Senate in Action
A complex decision. Eight stakeholders. No meeting needed.
I want to show you something that happened this week. A real decision, a real group of people, a real disagreement — and a Senate that worked better than the meeting it replaced. I can't share the topic. I can show you everything else.
The Setup
A complex technical decision. Eight stakeholders. Multiple organizations. Real money on the line, real security risk, real disagreement about the right path.
The calendar invite existed. The meeting was scheduled. An hour with eight people — some in different time zones, some with competing interests, none of them having read the same documents.
The meeting would have produced: a summary of confusion, three action items nobody owned, and a follow-up meeting.
Instead, someone said: let's do a Senate.
What Is a Senate
A Senate is a structured async decision chain. One question. A fixed routing list. Each participant adds their position and forwards the full chain — every prior argument intact — to the next person. The decision-maker receives everything and calls it.
Named votes. On the record. Dissent preserved.
- You read what came before you
- You add your argument — for, against, or with conditions
- You forward the complete chain
- You do not schedule a meeting
Each step takes approximately 30 minutes with AI assistance.
What breaks a Senate: The first response to "no meetings, write your position" was a meeting invite. That's the failure mode. The format only works if the discipline holds — no position, no forward. The moment someone substitutes a calendar event for a written argument, the chain stops and the thinking evaporates. You can see it happening in real time, because the token doesn't move.
Why a Senate is serial — and why the order is a craft choice
Senates run in sequence, one participant at a time, and the order of participants is not arbitrary. Both of those facts are doing work that's invisible from inside a single round, so they need to be stated plainly.
Why serial
Serial protects the participant from interruption and forces them to make their point fully. In a meeting, you get fifteen seconds before someone interjects, your argument gets truncated mid-construction, and you finish with the version you could fit into the gap, not the version you actually had. Serial removes the gap. You write until your argument is complete. Nobody talks over you. The token moves only when you're done.
The cost of an interruption-free turn is that you have to make your case well, because nobody is going to bail you out mid-thought. The benefit is that the case in the chain is the full case, which is what the next participant deserves to read.
Counter-arguments are still allowed and frequently necessary. But if you find yourself needing to counter a previous round, that's evidence the previous round didn't make its point as well as it should have. A tight argument leaves the next participant adding nuance, not pushing back. Counters are signals — they fire when an earlier link wasn't tight enough. Track that. Tighten the next round you write.
Routing order — the craft choice the organizer makes
Most meetings get routed by availability or by org chart. A Senate gets routed by what produces the best thinking. The organizer makes the routing call deliberately, before the token goes out. The structure of the routing list is a structural argument about how the disagreement should unfold.
- Pressing party first. Whoever is pressing for the decision — or for a side, if sides exist — leads. Their argument gets to be the full proposal, uninterrupted. The pressing party is usually the proposer, but not always. Sometimes the proposer is a neutral organizer between two pressing parties; in that case the pressing party leads, not the organizer.
- Counter-position next. Someone with a different idea responds early enough that the chain develops the disagreement before peripheral participants are asked to weigh in. Without this, peripheral folks end up reacting to the proposer's frame without seeing there was a counter at all. The disagreement gets buried; the chain becomes a chorus of agreement-with-the-proposer; the decision-maker reads a misleading consensus.
- Senior leads, never sandbags. Senior participants — the ones with the most context, the longest history, the deepest implementation knowledge — go early in the chain. Not late. Standard meeting-room dynamics put seniors last because their voice has the most weight, but the Senate format reverses this on purpose. With seniors leading, juniors receive the experienced framing in writing, available to read at their own pace, before they're asked to weigh in. They can build on it. They can't be ambushed by it. Putting seniors last produces ambush; putting seniors first produces apprenticeship.
- Roles interleaved when sides exist. If there are sides — product vs engineering, security vs feature, central vs regional — participants from each role alternate. Otherwise the chain becomes a one-side block followed by another-side block, which is two parallel monologues, not a dialogue. Interleaving forces each round to engage with the prior round's actual argument rather than continuing the prior round's own side.
The override — closest-to-implementation early
One legitimate exception to the senior-first rule: when the load-bearing input is ground truth from the implementation, not framing or strategy. In that case, the person closest to the actual code or system goes early, even if they're junior, because the seniors need to respond to what's actually happening, not to a sanitized summary of it.
This is rare and the organizer should be deliberate about it. Default to senior-first; override only when implementation reality is the thing the chain needs to absorb before it can produce useful framing.
The protective effect
The serial-with-senior-first structure is also a junior-protection mechanism, not just a sequencing one. In a meeting, a junior's incomplete argument can be demolished in real-time by a senior with information the junior didn't have. That demolition could be educational, but most of the time it isn't — it's public correction that punishes the junior for not knowing what they didn't know. The junior learns to stay quiet. The org loses a voice it should have heard.
The Senate format prevents that. Seniors put their information on the table first, in writing, available to everyone, before juniors are asked to speak. The junior reads, learns, and writes their position with the senior's framing already absorbed. No ambush. No surprise reveals. No cudgel. The junior contributes from a position of having read the thread, not a position of being caught short.
