Skip to content
gentic.news — AI News Intelligence Platform
Connecting to the Living Graph…

Listen to today's AI briefing

Daily podcast — 5 min, AI-narrated summary of top stories

A developer at a laptop in a modern office, Y Combinator logo on monitor, focusing on financial dashboards instead…

Stop Building Interfaces: How YC Runs Finance

Claude Code's agent registry model lets non-coders run operations via deterministic tools — YC's 350+ tool ecosystem proves it. Shift from building interfaces to building small tools.

·Jun 2, 2026·3 min read··29 views·AI-Generated·Report error
Share:
Source: blog.vtemian.comvia hn_claude_code, devto_claudecode, medium_agenticMulti-Source
How do I set up a Claude Code agent registry for non-code teams like finance or legal?

YC replaced the traditional dev-to-finance workflow by giving teams read-only SQL access and a registry of 350+ deterministic tools, letting Claude Code agents handle operations directly — no new interfaces built.

TL;DR

Claude Code's real power is as an agent orchestrator for non-code teams — YC's 350+ tool registry proves it.

What Changed — The Inversion

Claude Code isn't about code anymore. That's not a prediction — it's what's happening right now at Y Combinator and dozens of other organizations.

YC's internal playbook, shared by Pete Koomen, Garry Tan, and Jared, reveals a complete inversion of how software gets built. Instead of finance teams describing workflows to engineers who build purpose-built UIs, YC now gives every team:

  1. Read-only SQL access to their production database
  2. A registry of 350+ deterministic, well-scoped tools
  3. Claude Code agents that orchestrate those tools

The first tool that changed everything? Read-only SQL. Jared built it, felt like he was breaking the rules, pushed it out late at night. It worked extremely well.

What It Means For You

If you're a Claude Code user, this shifts your job dramatically. You stop building interfaces — dashboards, ERP screens, finance SaaS. You start building small, deterministic tools and putting them in a registry that both internal agents and Claude Code on individual laptops can consume.

Koomen's insight from "AI Horseless Carriages" captures it: "AI-native software is the agent wrapping deterministic tools, not deterministic software wrapping an AI." Control shifts from the developer to the user.

The effect is Jevons paradox in action. When asking a complex question stopped costing several hours of someone else's SQL time, people didn't just answer old questions faster. They asked far more questions, and far harder ones, because asking became cheap.

Try It Now — Building Your First Agent Registry

Here's how to apply this with Claude Code today:

1. Start with a single read-only tool

Create a simple MCP server that exposes read-only database access. Here's a minimal example:

{
  "mcpServers": {
    "analytics": {
      "command": "node",
      "args": ["path/to/mcp-server.js"],
      "env": {
        "DB_URL": "postgresql://readonly:token@host/db"
      }
    }
  }
}

Then in your CLAUDE.md:

## Agent Registry

Available tools for non-code teams:
- `analytics/run_query`: Execute read-only SQL. Returns JSON. Use for finance, ops, HR queries.
- `analytics/list_tables`: Show available tables with descriptions.

2. Add tools as teams request them

YC started with ~20 tools. Now they have 350+. Each team adds their own. Your registry grows organically.

3. Let Claude Code orchestrate

When a finance person asks "Show me all journal entries from last quarter that haven't been reconciled," Claude Code calls analytics/run_query with the SQL it generated, gets the result, and formats it — no dashboard needed.

The Shift In Your Workflow

  • Before: Finance describes workflow → you build a UI → they use it → you maintain it
  • After: Finance asks question → Claude Code calls your tools → they get answer → you add more tools

This is why people running factory floors, legal teams, HR, and operations are the best Claude Code users right now. They don't think of it as software. They think of it as the thing that does the work.


Source: blog.vtemian.com

Source: gentic.news · · author= · citation.json

AI-assisted reporting. Generated by gentic.news from multiple verified sources, fact-checked against the Living Graph of 4,300+ entities. Edited by Ala SMITH.

Following this story?

Get a weekly digest with AI predictions, trends, and analysis — free.

AI Analysis

Claude Code users should immediately audit their current workflow for interfaces they maintain that could be replaced by a tool registry. Every dashboard, every internal ERP screen, every carefully designed flow is a candidate for deletion. Instead, expose the underlying data and operations as deterministic MCP tools, then let Claude Code agents handle the orchestration. Second, adopt YC's approach of starting small. Build one read-only SQL tool first. Let your non-code teams discover the power of asking questions directly. The registry will grow organically as they request more capabilities. Track which tools get used most — those are your priorities for expansion. Finally, update your CLAUDE.md to document your agent registry. List each tool, its input/output schema, and example use cases. This makes your Claude Code instance discoverable by your whole team, not just developers.
This story is part of
The Agentic Pivot: How Claude Code Is Forcing a Reconfiguration of the AI Stack
Anthropic's developer tool is becoming the connective tissue between models, infrastructure, and autonomous workflows, challenging OpenAI's application-first strategy.
Compare side-by-side
Pete Koomen vs Garry Tan
Enjoyed this article?
Share:

AI Toolslive

Five one-click lenses on this article. Cached for 24h.

Pick a tool above to generate an instant lens on this article.

Related Articles

From the lab

The framework underneath this story

Every article on this site sits on top of one engine and one framework — both built by the lab.

More in Opinion & Analysis

View all