How it works

From one meeting to your developer AI’s plan

Bring in the conversation, find what changed, make a new version from only what you pick, get your client’s sign-off, and let your developer’s AI pick it up — PLARO’s five steps.

Measured times
Comparison 27–35 s — a two-week chat sample, timed 4 times
Slowest AI draft 42 s — our own test

Five steps

  1. ① Import conversations

    Import conversations · meetings

    Drop or paste a chat export, a meeting transcript or a recording. PLARO recognises the format and keeps who spoke and when.

    Details →

  2. ② Find what changed

    Find what changed

    It compares the conversation with your current requirements line by line and returns what changed, what’s missing and what’s open, each with its evidence line.

    Details →

  3. ③ New version

    Requirements · versions

    Choose Apply, Edit and apply, or Ignore on each card and you have v2. Earlier versions and numbers stay as they were.

    Details →

  4. ④ Client sign-off

    Client sign-off · documents

    Send only the changed lines to your client and get [I confirm] without an account. The extra-work quote and change agreement come from here too.

    Details →

  5. ⑤ Developer AI

    Developer AI connection

    Confirmed changes go to your developer’s Claude Code or Codex. When they talk to it, it is first told “v1 → v2” and plans the changes.

    Details →

What PLARO does

  • Finds what changed, what’s missing and what’s open in conversations, with evidence lines
  • Makes a new version from only what you pick, and logs every changed line
  • Client confirmation records · record fingerprints · extra-work quotes · change agreements · acceptance records
  • Gives your developer’s AI read-only access to the confirmed requirements

What it doesn’t do

  • Meeting-minute summaries
  • Contracts · e-signatures · tax invoices
  • Task management (assignees · progress)
  • Sending anything to clients or developers on its own
  • Letting the AI fill in amounts

Testing

Our own test — method and limits

We measure how right the AI’s findings are with fictional meetings we wrote ourselves. The answers are fixed before the runs.

Meeting comparison

4 fictional meetings (online store 45 min · hair salon 40 min · marketing site in English 40 min · a tricky meeting with reversed decisions 75 min) · 31 answers · 23 traps · 3 runs — 98.8% found (lowest meeting 86%) · 0 wrong findings · slowest comparison 47 s

October 8, 2026 · the model in use now

Chat comparison time

A two-week chat sample with small talk mixed in (5 real changes · 1 open question), compared 4 times — all 6 found · 27–35 s · 2–4 unneeded cards each time

October 10, 2026

Requirement drafts

2 drafts × 3 runs — 100% of the document’s facts · slowest draft 42 s

October 8, 2026 · the model in use now

We wrote these meetings and answers ourselves, so results in real meetings may differ. That is why findings are always drafts, and a person decides what goes in.