answer the hard questions. The highest-value interactions have always depended on people. The gap between what a product can do and what customers understand has quietly become one of the biggest constraints on growth. That wasn't a limitation of the work. It was a limitation of the tools. Sable changes the available toolkit. Their AI employee, Aidan, can see your screen, click through a live product, and explain it in real time. Not a recording of the expert—the expert, at the marginal cost of software. As AI gains vision, computer use, and real-time voice, expert guidance no longer has to be limited to a company's largest customers or highest-value interactions. Companies can deliver live, personalized expertise at a scale that wasn't previously possible. That's why Sable is already in production with category-defining AI companies like Notion and Decagon, as well as large public companies, with more than 150 customers on the waitlist ahead of public launch. As models improve, new possibilities open faster than any roadmap can absorb. We're proud to partner with Nim Ravid, Linda He, Leon Chen, Itamar Rocha Filho, and the entire Sable team as they build the future of customer interaction. Read our investment thesis below: https://lnkd.in/erq5V8G6

@Bobyard's march to unleash productivity across the building trades continues with today's launch of Bobyard for Finishes Trades!

We're proud to announce our investment in Generalist and their $400M round. Robotics is approaching the same inflection that large language models brought to software: the shift from narrow specialists to general-purpose foundation models that improve across every task at once. The crux of that shift is dexterity. It's the hardest unsolved problem in robotics, and the clearest path to everything that follows. Cracking it takes a rare combination of research depth and commercial focus, the kind that almost never coexists in one team. Generalist has both: a founding team among the most-published researchers in the field, relentlessly focused on what actually ships and compounds. We've followed their progress across successive generations of models, and the signal has been unmistakable: rapid adaptation to new robots and tasks, strong sample efficiency, and real commercial traction. Proud to be on the journey with Pete Florence, Andy Zeng, and the team. More on our conviction here, from Alex Kolicich and Vivek Gopalan: https://lnkd.in/gYSgzKrv

Thermal protection up to 600 F, now in convenient sprayable form…Cambium is cooking!

Today we’re launching Bobyard for electrical. It's our first trade beyond landscaping, and the first of many more trades to come. If you’ve followed me for a while, you know that my philosophy is to tackle problems that don’t have an easy solution, which is why we set our sights on electrical after landscaping. Electrical estimating is some of the most manual work in all of construction, tracing wire runs, counting fixtures, and flipping between sheets to cross-reference a single circuit. So, we built AI that reads electrical drawings the way an estimator does, but all at the same time, and in minutes. We’re stoked for electrical contractors to try it out and what it means to offer Bobyard to more trades. This launch is huge because it’s the next step to build custom AI models and products for every trade in construction, trained on the drawings each trade actually works with. And electrical is proof that our approach translates across them.