Merit

The problem with job search tools

Every tool in this category is one of two things: a job board with a chatbot bolted on, or an auto apply cannon that sprays a generic CV at everything. Neither survives contact with how hiring actually works, because the bottleneck was never finding postings. It is tailoring each application well enough to matter, three hundred times, without burning out.

Merit is a self initiated concept built as the third thing: an agent that does the reading and the drafting, and hands every judgement back to you. It reads your CV once into evidence lines you can edit, runs nightly against the market, drafts only where your evidence is strongest, and never sends anything without approval.

Designing what the agent may claim

The hard part was not the matching. It was deciding what the product is allowed to say. A match score with no reasoning is a horoscope, so Merit never shows one. Match strength is a band, and every claim behind it is split into what the agent is sure of, each with the CV line it came from, and what it is guessing, marked as inferred with its source one tap away. The one real gap in your evidence is named in the same list, because a tool that hides your weaknesses is selling you something.

That principle set the interface language. Evidence is quoted, never summarised. Limits are visible and priced: raise your comp floor to 95k and the live match count drops from 14 to 8 before you commit. And the weekly cap is a feature, not a restriction, because quality collapses at volume and the product says so.

Restraint as the feature

The agent has three levels of agency, and the default is the middle one: it drafts, you approve. Turning on autonomous sending takes a consent dialog that states its own boundaries. Everything the agent does lands in a ledger where each action carries an undo, and everything it sends holds a live 30 minute revert. Memory is plain sentences, learned from your decisions, deletable one by one.

None of that is decoration. Agent governance is the design problem of the moment, and Merit’s answer is that trust is built from receipts, not from a friendly tone.

Why it was animated

An agent product is judged on its handovers: the moment a draft becomes a send, the receipt that follows, the undo counting down. Static frames cannot show whether those moments feel considered or abrupt.

So all eleven flows above were built in code rather than prototyped in a design tool, with a virtual clock so every frame is deterministic, and rendered at 60fps. Sheets and keyboards move on real spring physics, screens hand over through shared elements instead of slides, and the camera holds still at the start and end of each shot with a single slow move between, so the interface carries the story. Every figure reconciles across the whole series: the 23 evidence lines, the 3 of 5 weekly cap, the 8 of 9 requirements evidenced, and the overnight run that read 212 postings to draft 3.