Y Combinator Summer 2026
Requests for Startups
16 problems YC wants founders to solve. Mapped to existing companies, solutions being built, and market gaps still wide open.
Read the official RFS at ycombinator.com/rfs16
Problems
8
AI / Software
5
Hardware / Space
60+
Companies Mapped
The Problems + Landscape
Each card: YC's problem statement, who's already building, and what's still missing.
#01
Garry Tan
AI for Low-Pesticide Agriculture
"Modern agriculture runs on chemicals. Pesticide residues are everywhere: in food, in water, in soil."
Gap
All hardware-first. Pure-software layer (satellite imagery + ML predictions on existing farm equipment) is wide open.
#02
Jon Xu
AI-Native Discovery Engines
"For centuries, scientific discovery has run on the same loop: hypothesize, experiment, interpret, repeat."
Gap
Existing tools assist literature review only. Nobody automates the full loop: hypothesis generation, experiment design, result interpretation, next-step recommendation.
#03
Gustaf Alstromer
AI-Native Service Companies
"AI models are now able to do complex work far beyond engineering. These companies sell services rather than software."
Companies
Solution
AI Tax Filing: Automate expat tax optimization (NL/DE/US cross-border). Replace human advisor with agent that knows your full financial picture.
Gap
US-focused market. EU expat tax, cross-border pension optimization, Dutch Box 3 strategy massively underserved.
#04
Ankit Gupta
AI Personalized Medicine
"Intelligent agents are enabling a new level of personalization in medical care."
Gap
Consumer personalization is vaporware. Real gap: clinical decision support integrating with existing EHR workflows.
#05
Tom Blomfield
Company Brain
"The biggest blocker to AI automation is no longer the models. Now the blocker is the domain knowledge."
Companies
Solution
Structured Memory Protocol: Not search. Typed, layered persistence that makes knowledge executable by agents. Auto-capture decisions, corrections, context. Agent-invokable institutional memory.
Gap
Glean is search. Guru is wiki. Nobody makes knowledge executable by agents. "Skills file" framing is new: turn institutional knowledge into agent-invokable capabilities.
#06
Tyler Bosmeny
Counter-Swarm Defense
"The next wave is worse: not one drone, but swarms."
Companies
Gap
Existing counter-drone is one-at-a-time. Many-vs-many coordination (swarm-vs-swarm) is a research frontier.
#07
Ankit Gupta
Dynamic Software Interfaces
"Before AI, users of a piece of software all interacted with the same interface." AI could enable radical per-user customization.
Solution
Adaptive Agent Interface: AI that changes its own behavior, communication style, model tier, and toolset based on learned user context. Not generating UI, but UI that adapts itself continuously.
Gap
v0/Builder generate static UI from prompts. Nobody builds interfaces that adapt themselves continuously from observed behavior.
#08-10
Hardware + Space
Electronics in Space, Hardware Supply Chain, Industrial Space Capabilities
Three hardware/space ideas: inference chips for orbit, faster US prototyping, and lunar resource processing.
Key Companies
#11
Diana Hu
Inference Chips for Agent Workflows
"Most AI chips are designed for 'prompt in, response out.' Agents don't work that way."
Companies
Gap
Current chips optimize batch throughput. Agent workloads are latency-sensitive, branch-heavy, stateful. Nobody optimizes silicon for the agent loop pattern.
#12
Jared Friedman
SaaS Challengers
"AI has collapsed the cost of producing software by 10-100x, and that changes everything."
Companies
Gap
Every vertical has a Cursor-moment waiting. ERPs, CAD, SCADA, industrial control wide open. Pattern: take 10-year-old tool, rebuild with AI at core (not bolted on).
#13
Aaron Epstein
Software for Agents
"The next trillion users on the internet won't be people, they'll be AI agents. While everyone else is building agents, the biggest opportunity might be building the software those agents depend on."
Companies (11)
Solution
Agent Intelligence Layer: Model routing (auto-dispatch to cheapest capable model, <50ms). Skill auto-discovery (agent finds its own tools). Structured persistent memory (typed, pruning, cross-session). Multi-agent bridging (Claude + Codex + Kiro collaborate on tasks).
Unfilled Gaps
1. Cost-aware model routing (nobody auto-selects tier per prompt)
2. Cross-model skill portability (Claude skills don't work in GPT)
3. Memory that prunes itself (existing = raw chat history dump)
4. Agent-to-agent handoff protocol (multi-vendor collaboration)
2. Cross-model skill portability (Claude skills don't work in GPT)
3. Memory that prunes itself (existing = raw chat history dump)
4. Agent-to-agent handoff protocol (multi-vendor collaboration)
#14
Harshita Arora
Startups That Want to Sell to Huge Companies
AI makes it possible for small teams to land major enterprise deals quickly by solving key problems for Fortune 100 companies.
Companies
Gap
Not a product gap, a motion gap. AI enables 2-person team to ship enterprise-grade solutions. Blocker is distribution + first deal, not capability.
#15
Diana Hu
Supply Chain 2.0 for Semiconductors
"A single advanced AI chip goes through 1,400 process steps, crosses a dozen countries, and takes five months to build."
Companies
Gap
Generic supply chain tools don't handle ITAR/EAR export controls or multi-tier fab-to-packaging visibility. Semiconductor-specific tooling barely exists.
#16
Diana Hu
The AI Operating System for Companies
Need for a connective layer that makes companies legible to AI by default. Integrating all business tools and data into one queryable system.
Companies
Solution
Unified Agent OS: Single intelligence layer connecting Slack + GitHub + Outlook + CRM with persistent memory, skill routing, and context continuity. Not a chatbot on search. An agent that acts, remembers, and gets cheaper over time via model routing.
Gap
Glean = search. Moveworks = IT tickets. Nobody builds the brain layer that persists knowledge + executes across all tools. The OS framing (not assistant) is unsolved.