← The AI Hype Audit — all 115 verdicts
PARTLY "You're not supposed to prompt Claude — build a system that prompts itself" — a real Anthropic idea, wrapped in a caption that can't decide who said it
The claimAn Anthropic engineer said you're not supposed to prompt Claude — you're supposed to build a system that prompts itself. In 45 minutes 'she' shows exactly how Anthropic builds agents that remember, fix their own mistakes, and get smarter with every run. Beats any paid course on agents.
The idea is real and traceable. Boris Cherny — the creator and Head of Claude Code at Anthropic — has said versions of this on the record, most directly in a podcast interview: he doesn't prompt Claude anymore, he writes loops that prompt Claude and figure out what to do. The tooling behind it is real too: Claude Code ships loops, memory files, subagents, and dynamic workflows, and Anthropic publishes free engineering guides on agent harnesses. What's fake is the wrapper. This exact caption has circulated on X since June — same wording, but the video runtime mutates between reposts (4, 16, 24, 30, 40, and 45 minutes) and the engineer flips between 'he' and 'she' depending on who copied it. That's the signature of accounts recycling a template without watching anything. And true to form, no version of the post links the actual talk.
What holds up
- The core idea is real — Boris Cherny, Head of Claude Code, describes this workflow on record
- His actual quote: 'I don't prompt Claude anymore... my job is to write loops'
- The tooling exists: Claude Code loops, memory files, subagents, dynamic workflows
- 'Beats a paid course' is fair — every real source behind this is free
- Nothing for sale on this repost, at least
What doesn't
- The quote is a punched-up paraphrase, not what he said
- Runtime mutates across reposts of the same caption: 4, 16, 24, 30, 40, 45 minutes
- The engineer flips gender between reposts — 'she' here, 'he' elsewhere. It's Boris.
- Word-for-word copy of June X engagement posts; the actual talk is never linked
- 'Save it for later' save-farming — the bookmark is the product, not the information
The catch
The caption is a template, not a report. The identical wording has bounced around X since June with the runtime and even the engineer's gender changing between copies — proof nobody in the chain watched a talk. The real sources are free and better than the clip, but linking them would end the scroll, so no repost ever does.
How to actually do it
- Watch the real thing: search YouTube for Boris Cherny's Acquired podcast interview — the loops quote comes from that conversation, free, no course attached
- Read howborisusesclaudecode.com — a compilation of his actual published workflows, including recurring loops and dynamic workflows
- Read Anthropic's free engineering posts at anthropic.com/engineering: 'Effective harnesses for long-running agents' and 'Building agents with the Claude Agent SDK'
- Then build the minimum loop in Claude Code: (1) a CLAUDE.md memory file so runs don't start from zero, (2) subagents so one agent isn't doing everything, (3) an explicit stop condition — a loop without one runs until the bill stops it
Handed over the real sources plus the three-piece loop recipe the caption gestures at: memory file, subagent split, stop condition.
- Confidence
- High
- Posted by
- a tech-clip aggregator account, ~18k followers, nothing for sale
We test hype for free. We build the real thing for a living.
Thirty minutes, no pitch — and you'll leave with something useful either way.
Book a call with Todd or start with the free Business Checkup →The Verdict Weekly
Three verdicts every Friday. Free forever, unsubscribe anytime, no spam — that would be ironic.
© Schreier Group · schreiergroup.com · See a wild AI claim? Drop it here and we'll test it.