How Three YC Startups Built Their Companies with Claude Code

If you have been watching how fast early-stage startups are shipping products today, you already know something has changed. The gap between idea and working product, which used to take months and a full engineering team, is collapsing fast.

Claude Code is a big part of why. It is a terminal-based agentic coding tool from Anthropic that reads your codebase, plans implementation, writes code across multiple files, and iterates on its own all from a single prompt. For YC startups using Claude Code, this has not just improved productivity. It has changed who can build a company in the first place.

Y Combinator has launched over 5,000 companies with a combined valuation of over $800 billion since 2005, including Airbnb, Stripe, and DoorDash. Today, three of its portfolio startups HumanLayer, Ambral, and Vulcan Technologies  are showing exactly what it looks like to build a company with Claude Code at the centre of your development workflow.

This is their story. Let us get into it.

Quick TL;DR Summary

• HumanLayer (YC F24): Built their entire platform with Claude Code, pioneered context engineering practices, and launched CodeLayer to help teams scale AI-first engineering.
• Ambral (YC W25): A solo CTO used Claude Code for development and Claude’s Agent SDK to build a multi-subagent research engine powering AI account management.
• Vulcan Technologies (YC S25): Two non-technical founders built a working government prototype in weeks using Claude Code, won a state contract, and saved Virginians over a billion dollars annually.
• The pattern across all three: Claude Code compresses weeks of work into hours, enables non-engineers to ship real products, and scales across entire engineering teams.
• What this means for founders: Agentic coding is rewriting the startup playbook from how you hire to how you build, review, and ship software.

What is Claude Code?

Claude Code is Anthropic’s agentic coding tool. Unlike standard AI chat tools where you paste code back and forth, Claude Code runs directly in your terminal and has access to your entire codebase from the start.

It can read files, write across multiple files simultaneously, run commands, fix errors, and complete entire feature builds from a single natural language prompt. It does not just generate snippets  it plans, implements, and iterates.

For YC startups using Claude Code, the practical impact is significant. Development cycles that previously took a week can compress into hours. Solo engineers can manage codebases that would normally require a team. And founders without engineering backgrounds can ship working prototypes.

HumanLayer: From SQL Agents to Scaling AI-First Engineering Teams

The Problem That Started Everything

Dexter Horthy was building autonomous AI agents to manage SQL data warehouses when he ran into a wall that every agent builder hits: companies were not comfortable giving AI applications unsupervised access to sensitive operations.

“Our MVP was an agent that would coordinate with humans in Slack and could do basic cleanup, like dropping any table that hadn’t been queried in 90+ days,” Horthy explained. “We weren’t comfortable with an AI application running raw SQL unsupervised, so we wired in some basic human approval steps.”

That insight that the most useful agentic functions are also the riskiest  became HumanLayer’s core product. In August 2024, Horthy built an MVP, demoed it to startups across San Francisco, and landed his first paying customers. That traction got HumanLayer into the YC F24 batch.

Building Everything with Claude Code

Through Q1 2025, the HumanLayer team talked to dozens of engineering teams and realised something bigger: every team had rolled their own agent architecture, and there were no shared patterns for how to build reliable agent systems.

This led Horthy to publish “12-Factor Agents: Patterns of Reliable LLM Applications” in April 2025. The guide went viral and became a foundational reference for the discipline of context engineering.

When Anthropic launched Claude Code, Horthy’s team was already using Claude models heavily. They moved immediately. “We just wrote everything with Claude Code,” Horthy said. “When the Claude Agent SDK launched with Opus 4 and Sonnet 4, enabling headless agent execution, we knew this was going to be a big deal.”

CodeLayer: Scaling Claude Code Across Teams

HumanLayer’s current product, CodeLayer, helps engineering teams run multiple Claude agent sessions in parallel using worktrees and remote cloud workers. But the real discovery was not about tooling — it was organisational.

“Once you have multiple people on your team shipping AI-written code, you have a completely different type of problem,” Horthy explained. “It’s a communication, collaboration, tooling, and management problem. You have to rewire everything about how your team builds software.”

Since Q4 2025, HumanLayer has closed several large pilots with engineering teams of all sizes — all of it built with Claude Code.

Ambral: Building Production Systems with Subagents

The Problem: Account Management Does Not Scale

Jack Stettner and Sam Brickman founded Ambral to solve a problem every B2B startup eventually hits: the founder-level customer intimacy that drives early growth becomes impossible to maintain as the company scales.

Account managers routinely handle 50 to 100 accounts at once. Customer context that once fit in a founder’s head gets scattered across CRM systems, Slack threads, meeting transcripts, and product usage logs. “You can’t give an effective account management experience with 1/50th of someone’s attention,” Stettner explained.

Ambral synthesises signals from all of these sources into AI-powered models of every customer account, surfacing who needs attention and why — and catching early signs of churn before they become problems.

The Development Workflow: Opus for Thinking, Sonnet for Building

As the sole engineer at an early-stage startup, Stettner needed to move fast without breaking things. He built a precise three-phase development workflow using different Claude models for different jobs:

• Research phase (Opus 4.1): Deep research on the background needed for a feature. “I think the most important thing is doing research before you plan,” Stettner said. “Have Claude do research for you and create a large, long research document.” He runs multiple subagents in parallel to research different areas of the codebase simultaneously.
• Planning phase (Opus 4.1): Create a structured plan with discrete phases for implementation. Stettner reviews and revises this plan manually, then discusses edge cases with Opus before locking it in.
• Implementation phase (Sonnet 4.5): Execute each phase of the plan systematically. “Sonnet 4.5 has been absolutely killer in terms of being able to go and implement these plans,” Stettner said.