If you're organizing a Senate, the routing list is your most important decision. Make it deliberately. Pressing party first. Counter next. Senior over junior. Interleave roles. Override only when ground truth needs to lead. The chain does the rest.
Don't block the flow. Come back when you can.
The other failure mode is rarer and quieter, but worth naming because it produces low-quality input dressed up as participation: the participant who engages under bandwidth pressure they don't actually have, throws together 30 minutes of half-context response at end-of-day to keep the chain moving, and ends up contributing something they wouldn't have written if they'd had a real hour to think.
That's a protocol failure, not a participant failure. The protocol implied engage now or your input is lost. The participant did the right thing under the wrong rule.
The fix is explicit: defer-and-rejoin is a legitimate move. If the token reaches you and you don't have the capacity to engage well, you say so — pass the token to the next participant in the routing list, note that you'll come back, and rejoin the chain when you actually have time to read it carefully and write something worth reading. Sequence matters; speed matters; quality of each link matters more than either.
- The token isn't a deadline. It's an invitation to engage when you can engage well.
- If you don't have the bandwidth, say so and pass it forward. Don't rush a half-context response.
- Rejoin the chain when you have time. Reference earlier rounds; add what you would have added if you'd had time the first pass.
- The decision-maker can wait for late-arriving high-quality input or close without it. That's their call, not yours.
Senates work because each link adds real signal. A rushed link adds noise dressed up as signal, which is worse than no link at all. Better to defer with honesty than participate with pressure.
What Actually Happened
The token went out Sunday evening. By the time the meeting was scheduled to start Tuesday morning, four of seven positions were complete.
Here is what those four rounds produced — without revealing the topic:
The proposer laid out the decision, the options, and a clear position. Cost of inaction quantified. The ask made explicit. Architecture described in enough detail to evaluate.
The second participant didn't just agree or disagree. They reframed the entire situation — providing context the first participant didn't have. The decision as framed was already out of date. Here is the actual current state. Here is why it is worse than described. Here are the prerequisites before any architecture can be selected. Conditional hold.
The third participant built on the reframe. Pushed back on the timeline — one week was optimistic, three weeks was realistic. Named the right architecture from the options on the table. Flagged that prerequisites with no owners will not resolve themselves. Proposed conditions: named owners, deadlines, contractual SLAs, team structure. Also named an open question the chain hadn't answered — and said the decision-maker probably shouldn't decide without knowing.
The fourth participant did something nobody in a meeting would have done. They disaggregated the question. The chain had been treating one decision as one decision. It was actually two — with different security models, different approval histories, different owners. They named both. They argued hold on the first, proceed on the second, citing an eight-month-old approved design that the confusion had swept up unfairly. They also named the specific document that was causing everyone to argue past each other — and formally retired it from the debate.
The Score
By the time the meeting was scheduled to start:
- Two separate decisions, cleanly separated for the first time
- An outdated document named and formally retired from the debate
- Three architecture options on the table with named tradeoffs
- A critical technical constraint surfaced — not from the proposer, from the person closest to the implementation
- Prerequisites identified, with the explicit condition that they need named owners before the hold begins
- An open question surfaced that changes the urgency calculus entirely
- A decision record that will survive the next reorg, the next hire, and the next time someone asks "wait, why did we do it this way?"
The meeting would have started at 10am. People would have been catching up on the documents for the first twenty minutes. Someone would have asked a question already answered in the chain. Someone else would have circled back to a point that had already been resolved.
Why AI Makes This Work
Each participant had approximately 30 minutes. AI did the drafting. The human provided the direction, the domain knowledge, the judgment — and approved the output.
This is the important part: the quality did not come from AI knowing the answer. The quality came from each participant being forced to commit to a position in writing before passing the token. That discipline — no position, no forward — is what generates the thinking. AI just removes the friction of turning that thinking into prose.
Without AI, 30 minutes is not enough time to read the prior chain, form a position, and write something worth reading. With AI, it is.
The result is that eight smart people, each spending 30 minutes, produced more structured thinking than an hour-long meeting would have. Asynchronously. On their own schedules. Without a single Zoom room.
The Economics
Eight people × 30 min Senate = ~4 hours of work
Meeting output: action items, none owned, follow-up meeting scheduled
Senate output: decision record, named positions, dissent preserved
The Senate is cheaper. The Senate produces an artifact. The meeting produces a memory that degrades immediately and disappears entirely when someone leaves the company.
The Close
The meeting is still on the calendar. It may still happen. But if it does, it will be a different meeting — shorter, sharper, with everyone already knowing where everyone else stands. No catching up. No rehashing. Just the remaining disagreements, if any.
That is the best version of a meeting: the one that only needs to exist for the things a meeting is actually good at. Relationship. Nuance. The stuff that doesn't fit in writing.
Everything else? The Senate already handled it.
Disclosure: prose drafted by Claude (claude-sonnet-4-6) with direction and approval from Bill Berger.
← The Senate Pattern · YOU++