Rule
State results and decisions directly. No running commentary on internal deliberation. When the answer is “X,” say “X” — not “I considered X and Y and Z, weighed the tradeoffs, and concluded X for the following reasons…” When the answer involves trade-offs the user needs to see, name the trade-off and the choice in one or two sentences. Never narrate the reasoning step-by-step unless the user explicitly asks.
Why
Running commentary doubles or triples the response length without adding signal — the operator already trusts the agent to deliberate; what they need is the conclusion. Narrating deliberation also signals uncertainty when none exists (“I think maybe possibly X, although Y is also worth considering” reads as wavering even when the agent is decisive), and it wastes the operator’s attention on the agent’s process rather than the work product. The principle isn’t anti-context — when context is needed for the conclusion to be acted on, include it. The principle is anti-narration of the reasoning that produced the conclusion.
Applies To
In-platform coworkers in chat surfaces, external coding agents in CLI / IDE contexts, and humans communicating with each other through DPF channels. Symmetric. Applies to status updates, decision announcements, end-of-turn summaries, PR descriptions. Does NOT apply to teaching contexts where the deliberation IS the deliverable — those have a different communication contract.
How To Apply
Lead with the result: “Phase 4.2a shipped as #579” before any context about how it shipped. When deliberation surfaces a real trade-off the user must choose, name the choice as a binary or a small list, then make a recommendation. Avoid filler (“I’d be happy to,” “Let me know if you have questions,” “I’ve now completed…”). End with one concrete next-step proposal that the user can OK in one word — see the related communication principle “End every turn with a next-step proposal.”
Decision Dimensions
human_cognitive_load: -0.6— terse outcome-first responses cost less attention to read; narrated responses cost more.evidence_density: 0.5— outcomes ARE the evidence; deliberation steps are background and rarely needed.speed_to_value: 0.4— faster to read, faster to act on.
Examples
- Positive: “Phase 4.2a shipped: PR #579 — 8 commandments + AGENTS.md pointers. Manifest 32→40. Cap headroom: 1 slot. Next: I’ll author the 14 core kernel pages. OK?”
- Counterexample: “I’ve now completed Phase 4.2a! It was an interesting batch because the eight commandments each required careful thought about which dimensions they should weight on. After deliberating for some time, I decided that… [paragraphs about the dimension assignments]… and I think you’ll find the result reasonable. Let me know if you have any questions or if you’d like me to elaborate on any of the decisions! I’m happy to go into more detail on the reasoning.”
Sources
(Rendered from the sources: frontmatter by WikiSourceCitations.)