How the best product teams work
The workspace that knows your product.Generic AI starts from zero. You don’t.
Every doc, meeting, and prototype feeds one context layer, so every output starts from everything your team knows and gets sharper the longer you work.
The morning brief · your product day, already loaded
One context layer · every room
Your tools store the docs. None of them remember why.
The context layer
Generic AI knows nothing about your product.Inversion accumulates it.
Every document, meeting, decision, prototype, and persona feeds one context layer. Every output draws from it — grounded in your users, your constraints, your calls. The work gets sharper the longer you work.
Month one · month two · month three
Your context layer
One compounding memory of what you know, decided, and shipped
The operating system for product
Every room of product work.All of it reading from your context.
Room 01 · Documents
Docs grounded in everything you know.Cited from the ledger.
PRDs, specs, and strategy docs drafted against your context layer — your users, your metrics, your settled calls pinned in the margin. Not a blank page with a prompt box.
A real doc · decisions cited inline, live in the margin
Room 02 · Prototypes & feedback
Clickable by tonight.Feedback at scale by morning.
Describe the flow and get a working prototype, grounded in your product's patterns. Then ask for feedback with a link — users answer like a form. Reactions and comments at scale, without begging for a call.
The prototype studio · present, collect, iterate
Feedback · 47 overnight
Shared as a link · answered like a form
Room 03 · Personas
Talk to your users before you book them.
Synthetic personas built from your real research — interviews, surveys, support threads. Pressure-test a flow at 11pm and walk into the real call already knowing where it bends.
“If batch invoicing means no more exporting 200 receipts a month, I switch tomorrow.”
Built from 6 interviews · 2 surveys
“I don't read docs. If the first run isn't my actual problem, I'm gone.”
Built from 9 support threads
Ask them anything · they answer from your research, and cite it
Living profiles · built from your research, cited back to it
Room 04 · Meetings & routines
Meetings in. Your day, loaded.The recurring work runs itself.
Transcripts come in; decisions, owners, and open questions come out while you're still in the room. The morning brief loads your day from every source, and routines turn the work you repeat into scheduled output — every external send waiting on your approval.
Routines · set it once, run it forever
Roadmap review · live
Extracted while the meeting runs
Room 05 · Competitor intel
Rivals, tracked against your own calls.
Name a target and Oracle watches it every week, filing what actually changed, cited to the source, against the decisions those changes threaten. A rival shipping something never ambushes a settled call.
Move detected · Dec 4
Atlas shipped flat seat pricing
Their pricing page dropped usage tiers entirely. Positioning now leads with “predictable bills.”
Threatens a settled decision
Every change cites a URL
Room 06 · Library
Install how other PMs work.
Templates, prompts, skills, and routines published by working product people. Install theirs in one click; publish yours back when they're proven.
Prompts, skills, workflows · published by working PMs
The library · published by working PMs
Room 07 · The org layer
Your private repo.The org's brain. A review between.
Everything you build lands in your private brain first. Promoting it to the team is a pull-request-style review — a diff, approvers, a merge — so the shared context stays trustworthy because someone said yes.
The private brain · draft → review → live
Proposal · Pricing FAQ v2
Room 08 · Oracle
An agent over all of it.With an approval gate.
Oracle plans, searches, drafts, and proposes — across your docs, ledger, GitHub, and the web — then waits at the gate. Nothing touches your ledger or your tools without your sign-off.
▸ Three reviews due this week. Pulling the numbers.
read analytics · activation, margin, beta NPS
read github · 4 PRs merged since Monday
Oracle proposes a decision
Sunset the weekly digest experiment
41% opens, 1.9% resurrection across 8 weeks. The habit has to form in the first session.
nothing lands without your sign-off
Oracle · nothing lands without sign-off
Every tool in your stack is a system of documentation.
Inversion is your system of judgment.
Decisions fade. The reasons fade faster.Inversion keeps both — and checks what happened.
The context layer’s spine: decisions extracted from your meetings and docs, cited in everything new, tied to the commits that shipped them, and brought back on review day with the number. Settled as Worked, Didn’t work, or Mixed.
The ledger · live · what you decided, and what happened
Decision detected
Cap the beta at 25 seats
Pricing sync · 41:22 · review in 3 days
The flagship · In the editor
Your next PRD cites the ledger.Contradictions get flagged before they ship.
Pulling from your context
PRD · Billing · Draft
Payments PRD, Q3
Billing friction is our second-highest churn reason at 21%. We price in EUR cents so spend stays legibleEUR-cents billing and we hold the 44% activation floorproblem-first first run while the rewrite lands.
For launch we bring back the template gallery as the default first run, giving new users breadth to browse before they commit to a real problem.
Contradicts a settled decision
First run generates a real PRD from the user’s own backlog item
WorkedSettled Feb 12 · Activation 31% → 44% in four weeks
Fig. 01 · The refusal flag, firing — nothing ships silently
The flagship · Review day
On review day, the number comes back.And the call gets settled.
You set the expectation when you decided. Inversion returns on the date you named with the metric, the receipts that shipped it, and one question: did it work? The answer compounds into the context layer.
Review due today · set when you decided
Cap the beta at 25 seats
Inversion came back with the number
Review day · expectation, number, receipts, verdict
Day One Pack · what the context layer buys you
A new PM, productive on day one.
Six months of judgment, readable in one sitting: every decision with its outcome, the receipts that shipped it, the reviews still open, and who to ask about what.
- 14 decisions, with outcomes
- 3 open reviews and their dates
- What shipped, with the receipts
- Your first 30 days, oriented
Generated for Ana · Tuesday, 9:02
Payments · the story so far
Six months
EUR-cents billing
PR #214WORKEDPayments deferred to Q3
MIXEDWeekly re-engagement digest
DIDN'T WORKBeta capped at 25 seats
Review MonOPENRead first
Billing migration notes · Q1 retro · Webhook patterns
The past, readable · generated from the ledger
The advantage compounds
Stop starting from zero.Your context is the moat.
Every doc, meeting, and decision you add makes every output sharper. The teams who start now will be impossible to catch.
Free to start · no credit card