What YC’s Latest Batch (P26/X26) Reveals About Where AI Is Heading

Y Combinator’s most recent batch (P26, formerly referred to as X26) paints a clear picture of the next phase of AI: not flashy new foundation models, but the messy, practical infrastructure and business models needed to make agents actually useful at scale.

According to a detailed analysis of the ~205 companies in the batch from Extruct AI, the focus has shifted from “build the model” to “build the world around the model.” Here’s the breakdown of the major themes.

1. Dev Tooling Is Being Rebuilt Around Agents

No one is trying to build the next Cursor or Devin from scratch. Instead, founders are creating the surrounding tooling that coding agents actually need to operate reliably in production.

This includes:

Test harnesses and verification systems (e.g., TesterArmy, Armature);
Secure sandboxes and execution environments (e.g., Indexable, StableBrowse);
Monitoring and runtime infrastructure for long-running agents.

A growing “trust” subcategory is particularly interesting: tools that gate every agent action and even sell insurance on agent behavior (examples include Mount and Klaimee). The message is clear — before agents can be trusted with real work, the industry needs guardrails, observability, and accountability layers.

2. Data Marketplaces and Brokers Are Booming

There’s a clear cluster of companies focused on sourcing, licensing, and brokering high-quality multimodal data for training LLMs. YC saw around five such projects in this batch.

What YC’s Latest Batch (P26/X26) Reveals About Where AI Is Heading One standout is Luel, which raised a massive $31.2M seed — one of the largest in YC history for this category. Business models in the space generally fall into dataset licensing from publishers, data marketplaces, or IP brokerage.

The category is heating up fast outside YC as well. Interestingly, YC passed on Mercor back in 2023 — the company that largely defined this category and has since reached a ~$10B valuation. The report notes that many VCs are now actively chasing similar plays.

3. SaaS Isn’t Dead — It Just Got Rebranded as Vertical AI Agents

Roughly 35 vertical-focused companies in the batch are building AI agents for specific industries rather than horizontal tools.

Breakdown:

Healthcare: 9 companies;
Manufacturing/Industrial: 11 companies;
Finance: 7 companies;
Legal/Gov/Compliance: 5 companies.

These aren’t generic chatbots. They’re purpose-built agents solving real operational problems in regulated or complex domains. The era of “vibe coding” general-purpose tools hasn’t arrived here yet — these companies are still focused on deep vertical integration.

4. The Rise of AI-Native Service Companies

What YC’s Latest Batch (P26/X26) Reveals About Where AI Is Heading This is one of the more interesting new directions YC is actively backing. Instead of (or in addition to) selling software, some founders are building AI-powered service businesses.

Why? Selling traditional SaaS in many verticals remains difficult, and someone still needs to deploy, operate, and take responsibility for agents. Service models also command higher willingness-to-pay, especially in regulated industries.

Standout examples:

Clara Health — AI primary care doctor
Hedge Specialty — AI-powered specialty insurance
Alchemize — AI customs brokerage

Successful patterns identified in the report:

Solve an important but non-core function for the customer;
Target areas where clients already pay for human services, consultants, or BPO;
Focus on repeatable work that AI can handle at scale;
Align with roles that are already being automated or consolidated in traditional SaaS buying centers.

If your vertical SaaS isn’t gaining traction, this model is worth serious consideration.

5. Infrastructure for Prediction Markets

A small but fast-growing category: tools built around prediction markets like Kalshi and Polymarket. The number of companies in this area grew from roughly 1 in the previous W26 batch to 5 in P26.

These include arbitrage finders between platforms (e.g., Oddpool), market data feeds (e.g., Dome), and professional trade execution services (e.g., River Markets). It’s classic “picks and shovels” for a growing betting/information market ecosystem.

6. The Self-Managing Company

A genuinely new and ambitious theme: companies whose core operations are largely run by AI agents.

Several startups in the batch are building agents that handle everything from campaign monitoring and landing page creation to customer support. Flagship example in the broader narrative is Polsia — a solo-founder company with zero employees and nine agents that raised $30M at a $250M valuation. It charges $49/month plus 20% of client revenue.

The vision: “ChatGPT, build me a $1M ARR SaaS company — and don’t make mistakes.”

7. Hardware Is Shifting Toward Defense

What YC’s Latest Batch (P26/X26) Reveals About Where AI Is Heading Hardware still represents ~10–11% of the batch (~22 companies), but the focus has moved away from general robotics toward defense and attritable systems.

The core thesis: cheap, mass-producible systems can break an adversary’s economics. A $20K drone that forces the other side to spend $4M on interceptors creates asymmetric pressure. Companies like Tenet and Surtr exemplify this approach.

8. B2C Remains Small and Experimental

Consumer applications are the smallest slice — only about 13 companies (~6% of the batch). Personal finance agents (e.g., Gravy as an “AI CFO”) are seeing some interest, along with games, creator tools, and health. However, the report suggests this category lacks the “vital launches” seen in B2B and infrastructure plays right now.

9. Founder Demographics

Roughly 50% of founders are not originally from the US.
However, 83–91% ultimately relocate and register their companies in the United States.
Top countries of origin for international founders: UK (33), India (~29–33), France (21), plus a strong Nordic cluster.
Notable examples include fintech Hevn, agentic observability play Superlog, and Sardor — reportedly the first YC founder from Uzbekistan, building in robotics.
Strong academic pipelines from Stanford/MIT domestically and TUM/Oxford from Europe.
Solo founders make up 13% of the batch.

The Big Picture

YC’s P26 batch shows AI moving from experimentation to operational reality. The winners aren’t just building better models — they’re building the trust layers, data pipelines, vertical agents, service wrappers, and execution infrastructure that will actually let agents do meaningful work at scale.

The future looks less like “one AI to rule them all” and more like thousands of specialized agents operating within well-defined guardrails, supported by new data economies and, in some cases, entirely AI-run companies.

The full detailed report and company list is available at extruct.ai/research/ycx26/.