What YC S24 Startups Are Building with AI

I've been in the YC S24 batch. I know what the hallway conversations sound like, what problems everyone is obsessing over, what tools keep coming up. Here's my honest read on what this batch is actually building with AI — and more importantly, what it says about where we are in the cycle.

The short version: infrastructure is mature, applications are multiplying, and the most interesting stuff is at the intersection of AI and workflows that used to be entirely human.

The Infrastructure Tier Is Largely Settled#

A year ago, there were a dozen YC companies building "better embeddings" or "faster vector search" or "more reliable RAG pipelines." Those companies still exist, but they're not where the energy is in S24. The infrastructure layer — models, serving, embeddings, orchestration — has largely been commoditized.

Founders in S24 who are still pitching infrastructure are pitching it as a layer under something more specific: "faster inference for vision models for retail checkout" is a business. "Faster inference" is a component.

This matters for application builders: the infrastructure you'd have needed to build two years ago now just exists. You can focus on the domain problem.

The Vertical AI Wave Is Real#

The biggest cohort of S24 companies is applying AI to a specific vertical domain — healthcare, legal, logistics, construction, finance. These companies share a few traits:

• Deep domain knowledge from the founders (often ex-practitioners)
• AI doing a task that used to take a highly-trained specialist
• A workflow that's obviously broken and hasn't been touched in 20 years

The pattern works because vertical AI can be wrong 10% of the time and still be dramatically better than the existing process. A legal contract review tool that's 90% accurate is useful because human review was 100% expensive.

What surprises me is how unsexy many of these verticals are. Nobody's excited about crop yield optimization until you see what it means for food supply. The boring verticals are often the best businesses.

AI-Native Tools Are Different from AI-Added Tools#

There's a meaningful distinction emerging between products that were built AI-native versus products that bolted AI on. The AI-added products feel like a chatbot in the corner of a screen. The AI-native products feel like a different category of software.

DenchClaw is built AI-native. The data model was designed so an AI can read and write it. The agent is not an afterthought — it's the primary interface. You don't use a UI to update your CRM and also happen to have a chatbot; you talk to the agent and the agent manages the CRM.

This is the right architecture for an AI-first world. The companies in S24 that are winning are the ones where AI is in the critical path of the core product, not decorating it.

The Local-First Opportunity#

One theme I see specifically in S24: growing discomfort with cloud-native AI tools and the data they process. Founders are sophisticated enough to know that everything going through a SaaS vendor's servers is a potential liability — for their customers, for their own sensitive business data, for their competitiveness.

Local-first AI is a real opportunity, and not just for privacy-conscious individuals. Enterprise customers increasingly want their data on their own infrastructure. Healthcare companies are legally required to keep patient data local. Law firms don't want their case notes in someone else's embeddings index.

DenchClaw is betting on this trend. The entire stack — DuckDB, the agent runtime, the app platform — runs on your machine. The only outbound traffic is API calls to model providers, and you control which ones you use.

I think local-first AI is going to look more and more prescient over the next three years.

What I Don't See Enough Of: Workflow Thinking#

The gap I notice in most S24 AI products is workflow thinking. A lot of companies are building point solutions — "AI for task X." That's fine. But the 10x products are the ones that understand the workflow X is embedded in and optimize the full workflow.

HubSpot didn't win because it did email better than competitors. It won because it understood the full marketing and sales workflow and built a system of record for the whole thing. The AI equivalent is building for the workflow, not the task.

That's what we're trying to do with DenchClaw: not just an "AI that answers questions about your CRM," but a system that manages the full relationship workflow — from first contact to closed deal to customer health.

The Tooling Used by S24 Builders#

If I had to characterize the AI tooling stack of a typical S24 technical founder, it looks like this:

• Coding: Cursor (AI-assisted IDE) for almost all code
• Search: Perplexity for technical research, avoiding Google where possible
• Writing: Claude for long-form, GPT for quick tasks
• CRM/operations: Increasingly, DenchClaw — which is genuinely gratifying
• Deployment: Vercel, Railway, Fly.io depending on use case

The interesting observation: the coding productivity gains are real and compounding. Founders in S24 are shipping at a pace that would have been impossible with traditional tooling. A 2-person team can execute what a 10-person team could before.

That changes the calculus on which markets are worth attacking. You can now go after smaller markets with niche workflows because the marginal cost of building is much lower.

Where DenchClaw Fits#

We're not a vertical AI company. We're building the operating layer for knowledge workers and founders — the thing that sits underneath all your domain-specific tools and gives you a unified, AI-native workspace with your data.

Think of it as the foundation, not the application. Your vertical AI tool might help you draft legal documents. DenchClaw is where you track which clients are getting those documents, who approved what, and when you need to follow up.

The foundation layer is less flashy than vertical applications. It also has much higher long-term retention and a much clearer data moat.

Frequently Asked Questions#

What were the most common types of startups in YC S24?#

AI applications accounted for a large majority of the batch. Within AI, the biggest categories were vertical applications (healthcare, legal, logistics), developer tools, and enterprise automation.

How does DenchClaw compare to other YC S24 companies building CRM tools?#

DenchClaw is the only local-first, open-source AI CRM in the YC ecosystem we're aware of. The alternatives are either cloud-native SaaS tools or point solutions. The combination of local-first + open-source + AI-native is genuinely differentiated.

Is local-first AI a real market or a niche?#

It starts niche (power users, developers, privacy-conscious founders) and expands. The same thing happened with self-hosted software in general. The enterprise market for on-premise AI is enormous — DenchClaw is starting from the bottom up, with individual founders, and growing toward teams.

What's the best way for a first-time founder to apply to YC?#

Build something real before you apply. The number one predictor of getting into YC is having a product that users actually use. Everything else — the idea, the team, the market — matters less than demonstrated traction and genuine user love.

Ready to try DenchClaw? Install in one command: npx denchclaw. Full setup guide →