← The AI Hype Audit — all 576 verdicts
PARTLY
Partly true: asking Claude Code or Codex to find the implicit state machines in your app is a genuinely good refactor idea, but 'always, without breaking anything' is a wish, not an instruction
The claimVibecoders, open your messiest project and paste one prompt into Claude Code or Codex: find all of the implicit state machines in my app and make the best decision that elevates user experience, developer experience and agent experience, always, without breaking anything.
The idea underneath is sound and old. Messy apps often track one thing with several loose flags (isLoading, hasError, isOpen, step = 3) that can land in combinations that should be impossible, like loading and errored at once. Turning those into one explicit state with named transitions removes a whole class of bugs, and coding agents are good at spotting the pattern. 'Agent experience' (AX) is also a real term, coined by Netlify's CEO in January 2025 for how easily AI agents can use a product. The prompt itself is the weak part. It asks for every state machine in the whole app at once, gives the agent permission to decide the fix on its own, and ends with 'without breaking anything,' which no prompt can guarantee. In a messy project with few tests, that is a recipe for a large, hard-to-review diff that changes behavior quietly. Run it as an inventory first, then fix one machine at a time behind tests.
What holds up
- Implicit state as scattered boolean flags is a well-known bug source; explicit states fix it
- Claude Code and Codex can find and refactor this pattern when the scope is small
- 'Agent experience (AX)' is a real term, introduced by Netlify's CEO in January 2025
- Explicit states also make code easier for the next agent or developer to reason about
What doesn’t
- 'Without breaking anything' is not something a prompt can enforce; only tests can
- 'Find all' across a messy app invites a sprawling diff nobody will review properly
- 'Make the best decision' hands design choices to the agent with no review step
- Not every flag is a state machine; over-applying it adds ceremony, not clarity
- UX, DX and AX together is a slogan; the prompt gives the agent no way to measure any of them
The catch
The prompt promises safety it can't deliver. 'Without breaking anything' only means something if tests exist to catch the break, and the 'messiest project' this post targets is exactly the one that doesn't have them.
How to actually do it
- Commit or branch first, so every change can be undone: git checkout -b explicit-state
- Step 1, read only. Paste into Claude Code or Codex: 'List the implicit state machines in this app: places where several flags or variables together describe one thing's state. For each, give the file, the flags involved, the impossible combinations they allow, and a risk rating. Do not change any code.'
- Pick the single highest-risk item from that list. Ignore the rest for now.
- Step 2: 'Before refactoring [item], write tests that pin down its current behavior, including the edge cases from your list. Run them and show me they pass.'
- Step 3: 'Refactor [item] into one explicit state (a union type or reducer with named transitions). Keep behavior identical. Run the tests. Show me the diff.' Review the diff yourself before merging.
- Repeat steps 2 and 3 for the next item. One state machine per pull request.
We rewrote it as a safe three-step version you can paste today: inventory first, then tests, then one refactor at a time.
- Effort
- 15 min for the inventory; 30-90 min per state machine refactored safely
- Cost
- $0 beyond the Claude Code or Codex plan you already have
- Stack
- Claude Code or Codex, git, your project's test runner
- Confidence
- Medium
- Posted by
- a vibe-coding tips post shared as plain text, no link; the source link is the Agent Experience definition we checked
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.