← The AI Hype Audit — all 181 verdicts
PARTLY “This is how you build an AI Agent with Claude” — that's a Project, and a Project is not an agent
The claimHow to build an AI Agent with Claude: 1) create a new Project with a title and description, 2) add a system prompt with instructions, 3) give tool access so your agent can use tools in its loops, 4) get your agent to work by prompting it. Pro tip: run your agents on Talk To Me Data and never worry about tokens, context windows, or fixing your agent.
The steps are a real, accurate description of setting up a Claude Project: persistent custom instructions applied to every chat, project knowledge, and — via connectors and MCP — genuine tool access Claude can use in multi-step loops within a conversation. Following them gets you a useful, specialized assistant. Calling it an “AI Agent” is terminology inflation. A Project is a workspace: it runs only when a human prompts it, has no autonomous execution loop, no schedule, no triggers, no ability to work between conversations, and no deployment. Step 4 — “get your agent to work by prompting it” — quietly concedes the whole case: an employee you must poke awake per task is an assistant. The closer is also just an ad: the video funnels to a paid hosting product with promises (“never worry about tokens, context windows or fixing your agent”) no vendor can actually keep.
What holds up
- The Project setup steps are accurate — custom instructions, knowledge, and connector tool access all work as described
- Claude genuinely does multi-step tool-use loops inside a response
- A Project with a sharp system prompt is one of the highest-leverage five-minute setups in Claude
What doesn't
- A Project runs only when prompted — no autonomy, no schedule, no triggers, no working between conversations. That's an assistant, not an agent
- “Get your agent to work by prompting it” concedes the inflation in the post's own words
- The “PRO TIP” is a vendor pitch, and “never worry about tokens, context windows or fixing your agent” is a promise no host can keep
The catch
The word “agent” is carrying the whole video — rename it “a really good specialized assistant” and both the wow and the hosting upsell evaporate.
How to actually do it
- Do the setup — it's real: a Project with a sharp system prompt and the connectors you actually use is a genuine upgrade
- Call it what it is, and you'll use it correctly: an assistant you direct per task
- If you want an actual agent — runs on its own, on a schedule, with tools — that's Claude Code scheduled agents or the Agent SDK: a real build, not a Project setting
Follow the steps, keep the correct noun, skip the hosting pitch.
- Confidence
- High
- Posted by
- an AI-sales content account whose “pro tip” is a paid hosting vendor's pitch
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.