YC Just Told You What to Build. Here Are the Ideas Worth Stealing.

Every few months, Y Combinator publishes a list of ideas it wants to fund. Most founders skim it. The smart ones treat it as a roadmap. The Summer 2026 edition is different from previous years. It is more specific, more technically ambitious, and more honest about where YC partners believe the next generation of generational companies will come from. This is not a list of trends. It is a list of gaps. Problems that are now solvable because AI, robotics, and biology have crossed a threshold that makes what used to be impossible suddenly tractable. Read it as a founder. Every idea on this list has top-tier investors already waiting to write checks. Modern agriculture is stuck in a bad loop. Farmers spray more chemicals to fight adapting pests, costs go up, margins go down, and the pipeline for new solutions is slower than ever. What changed: AI can now identify individual weeds and pests in real time. Sensors and cameras are cheap enough to deploy at scale. Robotics can treat one plant instead of blanket-spraying an entire field. And biology has caught up with microbes, peptides, and RNA-based solutions that can replace entire classes of synthetic chemicals. Garry Tan wrote this one personally. The company that cuts pesticide use by 90% while helping farmers grow more food is a generational company, not just a good business. This is the category shift most founders are still missing. 2023 to 2025 was the era of AI copilots. Tools that help humans do their jobs faster. The next era is companies that skip the human entirely and just do the work. Gustaf Alströmer is looking for startups that sell the service, not the software. The total spend on services is many times larger than the spend on software. And because most services are already outsourced, replacing them with an AI-native product is structurally easier than replacing software. The categories YC is most excited about: ▫️ Insurance brokerage ▫️ Accounting, tax, and audit ▫️ Compliance ▫️ Healthcare administration If you are building a company that replaces a service rather than improving it, YC wants to hear from you. So do early-stage investors. Two revolutions are happening simultaneously. The cost of generating personalized diagnostics is collapsing. Genome sequencing has fallen faster than Moore’s Law. New diagnostics are entering the market enabling early detection of health signals that were previously invisible. At the same time, the cost of printing personalized genetic therapies is falling. mRNA delivery vectors now make n-of-1 medicines possible. The FDA is more open to letting patients try these procedures than it has ever been. AI agents like Claude Code can now analyze personalized health data — genome scans, EHR data, wearables — and generate highly accurate, user-specific suggestions. The future of medicine is not population-level treatment. It is one patient, one protocol. The biggest blocker to AI automation inside companies is no longer model quality. It is domain knowledge. Every company has critical know-how scattered everywhere. In people’s heads, in old email threads, in Slack, in support tickets. The company works because humans vaguely remember where that knowledge is. AI agents cannot operate like that. Tom Blomfield wants a system that pulls knowledge out of every fragmented source, structures it, keeps it current, and turns it into an executable skills file for AI. Not a search tool. Not a chatbot over documents. A living map of how a company actually works. Every company in the world will need one. The AI tools exist. The company brain layer does not yet. A swarm of cheap drones took out an AWS data center. Nobody stopped them. A Patriot missile costs $3 million. An FPV drone costs $500. All the cost advantage sits with the attackers. The defense stack built for individual drones will fail completely against coordinated swarms. Tyler Bosmeny wants founders building the counter-swarm stack: high-capacity interceptors, software that fuses every sensor into a single real-time picture, non-kinetic defenses, and attacks on the autonomy stack itself. The key insight: drone defense is starting to look less like operating a weapon and more like running a real-time distributed system. The winning companies will look more like Cloudflare than Raytheon. Every user of a piece of software currently interacts with the same interface. That is about to end. AI coding agents have gotten good enough to let users become their own forward-deployed engineers. Ankit Gupta imagines a world where your email client looks like a task list while a student’s looks like an events calendar — both built on shared primitives, radically different in execution. Software companies will ship primitives with full intention that users will heavily modify the final interface. The whole stack of software delivery needs to be rethought for this to work. That is the startup opportunity. Reusable rockets from SpaceX and Stoke Space are about to produce a massive increase in humanity’s capacity to put things in space. The bottleneck will be compute. Specifically inference chips — slightly optimized for mass, thermal performance, and radiation — built for a market that barely exists yet but is about to explode. If you are working at SpaceX or NVIDIA doing chip design, YC is explicitly calling you out. This is one of the most specific talent-targeted requests in the history of the RFS. In Shenzhen, a team can go from design to a new physical part in one day. In the US, the same loop takes weeks. That gap compounds across every iteration cycle. Nicolas Dessaigne wants startups that produce parts dramatically faster, enable rapid hardware iteration, and tightly integrate design, manufacturing, and logistics. The YC W26 batch already has early examples: Hlabs building actuators, Prototyping.io turning designs into mechanical parts in days. The next generation of great hardware companies will be built on much faster iteration loops. The infrastructure for that does not fully exist yet. The moon has silicon, aluminum, iron, and titanium. Electrolysis works better in low gravity. 3D printing from molten regolith is more efficient without the need for structural supports. Adi Oltean wants founders building the industrial base that makes long-term space presence economically viable. This is a decade-long bet, but the investors writing checks into it are thinking across that horizon. Most AI chips are designed for prompt-in, response-out inference. Agents do not work that way. They loop: calling tools, branching, backtracking, holding context across dozens of steps. Current GPUs hit 30 to 40 percent of peak utilization on these workloads because the work is bursty. Diana Hu wants chips designed for the agent loop: fast context switching between models, native speculative decoding, memory built for KV caches that persist across an entire execution graph. NVIDIA bought Groq for $20 billion because it saw this coming. The founder who understands both chip architecture and how agents actually execute has a rare window right now. The moat that protected legacy SaaS — millions of lines of code built over decades — is gone. AI has collapsed the cost of producing software by 10 to 100x. A five-person startup can now outbuild entrenched incumbents on specific workflows. Jared Friedman wants founders going after the products that seem invulnerable: chip design software, ERPs, industrial control systems, supply chain management. The last generation of great software companies was built by replacing on-premise with cloud. The next generation will be built by replacing legacy SaaS with AI-native software. If your pitch deck makes this case clearly, the investors are already looking for you. The next trillion users on the internet will be AI agents. They need completely different software. Agents are already browsing the web, doing research, making purchases, and managing CRMs — but they are doing it on top of software designed for h…