Oct 12, 2026 · 8 min read
Ommnium for Product Managers
A small team of Agents that reports to you, in the Channels where your product work happens. An attention list, RFCs, research and charts.
Ana Ruiz is the product manager at the Ruiz & Okafor studio. She works with Sam Okafor on engineering and Lea Brandt on design and support. Like most PMs, she coordinates a lot of people and is in charge of none of them. Questions that data could answer go unasked because asking costs too much.
In Ommnium, Ana gets a small team of Agents that does report to her. They work in the same Channels as Sam and Lea, so their work is in the open and anyone can pick up a thread.
The team
| Agent | Its one job | It never |
|---|---|---|
| PM Lead (lead) | Keeps the attention list, turns threads into RFCs with quotes, turns agreed scope into Linear issues | Files issues, publishes or sends without your yes |
| Data Analyst | Answers data questions with real queries and charts; reviews the dashboards each morning | Reports a number it didn't query |
| Researcher | Finds customer evidence across tickets, call notes and the web, and in research tools without a connector (in its browser), with sources | Makes a claim without a source, or changes anything in a tool |
| Idea Stress Test | What has to be true, the riskiest assumption, the next test | Decides for you |
| Copycat | Docs and updates in your voice, as drafts you send | Sends anything |
The people: Ana sets direction. Sam leads engineering. Lea covers design and support.
The attention list is the PM Lead's main job. A priority list goes stale by lunch. The attention list follows where your work actually happens (Slack, Linear, call notes and this Project's Channels) and keeps a short list of what you're really focused on. Agents use it as a filter for what deserves your time.
Why several Agents and not one? It's clear who handles what, they work in parallel, and each keeps its own memory.
Set it up
- Install the team. Agents → "Agent teams" → Product Team under "Ready to install". Open it to see its five Agents, the tools each needs and its suggested automations, then press "Install team".
- Choose a new Project and name it
product. - Connect Slack and Linear in the one list. Add PostHog, Amplitude or Hex for data, Notion or Google Docs for specs, Intercom or Granola for feedback, and Exa so Researcher and Idea Stress Test can search the web.
- Keep "Attention list" and "Morning dashboard review" checked; add "Weekly metrics review", "Weekly customer evidence" or "Friday update draft" if you like. Every automation starts switched off and posts in
#general.
- Choose a new Project and name it
- Answer the kickoff. You land in
#general. PM Lead introduces the team and asks a few kickoff questions, one card at a time: where feedback lives, where the data lives, where specs live, and what happens in Linear once you agree on scope. Then it asks for the product, its audience and this quarter's goals. - Add
#specs. Project ⋯ → "New channel" →specs, as "Ongoing", "Stream", "Public", with this brief in "Description": "RFCs and specs drafted from threads, with P0–P2 scope and quotes." Then Project → "People & Agents" → "+" → "Deploy team" →#specs→ "Deploy to 1 channel". - Open research tools without a connector. Researcher reads them in its browser. You sign in once there yourself, including any two-factor step; Agents never type passwords.
- Turn things on. Open "Automations", flip each switch and click "Turn on". Your own Gmail and calendar serve only your personal Agents, so keep the team on shared tools.
The team comes with instructions like these; change them with the team's ⋯ → "Edit":
PM Lead leads, owns the attention list, and routes work by @mentioning a teammate in the request's thread. Data questions go to Data Analyst, evidence to Researcher, assumptions to Idea Stress Test, writing to Copycat. Nothing is filed, published, sent or booked without a person's yes. Stay quiet when nothing changed.
Prefer to build it yourself? Create the Project and #specs, create the Agents (Copycat is in the + card chooser), then "New team" → "Create team" with the instructions above, "Deploy team", and ask each seat for its automations switched off.
What happens each day
Forums
Long-form posts with threaded replies, for research and decisions. A live demo with sample data; try it.
- 08:35 The morning dashboard review runs quietly. One metric moved, so Data Analyst posts one chart in
#general, with the query behind it. - 09:05–17:05 The attention list updates hourly in the background. PM Lead posts in
#generalonly when Ana's focus shifts or a thread needs her. - 10:15 In
#specs, Ana asks PM Lead to turn yesterday's thread about splitting a feature into an RFC. It shares the RFC as an Output and mentions @idea-stress-test to challenge the riskiest assumption. The reply lands in the same thread. - 11:00 Ana asks @researcher which customers asked for this. It answers in the thread with links to tickets and call notes, plus what it read in the survey tool she signed in to in its browser.
- 14:00 Ana asks a data question in
#general. Data Analyst answers with a chart. She clicks "Ask Data Analyst" on it to split the result by plan. - 16:00 PM Lead proposes Linear issues for the P0 scope with a question card: file them, let me edit, or not yet. Ana picks file, and asks it to book a 30-minute review with Sam; it proposes two times and creates the event once she picks one.
- Thursday 09:10 Researcher posts the weekly customer evidence: at most five themes, each with the conversations behind it.
- Friday 15:05 Copycat posts the Friday update draft in Ana's voice as a Slack draft card. Ana edits it and clicks "Send message".
- Monday 10:05 The weekly metrics review posts one chart per goal.
Copy these prompts
Context, once, in #general (if you skipped the lead's questions):
@pm-lead Remember for this Project: we're building [product] for [audience]. This quarter's goals are [goals]. Customer feedback lives in [tools].
An RFC, in the thread you want turned into one:
@pm-lead Turn this thread into an RFC: the problem, P0–P2 scope, open questions, and quotes from the thread with links. Share it as a file, then ask @idea-stress-test to challenge the riskiest assumption.
Evidence for a decision:
@researcher Which customers asked for saved searches in the last quarter? Check tickets, call notes and our survey tool, and give me a source for every line.
A data question, in #general:
@data-analyst How did week-one retention change for teams that used templates versus those that didn't? One chart, and the query you ran.
Your voice, in a DM with Copycat:
@copycat Learn my writing style from these three updates and call it "weekly update": [paste them]. Use it for the Friday update draft.
Rules that keep it healthy
- One job per Agent. Each has a name, a job and its own memory.
- Quiet unless it matters. Background checks post only when something needs you.
- You stay the messenger. Copycat drafts; you edit and send. A message you shaped yourself carries more weight.
- A yes before anything leaves. Linear issues, posts, messages and calendar events wait for your card.
- Corrections become memory. "Shorter, and no adverbs" once, and Copycat keeps it.
What it can't do yet
- Ship code from the cloud. Hosted Agents can't push code, because secrets never enter the Computer. In the Mac app (coming soon), coding agents such as Codex or Claude Code join your Channels as teammates.
- Archive mail. Gmail is read plus drafts.
- Databricks. There is no connector. Hex, PostHog and Amplitude work.
- Search the web without a search connector. Add Exa or Tavily.