Y Combinator named the areas in which it wants to see new startups

The Summer 2026 list opens with a common theme: AI has ceased to be a feature and has become the foundation. The Fund is looking for startups that are rebuilding software, services, and hardware — and advancing AI into the physical world.

AI for pesticide-free agriculture

Today, farmers are heavily dependent on chemicals, but this no longer works as it used to: pests adapt, more spraying is required, costs rise, and risks increase as well. Additionally, pesticide residues are found in food, water, and soil.

An alternative is emerging. AI can recognize weeds and pests in real time, while robots can treat them precisely without drenching the entire field. Technologies have become cheaper, making this scalable.

Parallel to this, biological solutions are developing: microbes, RNA, and genetics help plants defend themselves.

Combining all of this can reduce pesticide use by up to 90% while increasing yields. This represents a massive opportunity for the next generation of agtech companies.

AI-native service companies

AI is evolving rapidly and can already perform complex tasks, not just write code.

The previous path was: service → SaaS → AI copilot. Startups following this model were active in 2023–2025. The next step now is AI-native companies. They sell not software, but ready-made results. They don’t provide tools; they do the work for you.

Why this matters: the services market is much larger than the software market. Many services are already outsourced, meaning they are the easiest to replace with AI. These include:— insurance— accounting and auditing— compliance— medical administration. The focus is shifting: not improving services, but replacing them entirely. This is exactly what YC is interested in now.

AI-driven personalized medicine

Intelligent agents are opening a new level of personalization in medicine. It is now possible to analyze an individual’s specific data — diagnostic results, genomic scans, electronic health records, or wearable device data — to receive precise recommendations tailored specifically to them, rather than to an average patient.

Two important shifts are happening in parallel:
— diagnostics are becoming cheaper, with new tools for early disease detection
— personalized therapies are also becoming more accessible and easier to implement.

Together, all this should lead to a revolution in healthcare delivery. The abundance of data and intelligence will help patients more accurately assess their risks and democratize access to treatment for the most serious diseases.

The “Company Brain”

The main barrier to AI automation in companies today is no longer model quality, which has improved too rapidly. The barrier is now domain knowledge. In every company, critical knowledge is scattered everywhere. Some of it lives in employees’ heads. Some is buried in old emails, Slack threads, support tickets, and databases. Companies function because people vaguely remember where this knowledge lies and how to apply it. AI agents cannot do this.

For a business to truly run on AI, a “company brain” is needed. This is a system that collects knowledge from various sources, structures it, updates it, and turns it into clear instructions for AI agents.

Importantly, this is not just document search or a chatbot. It is a living map of how the company operates — how returns are processed, decisions are made, and incidents are handled. With such a system, AI will not just assist but reliably perform work.

Defense against drone swarms

The threat is shifting from individual drones to swarms — cheap, autonomous, and coordinated. Current defenses are largely ineffective against them.

The main problem is economics: intercepting a $500 drone with a million-dollar missile is not cost-effective. Also, defense systems today are fragmented and do not provide a unified picture. What is needed is not point solutions, but a full-stack defense against swarms:
— interceptors that can neutralize multiple drones simultaneously
— software that unifies all sensors into a single real-time system
— alternative methods, such as physical or technical means of disabling drones
— attacks on the drones’ autonomy itself, not just their communication.

The key idea: swarm defense is no longer about weapons, but about IT systems. Those who build “digital defense” rather than just hardware will win.

Dynamic interfaces

In the past, everyone had the same interface — at most, you could change the theme. Even “personalization” meant the same screens, just with different content. Therefore, software was built “for everyone at once,” even though people’s scenarios differ.

AI is changing the approach. Now, users can customize the product for themselves using agents. For one person, email is a task list; for another, it is a calendar.

The underlying foundation remains the same — basic components from developers.

Looking ahead: software will be built like a constructor that can be deeply customized, not just have its appearance adjusted.

Chips for space

Thanks to reusable rockets from SpaceX and Stoke Space, humanity is on the verge of gaining access to significantly larger launch volumes. This means that space will require massive new computing power — especially inference chips. These are chips slightly optimized for space conditions: mass, heat dissipation, and radiation resistance. According to YC partners, the market for them will be colossal, and they are best created by people with chip design experience at SpaceX or NVIDIA.

