∴ calibrating signal

Skip to content
∴ TRAZE

∴ work — 02 · Traze — self-initiated

Signal Scan — an agent that scopes AI projects

self-initiatedagentlead-genin-production

5

questions to a ranked opportunity map

<60s

from first answer to actionable output

24h

human follow-up promise behind every scan

The challenge

Every AI agency site says 'book a call'. Almost none shows AI doing anything. We wanted the first interaction with Traze to be a working demonstration: useful output before any commitment, sales qualification as a by-product.

The approach

A deliberately bounded agent: structured intake (industry, team size, where hours leak, current stack, goal), a constrained reasoning pass, and a typed output schema — three opportunities ranked by impact and effort, each mapped to an engagement shape. No open-ended chat, no hallucinated promises; when the model is unavailable it degrades to an honest hand-run fallback.

Bounded beats chatty

The scan asks exactly five things and returns exactly one artifact. Constraint is what makes agent output dependable enough to put your name on — the same principle we apply to every client agent.

The demo is the qualification

By the time someone books a call, we know their industry, their bottleneck and their appetite — and they know how we think. Both sides skip a discovery meeting.

Working in public

The scan runs on this site with its methodology documented in the open. Usage numbers join this page as they accrue — real ones, when they exist, not projections.

Stack

Vercel AI SDK

AI Gateway

Zod schemas

Edge-rate-limited API