As distributed teams grow the messaging overhead grows. This has been shown in study after study. It has also held true in LLM "teams". In teams where there is a coordination layer the messaging is typically O(n) and in distributed teams it shoots up to O(n²).

The O(n²) growth is a coordination cost that shows up again and again in team research.

The issue isn't the messaging itself. It is the lack of consensus/shared truth, messaging volumes are a symptom of trying to sync to that consensus. Typical organizations deal with this by adding layers of hierachy. As those layers solidify they harden into rigid top-down organizations. Eventually innovation and adaptation are killed.

To facilitate leaner, more nimble teams, while fikghting rigid hierarchies, we are experimenting with three approaches: workflows, ai embedding and dynamic voting.

Workflows structure effort for transparency and global visibility. This allows teams to define their process and define their deliverables. It allows other teams to query and understand the current state and process. It eliminates the "what is happening now?" question. Workflows are defined in markdown, so they're readable, portable, and easy to version.

Dynamic voting with delegation. This reduces independent truths that need to converge which drives consensus. Teams retain autonomy and decisions remain auditable. Votes can be delegated to subject matter experts in order to get the best outcomes. Every decision is recorded and the ai agent can be interrogated on it.

AI embedding allows agents to act as a witness and can summarize and answer questions for outside teams. This means putting AI agents directly inside the workflow as participants, not just as tools teams call out to. A human status update is one person's account, shaped by their memory, bias, and framing. An embedded AI agent instead witnesses the actual events, votes cast, documents moved, state changed, and reports the same underlying facts to every team the same way. That consistency is what makes it a shared-truth mechanism rather than just a faster summary. Agents can:

Move the work forward:

  • act on external APIs

  • act after humans vote

  • move a document through a workflow

  • change state

Support the humans doing the work:

  • review documents

  • interrogate actions and raise questions

  • reach out to individuals

  • facilitate collaboration on plans or other documents

No more status meetings!!!

At cards.statmeet.com we're building a system that implements all of the above. To get a flavor of the mechanics for workflows, markdown, and AI, an excerpt from the documents is below.

AI Agent States

Set assignee: ai on a state to have the AI automatically process cards when they arrive in that state. The AI can update fields, perform a transition, and post a comment explaining its reasoning.

states:

triage: Triage

assignee: ai

ai_prompt: Assess the severity of this bug report. Set priority field to critical, high, or low. Route critical bugs to the critical state and everything else to normal.

Transitions the AI can perform must have allowed: ai. Human users cannot perform allowed: ai transitions — they are AI-exclusive.

transitions:

triage -> critical: Critical Bug

allowed: ai

triage -> normal: Normal Bug

allowed: ai

If the destination state is also assignee: ai, the AI chains automatically without human intervention.

Requirements: The card owner must have an Anthropic API key configured in Settings. If no key is found, the AI posts a comment on the card explaining why it could not run.

AI Workflow Example

name: Bug Triage

description: AI-assisted bug triage and routing

states:

new: New

initial: true

assignee: any

triage: Triage

assignee: ai

ai_prompt: Assess severity based on the description. Set the priority field to critical, high, or low. Route critical bugs to the critical state, all others to normal.

critical: Critical

assignee: assigned

normal: Normal

assignee: assigned

done: Done

terminal: true

transitions:

new -> triage: Submit

allowed: any

triage -> critical: Critical

allowed: ai

triage -> normal: Normal

allowed: ai

critical -> done: Resolve

allowed: assigned

normal -> done: Resolve

allowed: any

  • -> done: Force Close