What the agent assumed
You gave a coding agent a vague brief. Everywhere you said nothing, it made a silent decision — a currency, a locale, a destructive shortcut. Flag the ones you'd want to review, then see which risky defaults you caught and which you'd have shipped. Runs entirely in your browser (0 uploads, works offline).
How to use it
- Read each silent decision the agent made on the under-specified brief.
- Flag the ones you'd want to review — leave the harmless defaults alone.
- Audit it: your recall (risky defaults caught) and precision (review budget not wasted on safe ones).
What this clears up (the fundamentals)
- A vague brief produces invisible decisions, not fewer — the agent must fill every gap, and it does so silently.
- Not all defaults are equal — English copy is a safe gap-fill; assuming USD for an India-first product is a wrong-market default.
- Destructive and security defaults are the danger — instant delete with no confirm, or a fully public prototype, are the ones to catch.
- The skill is auditing, not spec-writing — you'll never specify everything, so review the gap-fills with real blast radius before they ship.
Where it's used
A product and AI-collaboration drill — the fastest way to build the habit of auditing what an agent assumed rather than trusting it filled the gaps sensibly. It's a miatz build-lab concept playable; learn to build it yourself.
FAQ
Why does the agent make assumptions at all?
Because it can't proceed on a silent point without choosing something. A brief that doesn't specify a currency, locale, or confirmation step forces the agent to pick a default to produce anything.
Which assumptions matter?
The ones with real blast radius: wrong market (currency, date format), data loss (no delete confirm), or security (no auth). Aesthetic defaults like a brand colour are cheap to change and low risk.
How do I catch them in practice?
Treat every silent gap-fill as a decision to review. Ask the agent to list its assumptions, and check the ones that touch money, data, or access before shipping.
Is anything uploaded?
No. The assumptions and audit run entirely in your browser — nothing is transmitted or stored. Turn off your Wi-Fi and it still works.
Limits
A teaching model with a fixed set of illustrative assumptions — real agents make many more, context-dependent. The auditing reflex transfers directly.
Related
Part of the miatz playgrounds. Explore the rest from the playground catalog.
Bookmark this page (Ctrl+D, or ⌘D on Mac) — it works offline the next time you need it.