What Is YC Funding in Summer 2026? The First S26 Companies Are Live, and They Tell You What to Pitch for F26

The YC Summer 2026 batch is still running, but the first S26 companies have already appeared in YC's public startup directory. If you're preparing a Fall 2026 application, this is the most current answer available to the question every applicant asks: what is YC actually funding right now?

We went through the newest S26 additions. Here's what the data says, and what it should change in your F26 application.

Who are the first YC S26 companies?

Among the earliest S26 companies visible in the directory: Archal (an eval platform that builds stateful clones of SaaS APIs so you can test agents without touching production), Agentcard (debit cards AI agents can issue and spend autonomously), Context.dev (a single API that turns websites into clean, token-optimized context for agents), Codag (systems log compression for agents), Rex (AI agents running enterprise order-to-cash out of London), RealPact (AI paperwork agents for real estate brokerages, starting in New Hampshire), Care GP (AI agents running primary care clinic operations in Sydney), GUILD (an AI-driven "neoprime" defense supply chain contractor in New York), and Tsenta (a consumer job-matching agent built by two founders who applied to 3,000 jobs manually).

Nine companies, and every single one is an AI agent business. But the interesting part is not "YC funds AI." You knew that. The interesting part is the three patterns underneath.

Pattern 1: Agent infrastructure is splitting into narrow, almost absurdly specific layers

In Fall 2025, YC bet on agent infrastructure broadly. By Summer 2026, that category has fractured into picks-and-shovels niches that would have sounded like features a year ago: payments for agents (Agentcard), evals for agents (Archal), web context for agents (Context.dev), log compression for agents (Codag).

Log compression. That's a fundable company now, because when agents run continuously, observability data explodes and someone has to deal with it.

The lesson for F26: "platform for AI agents" is dead as a pitch. The funded shape is one painful, specific layer of the agent stack, named precisely. If you can't say which layer you are in one sentence, a YC reader in 2026 will assume you don't know.

This matters beyond YC too. The same week these companies surfaced, Deno shipped Claw Patrol, an open-source security firewall for agents. Incumbents are entering the obvious layers. Your application needs to answer why your layer is the one a two-person team can still win.

Pattern 2: Not "copilots" anymore. Agents that run entire operations in unglamorous verticals

Look at the vertical companies in the list: Rex runs order-to-cash, a $60B outsourced function, and is already managing over $500M in receivables from Fortune 100 customers. RealPact runs transaction paperwork for real estate brokerages. Care GP runs clinic operations for primary care. GUILD runs defense supply chain contracting.

None of these are assistants that help a human do the work. They are agents doing the work, sold as an outcome with an SLA. And the verticals are deliberately unsexy: receivables, deeds and parcel records, clinic admin, procurement paperwork. This is Paul Graham's schlep blindness thesis playing out in batch composition: the messier and more email-buried the workflow, the less traditional automation could touch it, and the more valuable a stateful agent becomes.

For your F26 application, the question to steal: what function do companies currently outsource to people, in a vertical you know unusually well, that an agent could own end-to-end? "AI copilot for X" reads like 2024. "We replace the outsourced X function and commit to an SLA" reads like S26.

Pattern 3: Teams of one to three, and second-time founders with proof

Team sizes in this group: Context.dev and Codag appear to be solo founders. Archal, Tsenta, Rex, GUILD are two-person teams. RealPact has three.

Two things follow. First, YC is visibly comfortable funding solo founders and duos in 2026, which makes sense when AI writes most of the code (Anthropic says Claude now writes 80% of its own). Second, the bar moved to evidence. Rex's founders spent four years building Sequence, an a16z-backed finance ops platform, to Series A before applying, and won the Vercel AI Accelerator out of 2,500 teams. Tsenta's founders applied to 3,000 jobs by hand before automating it. The pattern isn't credentials. It's earned insight you can date and count.

If you're a first-time founder, you can't manufacture a Series A history. You can manufacture the countable kind of proof: users, revenue, the 3,000 painful manual repetitions that explain why you, specifically, had to build this.

What should F26 applicants do with this?

Three concrete edits to make before the Fall 2026 deadline. State your layer or your function in the first sentence of your description, the way Agentcard ("debit cards for AI agents") and Rex ("AI operations partner for enterprise order-to-cash") do. Replace any "copilot" framing with the outcome you own and the metric you're accountable for. And put your earned insight in numbers: how many times you did the schlep manually, how much volume your agent already handles.

Then pressure-test it. The fastest way to find out whether your application reads like an S26 company or like 2024 leftovers is to have someone who has been through the process tear it apart. That's what we built YC Roaster for: YC alumni read your actual application and tell you, bluntly, where it falls into the patterns above and where it doesn't.

The S26 list will keep growing through the summer, and we'll keep tracking it. But the direction is already legible: specific layers, owned outcomes, small teams with countable proof. Write your F26 application like the directory already includes you.