AI development has defaulted to Python, but most of the world's application developers write JavaScript. Mastra is the TypeScript framework for building AI agents—from the team that built Gatsby—and it's betting the next wave of AI apps gets built where the web already lives.
Vision language models can describe almost anything in an image. The catch is speed: by the time most of them answer, the moment has passed. Overshoot is building the infrastructure that lets developers run any VLM on live video and get answers in under 200 milliseconds—faster than human reaction time.
The web was built for human eyes and human hands, not for software that reasons. Browser Use gives AI agents a reliable way to see, understand, and act on any website—an open-source project that became foundational infrastructure almost overnight.
Traditional contract review costs thousands of dollars and takes days or weeks. General Legal pairs proprietary AI with US-barred attorneys to deliver Biglaw-quality work at a flat fee, with a median first turnaround under three hours.
Frontier AI has run out of internet to read. The next generation of models—voice, robotics, video—needs data the public web doesn't have, collected with consent and clean provenance. Luel is building the marketplace to supply it.
Deepfakes are cheap, bots are convincing, and AI agents are transacting on their own. In that world, proving a real human is on the other end becomes foundational infrastructure. Didit is building it as a single, programmable API.
Most developers use one AI coding agent at a time. Superset lets you run 10 to 100+ in parallel on your local machine without conflicts. It's a quiet infrastructure shift that changes how software gets built.
AI agents that can browse the web are impressive. AI agents that can control an entire operating system are transformational. Cua is building the open-source framework to make that possible.
Financial services runs on complex software that AI agents can't navigate without specialized training data. Halluminate builds the sandbox environments and datasets that make computer use AI actually work in finance.
CPA firms are drowning in data entry while turning away clients they don't have capacity to serve. Magnetic Tax built an AI agent that scans documents and inputs data directly into legacy tax software with over 90% accuracy.
The return-to-office narrative dominates headlines, but the data tells a different story. Remote work is growing, and platforms like RemoteJobs.org are the infrastructure that makes it findable.
Y Combinator alumni no longer just build startups. They run funds, angel networks, and coordinated investing systems that increasingly shape the earliest rounds.
Most AI products do not fail because the model is weak. They fail because the model is forced to reason over messy, flattened data that has already lost its structure.
For twenty years, blog writing was one of the best ways to get attention on the internet. AI is changing the economics of content in ways that make people wonder if that era is ending.
Subscriptions were always a little artificial. AI is forcing software companies back to an older, more honest idea: charging people for what they actually use.
For twenty-five years the web ran on blurbs that invited curiosity. Now it runs on summaries that satisfy curiosity immediately. The click is disappearing—and that changes everything.
Credentialing delays cost hospitals up to $100,000 per doctor per month. Harbera is betting that AI browser agents can automate the entire workflow—and redefine a category in the process.