$ cat workshop.md

Build an AI Agent That Verifies Its Own Work

LIVEMARCH 202 HOURS25 SEATS

You'll implement a RALPH-style agent loop with autonomous verification, so your agent checks its own output before you ever have to.

Book a 15-min call →

$49 · Book a 15-min call — I'll confirm your spot.

$ cat problem.md

The Problem With AI Code Review

AI agents ship fast, but verification didn't get faster with them. Engineering teams are doing more review than they did before agents existed.

10×more PRs

You're using AI agents because they ship code faster than any human team.

91%more review time

Review didn't speed up with them. Someone still reads every output, so it actually got slower.

?%of outputs are wrong

You don't know which outputs are wrong, so you end up checking all of them.

The fix is giving the agent a verification step so it catches its own mistakes.

$ ./build --agenda

What You'll Build

Two hours of real code. You leave with a working verification loop you can run against your own codebase.

0:00

The RALPH loop

How autonomous agent loops work — and why most implementations miss the verification step.

0:30

Building the bash loop

Write a minimal agent loop from scratch. No framework. You own the infrastructure.

1:00

Adding autonomous verification

Wire in a verification layer. The agent runs, checks its own output, and corrects — without you.

1:30

Running it against your codebase

Connect the loop to real code. By the end, you have something that works on your backlog.

You'll leave with:

  • A working RALPH-style agent loop with autonomous verification
  • A verification layer you control and can extend
  • Patterns for catching regressions without manual review
  • A setup you can run against your own codebase today

$ claude-code-camp --instructor

Who's Running This?

Abhishek Ray

Trained 60+ engineers on Claude Code

Founder, Opslane (YC S24) · Ex-Robinhood Engineer

I built Opslane (parallel Claude sessions) and Codient (Claude Code in your browser).

I've lived in Claude Code for the last 6 months. Talked to engineers about what works and what doesn't.

Open-source: CLAUDE.md Templates — starter kit used by hundreds of engineers.

LinkedIn →X (Twitter) →

$ ./apply --seats 25

Apply for a Seat

25 seats. I talk to everyone before confirming a spot, so book a 15-min call and I'll send you everything you need to join on March 20.

Book a 15-min call →

$49 · Live on Zoom · March 20 · 2 hours · 25 seats