“I tried every coding tool, and I experimented with basically every model. I just think Anthropic’s models are the best at tool use right now, and that translates to code,” Stettner said.

How the Product Mirrors the Development Workflow

The architecture of Ambral’s product directly reflects how Stettner builds it. He used the Claude Agent SDK to build a research engine with a dedicated sub-agent for each data type usage data, Slack messages, meeting transcripts, product interactions with each sub-agent specialised in understanding its particular data source.

Vulcan Technologies: Empowering Non-Technical Founders to Ship Products

Building a Government Product With No Engineering Background

Tanner Jones and Aleksander Mekhanik co-founded Vulcan Technologies in April 2025 to solve a problem that has been accumulating for centuries: regulatory code complexity. Virginia’s legislative history stretches back over 400 years, and the buildup of regulations during that time has created one of the most complex regulatory codes in the United States.

There was just one problem. Neither founder had a traditional engineering background. Mekhanik studied ML and mathematics. Jones’ last programming experience was an AP JavaScript class in high school where he wrote code with pen and paper.

Despite this, the two built a working prototype for Virginia’s governor’s office by May 1st, 2025 and won the contract over established consulting firms.

“The entire prototype was made using Claude,” Jones explained. “This was pre-Claude Code. It was literally copy-pasting scripts into the web app, swapping out methods.”

What Claude Code Made Possible

After the prototype win, Vulcan hired a CTO with experience at Google on Gemini and Waymo. When Claude Code launched in June 2025, the team’s development velocity multiplied again.

Vulcan’s AI-powered regulatory analysis helped identify redundant and duplicative requirements in Virginia’s code. The result: the average price of a new home in Virginia dropped by $24,000, saving Virginians over a billion dollars annually. Virginia’s governor was impressed enough to sign Executive Order 51, mandating the use of agentic AI to power state regulatory review.

What All Three Startups Have in Common

Across HumanLayer, Ambral, and Vulcan, the same patterns show up regardless of industry or team size:

• Claude Code is not used for snippets it owns entire workflows. Every team uses it to plan, build, and ship complete features, not to autocomplete individual lines.
• Multi-agent architecture is the default. All three teams run parallel Claude sessions for research, planning, and implementation rather than using a single linear prompt.
• The bottleneck shifts from engineering to coordination. Once Claude Code is embedded in a team, the hard problem becomes managing AI-written code across a team, not writing the code itself.
• Non-technical founders can compete. Vulcan is proof that Claude Code removes the engineering prerequisite for founding a software company.
• Velocity is the compounding advantage. Weeks of work compressing into hours means more experiments, faster iteration, and more product decisions made earlier.

What This Means for Founders Building Today

The YC startups using Claude Code are not just building faster they are building differently. The assumptions that shaped how software companies get started, how many engineers you need, how long an MVP takes, what kind of background a founder needs all of those are being revised right now.

If you are building a company today, the question is not whether to use tools like Claude Code. It is how quickly you can build fluency with them, and how deeply you integrate them into your workflow before your competitors do.

The founders at HumanLayer, Ambral, and Vulcan did not wait to see how the technology matured. They built with it from day one, shaped their products around it, and used the speed it gave them to get to customers and traction faster than traditional development would have allowed.

If you want to learn more about building skills for Claude Code and automating your procedural knowledge, do not miss the chance to enroll in HCL GUVI’s Intel & IITM Pravartak Certified Artificial Intelligence & Machine Learning courses. Endorsed with Intel certification, this course adds a globally recognized credential to your resume, a powerful edge that sets you apart in the competitive AI job market.

Conclusion

In conclusion, YC startups using Claude Code are not just a story about faster development. They are a signal about where company building is heading.

HumanLayer built a platform and an entirely new product category. Ambral built a production-grade multi-agent research engine as a solo engineer. Vulcan built a government prototype as non-technical founders and won a state contract. All three did it with Claude Code at the centre of how they work.

The tools are available. The patterns are documented. The only thing left is to start building.

FAQs

1. What is Claude Code and how does it work?

Claude Code is Anthropic’s agentic coding tool that runs in your terminal and has direct access to your entire codebase. You give it a natural language prompt, and it plans, writes, edits, and debugs code across multiple files simultaneously without you needing to copy and paste between tools.

2. Which YC startups used Claude Code to build their companies?

Three YC startups have publicly documented how they built with Claude Code: HumanLayer (YC F24), which built its entire platform and launched CodeLayer; Ambral (YC W25), which used Claude Code and the Claude Agent SDK to build a multi-subagent account management product; and Vulcan Technologies (YC S25), which used Claude Code to ship a government regulatory analysis product as non-technical founders.

3. Can non-technical founders really build products with Claude Code?

Yes. Vulcan Technologies is direct evidence of this. Neither co-founder had an engineering background, yet they built a working prototype for Virginia’s governor’s office and won a government contract over established consulting firms using Claude before Claude Code even launched. Once Claude Code was available, their velocity increased further.

4. How does Claude Code differ from GitHub Copilot or Cursor?

GitHub Copilot and Cursor primarily assist with code as you write it inside an editor they suggest completions and edits. Claude Code operates agentically from your terminal: it reads your full codebase, plans multi-step implementations, runs commands, and executes entire feature builds end-to-end without requiring you to guide each step manually.

5. What is the Claude Agent SDK mentioned by Ambral?

The Claude Agent SDK is Anthropic’s toolkit for building agentic applications that use Claude models as autonomous sub-agents. Ambral used it to build a research engine where each data source Slack, usage data, meeting transcripts has a dedicated sub-agent specialised in that data type. The SDK enables headless, orchestrated agent execution without a human in the loop for each step.