Skip to main content
Product

A research run that ends in a backlog.

Jary is not a chat window with a product-manager persona. It's 11 defined stages that go out and read your market, argue for one focus, try to break it, and hand what survives to your repo.

01

Evidence, not recall

Nothing comes from a model's memory of the web. Every competitor, feature, and complaint is fetched during your run and attached to the stage that used it.

02

Citations you can open

Each claim carries the link behind it — a changelog entry, a pricing page, a support thread. If you disagree with a conclusion, you can go read what it was built on.

03

It argues both sides

The coach commits to a focus and then makes the strongest case against it, using the contrary evidence from the same run. A focus that can't survive that isn't ready.

04

You hold the gate

Research runs unattended. Anything that rewrites scope — splitting a feature, deferring one, changing acceptance criteria — waits for you to approve it.

The gap

Most ideas die between the doc and the repo.

Strategy lives in Notion. Tickets live in Linear. Code lives in GitHub. The translation between them is done by whoever has the smallest queue.

AI assistants suggest sentences. Ask one whether your competitor already shipped the thing you're scoping and it will guess, confidently, from a year-old snapshot of the web.

Jary goes and looks. Traffic, feature sets, shipped changelogs, complaint threads — pulled at run time, attached to the stage that used them, and cited in every claim it makes.

The run

Every stage, and what it comes back with.

Each one leaves evidence behind and unlocks the next. Rerun any single stage without restarting the run.

Group 01Define

What are we building, and for whom?

Runs on its own
01

Frame the idea

Turns a loose sentence into a target user, a job to be done, and one hard constraint. Everything downstream is measured against this, so a vague frame fails here instead of in stage nine.

  • your prompt
  • follow-up questions

Group 02Research

Who else is solving this, and where are they weak?

Runs on its own
02

Find competitors

Names pulled from the language people actually use for the problem — search queries, community threads, and app directories — so the list isn't just the two or three brands you already had in mind.

  • search queries
  • directories
  • community threads
03

Size the competition

Traffic and rank evidence separates real incumbents from names on a list. Nine competitors becomes three that matter and six you can stop worrying about.

  • traffic estimates
  • category rank
  • funding signals
04

Extract competitor features

Feature sets pulled from live product surfaces and compressed into the table stakes an MVP has to match before anyone will switch to it.

  • product pages
  • docs
  • pricing tiers
05

Read their trajectory

Changelogs and release notes say where incumbents are actually heading — which gaps they're closing this quarter, and which they've quietly walked away from.

  • changelogs
  • release notes
  • careers pages
06

Hear the complaints

Customer complaints grouped into recurring themes, ranked by how often they surface. The ones an MVP must solve to be adoptable, not the loudest one.

  • review sites
  • Hacker News
  • support threads

Group 03Decide

Which gap do we commit to, and does it survive challenge?

Stops for you
07

Gap analysis

Cross-references what incumbents ship, where they're heading, and what customers complain about — then names the openings nobody is serving, each with the evidence behind it.

  • stages 02–06
08

Pick the focus

Commits to a single focus and says how sure it is. You get the recommendation, the reasoning, and the evidence it leaned on — and you can argue with any of it.

  • stage 07
  • your input
09

Pressure-test

Turns around and makes the best case against the focus it just picked, built from the contrary evidence in your own run. If the focus survives, it earns a backlog.

  • contrary evidence

Group 04Build

Turn the decision into backlog you can ship.

Stops for you
10

Shape the backlog

Turns the surviving focus into features with acceptance criteria written before code is — sequenced by what blocks what, not by what came up first.

  • stages 07–09
11

Commit the build

Each shaped feature maps to a GitHub issue — or imports the issue you already have. When the PR that references it merges, the feature advances to shipped on its own.

  • your repo

11 stages · 4 groups · one session

Group 03 · Decide

It argues for a focus, then argues against it.

Gap analysis names the strongest unserved opening and the evidence behind it. Pick the focus commits to one. Then pressure-test turns around and makes the best case that the decision is wrong — using the contrary evidence from your own run.

If the focus survives that, it earns a backlog. If it doesn't, you found out before the sprint.

Stage 08 · Focus recommendation

Confidence

Multi-channel inbox for sellers under 200 orders/month.

Three of nine competitors are moving toward checkout and away from messaging. The complaint themes cluster on channel-switching, not catalog. Zendesk's per-resolution pricing prices out this exact band — 41 complaints reference cost at low volume.

Strongest case against

Bikko shipped WhatsApp Catalog sync six weeks ago and is the only competitor still investing in messaging. If they ship an inbox next, the wedge closes. What would have to be true: their hiring stays on messaging, not payments.

Group 04 · Build

When the PR merges, the spec closes itself.

Link a repo and each shaped feature maps to a GitHub issue — or import the issues you already have. Your backlog and your code stop drifting apart.

Jary · shaped feature

Unified inbox — WhatsApp + Instagram

  • Messages from both channels in one thread
  • Reply without leaving the thread
  • Unread count survives a refresh

GitHub · issue #142

Unified inbox — WhatsApp + Instagram

  • Acceptance criteria synced
  • PR #98 references #142
  • Merged 2d ago

PR #98 mergedissue #142 closedfeature advanced to Shipped

No ticket touched by hand

Bring an idea.
Leave with evidence.

Eleven stages, one session, no credit card. Watch it find your competitors before you finish your coffee.

Start with an ideaWatch a run · 2 min