What a Room actually is, and why it isn't a Slack thread
A Room is a scoped decision space with weighted evidence, open objections, and a recorded outcome. Here's what makes it different from a chat thread, and when a thread is fine.
Every company already has a place where revenue problems get discussed. It's a Slack channel, and it works surprisingly well right up to the moment it doesn't.
The failure isn't that Slack is bad. It's that a chat thread has three properties that are wrong for decisions: it never closes, it loses objections, and it forgets.
A Room is the same conversation with those three properties fixed.
Key takeaways
- A Room is scoped to one problem and it closes. Resolved or explicitly no-action-needed — those are the only exits.
- Evidence carries a weight label. Causal finding, strong association, hypothesis, or insufficient evidence. Not all claims in a discussion are equal and the format shouldn't pretend otherwise.
- Objections stay open until answered. In a thread, disagreement scrolls away and silence reads as consent.
- Customer-facing actions live as Plays that require approval. An agent can propose; a human decides.
- Closed Rooms become Business Memory. The outcome is stored with its holdout measurement, not with a narrative.
The three failures of the thread
It never closes
A Slack thread has no terminal state. It gets quiet. Quiet is ambiguous — it could mean resolved, deprioritized, blocked on someone, or forgotten entirely, and there's no way to tell from the outside.
Six weeks later someone asks "whatever happened with the repeat rate thing," and the honest answer is that nobody knows. The discussion happened. Whether a decision came out of it is unrecoverable.
A Room has exactly two exits: resolved, or no action needed. Both are affirmative states someone chose. There's no drift.
It loses objections
This is the failure I care most about.
In a thread, someone raises a serious concern at 4pm on a Thursday. Three other messages land on top of it. By Friday morning the conversation has moved on, and the objection was never answered — it was buried.
Nobody dismissed it. The medium dismissed it, by being linear and having no concept of an unresolved item.
In a Room, an objection is an object with a state. It stays visibly open until someone answers it or explicitly accepts the risk. You cannot close a Room with open objections without acknowledging them. That single constraint changes the quality of decisions more than any other feature we've built, because it makes the cost of ignoring dissent visible rather than free.
Strong association in the Rooms we see, the ones with at least one recorded and answered objection tend to produce decisions that hold up better over the following quarter. We label this strong association because we can't randomize whether teams disagree, and teams that argue may differ in other ways.
It forgets
Threads are searchable in theory. In practice, finding the thread where you decided something eighteen months ago requires remembering enough about it to construct the query — and if you remembered that much, you wouldn't need to search.
More importantly, a thread stores the conversation, not the conclusion. Even when you find it, you get forty messages and have to reconstruct what was decided and whether it worked.
A closed Room stores a decision document: what the problem was, what the evidence said, what was decided, what happened, and how it was measured. That's what goes into Business Memory, and it's what an agent reads before investigating a similar problem next year.
What's inside a Room
Four components.
Evidence, weighted. Every item carries a label — causal finding, strong association, hypothesis, insufficient evidence. This matters because in a chat thread a well-argued guess and a controlled measurement look identical, and the more confident writer usually wins. Weighting evidence separates the strength of a claim from the confidence of the person making it.
A thread, with humans and agents. Agents participate as first-class members. They post evidence, respond to questions, and flag when a claim contradicts something in Business Memory. They're identified as agents; nobody is pretending otherwise.
A decision document. Drafted by agents, edited by humans. It records why, what the expected impact is, what happens next, what the guardrails are, and who owns each action. It's the artifact that outlives the Room.
Plays. Proposed actions — a campaign, a retry-window change, a discount suppression, a product fix. Every Play that touches a customer or moves money requires a recorded human approval before execution. Agents draft them; agents never fire them.
The scoping rule
A Room is scoped to one revenue problem or one cohort. Not a project. Not a quarter. Not a channel.
This is a constraint we enforce, and it's occasionally annoying. If a Room's scope creeps — if the discussion about second-order decay turns into a general conversation about the checkout experience — the right move is to close it and open a new one, and Flolyt will nudge you toward that.
The reason is that scope creep is how a Room turns back into a Slack channel. A space that discusses everything closes on nothing, and the artifact it produces is a summary rather than a decision.
When a thread is the right tool
I want to be clear that we're not proposing you conduct your working life in Rooms.
Fast, low-stakes coordination. "Can someone check whether the export finished?" That's a thread. Putting it in a Room is ceremony.
Ongoing team communication. Standups, announcements, general awareness. Rooms are terrible at this because they close.
Exploration before a problem is defined. Sometimes you're poking at something and don't know if it's real. Do that in a thread. Open a Room when you have a scoped problem.
The test is whether there's a decision with a consequence and an owner. If yes, and if it involves more than one team, it belongs in a Room. If not, Slack is genuinely the better tool and we'd rather you used it.
What this looks like end to end
A concrete sequence:
- The Repeat & Decay agent detects a step change in cohort second-order rate and opens a Room. It attaches the cohort curve, the deploy timeline for the window, and an initial hypothesis labelled as such.
- Marketing and Product are pulled in automatically, because Retain is Marketing's stage and the candidate cause is a deploy.
- An engineer notes the deploy was behind a 50% flag. That converts the analysis: now there's a comparison group, and the evidence weight moves from hypothesis to causal finding.
- Someone from Finance raises an objection — the same period included a shipping-rate change that might confound it. The objection stays open.
- The agent checks: the shipping change applied to all customers, the deploy to 50%. The flag split isolates the deploy. Objection answered, and the answer is recorded.
- A Play is proposed: revert the checkout change. Product approves. It executes.
- The Room stays open through measurement — post-fix cohorts against pre-fix cohorts at matched maturity.
- The Room closes. The decision document, the measurement, and the answered objection go into Business Memory.
Fourteen months later, a similar signal appears. The agent investigating it reads that memory first and opens with the prior finding rather than starting from zero.
Frequently asked questions
How is this different from a Jira ticket or a Linear issue? Those track work to be done. A Room hosts a diagnosis and a decision, and it holds evidence with weights and open objections. A Room frequently produces tickets — but the ticket is the output, not the container.
Can I use Rooms without the AI agents? Yes. Humans can open Rooms directly. Agents accelerate detection and evidence assembly; the structure works without them, just slower.
What stops a Room from staying open forever? Rooms have owners and expected resolution windows, and stale Rooms surface for a decision. The intended answer to a Room nobody is working is "no action needed," which is a legitimate outcome — it's the ambiguity we're eliminating, not the possibility of deciding to do nothing.
Do agents have the same permissions as people in a Room? No. Agents can open Rooms, post evidence, and draft Plays. They cannot approve a Play, cannot send anything to a customer, and cannot close a Room. Those are human actions and the audit log records which human took them.
What if my team already coordinates well in Slack? Then keep doing that for coordination. The question is whether you can answer, for a decision made eight months ago, what the evidence was and whether it worked. If you can, you don't need this.
Start free at Flolyt
Connect one source and see your first leak with a number attached. Free under $500K in revenue, priced per company after that — never per seat.