← The AI Hype Audit — all 100 verdicts
PARTLY "The 4 Levels of Agentic OS 2.0 — Claude Code that runs your work" — a real workflow idea dressed in fabricated dashboard numbers
The claimThe 4 Levels of Agentic OS 2.0 — turn Claude Code from a CLI into a system that runs your work: Skills (workflows that loop and self-improve, 'Claude stops waiting for prompts'), Memory (an Obsidian second brain, cheaper runs), Interface (a dashboard, one click runs skills headless or by voice), Handoff (package the whole OS and share the repo — anyone runs it, no terminal).
The underlying idea is real and good — Claude Code plus skills, a memory vault, and reusable automations genuinely is a step up from typing one prompt at a time, and we run a version of exactly this. Two things pull it down to partly. First, the slides display fabricated dashboard metrics (numbers like '9,875,744' with no source) as if they were live results — that's set dressing, not proof. Second, 'workflows that loop so they self-improve' and 'Claude stops waiting for prompts' oversell autonomy: today's agents run the steps you define, they don't reliably self-improve unattended, and unattended loops are exactly where token bills and runaway behavior come from. The concept ships; the sci-fi autonomy and the invented dashboards don't.
What holds up
- The core stack is real: Claude Code + skills + an Obsidian memory vault + reusable automations is a legitimate, powerful setup
- 'Package it and hand off the repo' is genuinely how these systems get shared with a team
- We run a production version of this exact pattern — the direction is sound
What doesn't
- The slides show fabricated dashboard metrics with no source — decorative numbers presented as results
- 'Workflows that loop so they self-improve' oversells — agents run defined steps; reliable unattended self-improvement isn't here yet
- Unattended loops are where token burn and runaway behavior live — 'Claude stops waiting for prompts' is the risky part, not the feature
- Funnels to a YouTube channel + link-in-bio for 'the full builds'
The catch
The four-level framing is a decent mental model wrapped around invented proof and overstated autonomy. Build the levels — they're real — but verify every automated run, and don't believe the dashboard screenshots any more than you'd believe a stock photo.
How to actually do it
- Level 1 (Skills): codify 2–3 tasks you actually repeat into Claude Code skills — start narrow
- Level 2 (Memory): point Claude at an Obsidian vault with a CLAUDE.md that says how to file and recall notes — this part genuinely compounds
- Level 3 (Interface): a simple script or command to run a skill beats a fancy dashboard; add voice later if ever
- Level 4 (Handoff): document the setup so a teammate can run it — but keep a human approving anything that writes, sends, or spends
- Never let a workflow 'loop to self-improve' unattended — that's the sentence that runs up a bill or breaks something quietly
Build the real version of the four levels, with a human on the loop.
- Effort
- A weekend to a real Levels 1–2 setup; skip the dashboard until you feel the need
- Cost
- Claude plan ($20–200/mo depending on how hard you run it) + free Obsidian
- Stack
- Claude Code + Obsidian + a few verified skills + you on the approve button
- Confidence
- Medium
- Posted by
- a Claude-Code creator funneling to a YouTube + link-in-bio channel
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.