108+ Startups In, And Counting...
Hey there, investors and innovators!
Every quarter, hundreds of founders funnel into Y Combinator hoping to build the next Airbnb or Stripe. As a firm that invests in every YC batch, we get a front-row seat. Each week during the program, we're pulling back the curtain on what's actually launching.
We're now in the stretch between kickoff and Demo Day where founders start soft-launching product publicly, which makes this one of the best windows all year to see where the next generation of startups is placing its bets. As of this week, YC's public directory shows 108+ Summer 2026 companies that have already come out of stealth, with new ones added daily.
The Bigger Themes
The freshest launches in the S26 directory lean heavily into agentic infrastructure and vertical AI operations. That being said, these are the overall insights that seem to govern this batch so far:
- AI has moved from feature to foundation. YC's Summer 2026 Request for Startups spanned 15 categories across AI, hardware, defense, agriculture, and space; a notably more prescriptive wish list than in past summers.
- Agent reliability is the new wedge. A cluster of this week's launches (monitoring, evaluation, red-teaming, production fixes) is coalescing around making AI agents trustworthy in the real world, rather than just building more of them.
- Deep tech is getting real attention. Chip supply-chain tooling, space-adjacent infrastructure, and industrial robotics are showing up more than they have in recent batches; a signal worth watching if you follow hardware alongside software.
Why we're watching
We invest across every YC batch, which means the weeks before Demo Day are our highest-signal window. Here's what caught our eye this week from the Summer 2026 (S26) cohort.
Compiled from Y Combinator's public Startup Directory and Requests for Startups. Have thoughts on this week's drops? Hit reply, we read every response.
Fund Manager Picks
Fabraix
The world's frontier hacker for AI agents
Why it's interesting
Fabraix (YC Summer 2026) is a San Francisco-based AI cybersecurity startup building Nyx, an automated red-teaming agent that continuously finds security vulnerabilities in customer-facing AI agents. Nyx tests chat, voice, browser, and coding agents through the same user interfaces (no source-code access or special integrations), leveraging a library of 10,000+ classified jailbreaks and adaptive multi-turn attack strategies. The company says Nyx has uncovered vulnerabilities in agents at dozens of Fortune 500 companies and scored a 78% attack success rate on the AgentHarm benchmark (vs. 67% for GPT-5.6 Sol). Founders: Ahmed Aly (ex-Two first data scientist; built $1B+ fraud engine) and Ibrahim Abdu (ex-Meta SWE on AI agents; early engineer at Two).
OneCLI
The Identity Gateway for AI Agents
Why it's interesting
OneCLI (San Francisco; YC Summer 2026) is an open-source credential isolation gateway for AI agents. Instead of exposing real API keys or OAuth tokens to agents, OneCLI shows placeholder tokens and injects secrets at the network layer, reducing leakage risk from compromised or misbehaving agents. The product provides per-agent access controls, audit logs, and one-click integrations with Gmail, GitHub, Jira, Slack, and 50+ other apps. Founded by Guy Ben Aharon (first engineer at Argon Security, acquired by Aqua; ex-Aqua, Wix, IDF Unit 8200) and Jonathan Fishner (ex-Principal Security Architect at Axis; ex-Fiverr; IDF Unit 8200). Traction includes 2.5K+ GitHub stars, 300K+ downloads, and adoption as the default credential layer for NanoClaw.
Palette
An interactive AI video generation and editing platform
Why it's interesting
Palette (YC Summer 2026) is a San Francisco-based interactive AI video generation and editing platform for creating, personalizing, and maintaining multimedia content at scale. The product unifies text, image, and video generation, editing, and transformation in a single interface, letting users direct changes via natural-language prompts rather than juggling fragmented creative tools or manual production workflows. Palette targets use cases including customer experiences, product content, education/training materials, and internal communications, aiming to make professional-quality media production more accessible and scalable. The company is founded by Serena Pei (CTO; MIT CS). Traction, customers, and non-YC investors have not been disclosed.
OpenTrade
Investing for the rest of us
Why it's interesting
OpenTrade (YC Summer 2026) is a Seattle-based consumer investing startup building a retail trading experience aimed at “hedge-fund level” decision-making for everyday investors. The product experiments with a new UX: each morning users swipe through 10 market “theses” generated by proactive agents, helping investors quickly discover and act on trade ideas. By packaging research and idea generation into a lightweight daily workflow, OpenTrade targets faster, more guided investing versus traditional brokerage interfaces. The company is led by founder Ojas Kandhare. OpenTrade is early-stage (team of 3); no specific customer, revenue, or non-YC investor traction has been disclosed publicly.
LemonLime
No-code AI that learns your business and self-creates automations
Why it's interesting
LemonLime is a San Francisco-based workflow automation startup building no-code AI that “learns your business” and self-creates automations. The product connects to existing tools to unify scattered company context (“company brain”), then generates specialized AI agents for functions like sales, marketing, and support; it also recommends one-click automations for teams unsure where to start. Founded by Daniela Muñoz (Co-Founder/CTO; ex-Google/Microsoft SWE intern; dual BS CS/HCI, Carnegie Mellon; previously co-founded Confetti) and Jordan Zietz (former CEO/founder scaling consumer apps), LemonLime is a 5-person team in YC Summer 2026. No customers, revenues, or non-YC investors disclosed yet.
Robocurve
We measure how well robots can do real-world jobs
Why it's interesting
Robocurve (San Francisco; YC S26) builds open-source tools and independent, reproducible benchmarks to measure how well robots perform real-world jobs. Its platform aims to replace unverified demo videos and opaque in-house lab testing with standardized evaluations run on physical hardware and real-world setups, where sim-to-real gaps often appear. The team has shipped v1 of its open-source framework, Inspect Robots, and completed an initial pilot scoring a frontier model on a real robot. Founded by CEO Jay Chooi (Harvard stats/CS; ex-UK AI Security Institute, MATS; contributor to Inspect Evals) and CTO Aris Zhu (Harvard CS/physics; ex-Amazon AGI Labs/Robotics; robotics systems engineer).