Supply chain for hardware manufacturing

YC is increasingly investing in hardware — from wearables to robots. But making hardware in the US is still slow, especially compared to China. In Shenzhen, the path from design to part can take a day; in the US, it takes weeks. And with each iteration, this gap grows.

The problem is not just supply, but speed: China has a dense network of manufacturers and rapid coordination between design and production. YC is looking for startups that will accelerate this process: rapid part production, fast iterations, and a unified link between design, production, and logistics.

Industrial capabilities in space

YC wants to see startups that will develop industrial capabilities on the Moon and in space — primarily resource extraction: silicon, aluminum, iron, titanium. This can be done through electrolysis and 3D printing of complex structures from molten lunar regolith. In low gravity and without the need for supporting structures, this may prove more efficient than production on Earth.

Inference chips for agentic tasks

AI agents work differently than standard “query-response” models. They take many steps: calling tools, branching, returning, and maintaining context.

Current chips are not optimized for this. GPUs are loaded only 30–40% because the load is “bursty” — memory, I/O, and CPU constantly alternate. New chips tailored for the agentic cycle are needed: fast context switching, memory handling, and support for complex action chains.

The key point: it is not just the chip, but the software. As Groq’s experience showed, the compiler that efficiently utilizes hardware is decisive.

The next generation of winners will likely build both the chip and the software for it.

Attacking legacy SaaS

Many say AI will “kill” SaaS. For large companies, it is a risk; for startups, it is a huge opportunity. Previously, SaaS won because custom software was too expensive. Now, AI has reduced development costs by 10–100 times — and old advantages are disappearing.

How to attack the market:— create the same product but at a fraction of the cost— build an AI-native product from scratch — rethink the process itself, not just add a chatbot— combine multiple SaaS products into one— release an open-source alternative and monetize through services

The main thing is to go not only into simple niches, but into complex and “closed” ones: ERP, chip design, industrial systems. Places that were previously almost impossible to enter.

Software for agents

The next trillion internet users are not humans, but AI agents. They are already browsing websites, conducting research, making purchases, and managing CRM — but they do so on top of interfaces created for humans with mice: this is slow, unpredictable, and unreliable.

Agents need a completely different foundation. Instead of visual forms, buttons, and dashboards, they need machine-readable interfaces — APIs, MCP, CLI. They need detailed documentation that allows them to discover a tool, register, and immediately start using it programmatically, without human involvement in the loop. This means all existing software needs to be rewritten from scratch — for agents as the primary users. This new product will not come from large players who simply add “agent support.” It will come from startups that build everything for this audience from the start. While everyone else is building agents, the biggest opportunity is building the software that these agents will rely on.

Sales to largest corporations

Paul Graham’s classic advice is to sell to other startups: fast cycles, smart users, live feedback. But AI has had an unexpected side effect: for the first time in history, YC’s small startups are closing multi-million dollar pilots with Fortune 100 companies even during the batch. A startup’s first customer can now be one of the largest companies in the world — and this is no longer an exception.

Executives at the world’s largest companies are actively seeking teams that can solve key problems using AI — they are not hiding behind computers; they are ready to talk.

Supply chain for semiconductors

One AI chip goes through ~1,400 production stages, dozens of countries, and months of work — and all of this is still managed via Excel, SAP, and phone calls. In 2021, a shortage of one $300 chip halted $50 billion worth of car production. The problem is that companies cannot see the entire supply chain.

The situation is now even more complex: bottlenecks in chip packaging, memory shortages, and constant export restrictions. Meanwhile, new fabs are being built in the US, and supply chains need to be reassembled.

But tools are almost non-existent: there is no proper control over capacity, risks, and logistics in real time. This opens an opportunity for new startups that can manage the entire supply chain as a unified system.

AI operating system for companies

AI-native companies make all company data available to AI: calls, tickets, tasks, communications — everything is collected into a single system.

This turns the business into a “closed loop”: the system automatically tracks results, compares them with plans, and adjusts actions. As a result, teams work faster and produce more.

The problem is that assembling this is currently difficult — it requires manually linking Slack, GitHub, Notion, and other services. But there is an opportunity: to create a unified layer that combines all company context and turns it into a system that automatically analyzes and helps manage work.