cd /news/ai-agents/the-hard-part-of-an-agent-loop-isn-t… · home topics ai-agents article
[ARTICLE · art-50617] src=dev.to ↗ pub= topic=ai-agents verified=true sentiment=· neutral

The Hard Part of an Agent Loop Isn't the Loop

Boris Cherny, creator of Claude Code at Anthropic, stated in 2026 that his job is now to write loops rather than prompt models directly. A developer at Puzzle Quest Games describes the bootstrap problem of setting up agent loops, introducing Squad, a human-directed AI development team that runs through GitHub Copilot. The developer notes that Squad is experimental, Copilot-only, and that parallel sub-teams slowed throughput.

read2 min views1 publishedJul 8, 2026

Originally posted on the Puzzle Quest Games blog.

In 2026, Boris Cherny, the creator of Claude Code at Anthropic, told an interviewer he'd stopped prompting the model by hand: "I don't prompt Claude anymore... My job is to write loops," (OfficeChai, 2026). For about a week my feeds treated that clip as proof loops are the next thing in AI coding. I'd been running one since March 2026, and the viral cut skipped the only part that's actually hard.

The Bootstrap Problem. I first ran into this pattern as a "Ralph Wiggum loop," before Cherny's clip made the rounds. Claude Code's /loop

command and Copilot's "Automation" tab chase the same idea, but both are generic: wiring a real loop in Copilot CLI still means writing charters, skills, and routing from nothing, per project. That's boilerplate, the loop was supposed to remove it.

Squad is the bootstrap layer.

Human-led AI agent teams for any project. One command. A team that helps you move faster with your code.

⚠️

Alpha Software— Squad is experimental. APIs and CLI commands may change between releases. We'll document breaking changes in[CHANGELOG.md].

Squad gives you a human-directed AI development team through GitHub Copilot. Describe what you're building. Get a team of specialists — frontend, backend, tester, lead — that live in your repo as files. They persist across sessions, learn your codebase, share decisions, and help you move faster without giving up oversight.

Squad is a productivity tool for humans, not a replacement for engineers, reviewers, or decision-makers. People stay accountable for priorities, approvals, and final changes; Squad helps with coordination, repetition, and parallel execution.

It's not a chatbot wearing hats. Each team member runs in its own context, reads only its own knowledge…

A few commands scaffold an agent charter, then you hand each agent the skills it needs. A loop.md

file is read at the start of every cycle, self-directed or off a GitHub issue queue, and agents file their own issues when they find work outside scope. Mine runs 30 minutes on, 10 off (the break matters: less than 10 and the next cycle collides with a subagent still finishing background work).

Where it falls short: Copilot-only (no good Claude Code/Codex port), and subsquads (parallel sub-teams) slowed my throughput instead of raising it.

Anyone else running agent loops against a GitHub issue queue instead of a fixed prompt? Curious what your routing rules look like.

── more in #ai-agents 4 stories · sorted by recency
── more on @boris cherny 3 stories trending now
sponsored brought to you by zahid.host 4,200+ EU-deployed projects
reading about agents? ship yours in a single git push.

Run your AI side-project on zahid.host

EU-based hosting, git-push deploys, automatic HTTPS, no cold starts. Free tier with a custom domain — perfect for shipping the agent you just read about.

$git push zahid main
Live at https://your-agent.zahid.host
Get free account → Pricing
from €0/mo · no card required
LIVE [news/the-hard-part-of-an-…] indexed:0 read:2min 2026-07-08 ·