A product decision log tells an AI assistant which choices you have accepted, why you made them and when to revisit them. Record proposals separately, keep the evidence attached, and preserve the history when a choice changes. This helps the next session use your current direction without treating every old idea as an instruction.

For a solo founder, start with a Markdown record you can maintain. You can use the same practice in a shared product workspace; the value comes from clear decisions and useful evidence.

What belongs in a product decision record?

Record choices that materially change the next task: the intended customer, a pricing hypothesis, excluded scope, the next experiment or the conditions for releasing a pilot. Routine edits rarely need a decision record.

  • Question: the choice you needed to make.
  • Status: whether it is proposed, accepted or replaced.
  • Choice: a specific direction and its boundaries.
  • Reason and evidence: what supports it and what remains uncertain.
  • Confirmation: who accepted it and when.
  • Review trigger: new information that would justify reconsidering it.
  • Next action: the work affected by this choice.

An accepted experimental direction does not establish customer demand. You can choose to test a segment while leaving the segment's willingness to pay unverified. The validation guide explains how to connect that choice to observable evidence.

Copy the decision log template

Download the decision log template ↓

# Product decision record

A manual note template. In 1PC, save relevant choices in department documents or
review notes; this does not create an independent Decision object, approve a
department or trigger downstream updates.

ID: [stable reference]
Question: [the choice to make]
Status: Proposed
Last reviewed: [date]

## Choice
[One specific direction, including its boundaries.]

## Reason and evidence
- Reason:
- Observation or source, with date:
- What the evidence does and does not establish:
- Remaining uncertainty:

## Alternatives considered
- Alternative:
- Reason for excluding or deferring it:

## Acceptance
Accepted by: [leave empty until the responsible person accepts]
Accepted on: [date of actual acceptance]

## Review trigger
[New evidence or a change that would justify reconsidering the choice.]

## Affected work
- Next action:
- Relevant context or scope document:
- Work this choice excludes:

## Replacement history
Replaces: [earlier record, if any]
Replaced by: [later accepted record, if any]

## Working rules
Keep recommendations proposed until accepted.
Preserve the rationale when a choice is replaced.
Separate evidence from hypotheses and completion from planned work.
Record concise reviewed outcomes, not private reasoning or raw chat transcripts.
Do not include credentials or unnecessary personal/customer data.

This is a document format you control, not a JSON payload for a 1PC tool. Downloading it does not submit or store your decisions. Provide it to an AI client or save it in a workspace only when you choose to do so.

A founder decision with an explicit uncertainty

This AgentCRM example is hypothetical. The founder's capacity and choice are illustrative inputs; no interviews, customer quotes or market results are implied.

ID: DEC-001
Question: Should the first experiment cover agents or agencies?
Status: Accepted
Accepted by: The founder, for this example

Choice:
Start with independent residential agents.
Test the inquiry follow-up problem before building a full CRM.

Reason:
The founder can reach this segment and has 10 hours this week.

Evidence:
No completed interviews. Problem urgency is still a hypothesis.

Alternative considered:
Agency-wide administration, deferred because the test is broader.

Review trigger:
Initial interviews point to an agency-level buyer or a different problem.

Next action:
Review the interview outline and identify reachable candidates.
Do not contact them without a separate outreach instruction.

A new session can now suggest an interview plan that fits the accepted experiment. It should still ask for evidence before describing the segment as validated. It also has a clear reason to revisit the choice rather than treating it as permanent.

When a decision changes, keep the explanation

Suppose later research changes the intended buyer. Record the actual observations, their source and limits first. Then propose a replacement decision. The founder decides whether to accept it.

In a file-based log, mark the old record as replaced and link it to the new record. Keep the old rationale available, while making the current choice easy to find. An index can list the active decision for customer, scope and pricing so the assistant does not need to interpret the entire history.

A recommendation should stay proposed until accepted. Avoid silently rewriting “we may test agencies” into “agencies are our target customer.” Also avoid converting a completed draft into a completed experiment; those are different outcomes.

In 1PC, record decisions in scoped Knowledge and Task result reviews. Their saved revisions retain the reasoning; decisions are not separate records. The Task and Knowledge reference explains the supported behavior. A current revision check helps prevent a save from overwriting changes another session made.

Use decisions to guide the next task

Keep the latest relevant decisions with your founder context. Before asking for a plan, tell the assistant how to handle proposed choices, conflicts and missing evidence:

Read the current product context and relevant decision records.
Summarize the accepted customer, scope and experiment choices.
Keep proposed or replaced decisions separate.

Plan the next validation step using the accepted direction.
If new evidence conflicts with it, explain the conflict and propose
a review. Do not silently accept a replacement decision.

At the end, identify any record that needs an update.

For an MCP workflow, select the correct accessible product and load its current records before writing. A model's answer that says “saved” is insufficient: confirm the tool's successful result and the saved record when a write was requested.

Store a concise rationale, evidence and the choice. Private reasoning and raw conversations are unnecessary for this log. Omit credentials and personal/customer details that do not affect the decision.

Continue with the guide to product context across AI tools to carry the decision into another session. Use the file-versus-workspace comparison when maintaining the current version becomes the problem.

Make the current choice easy to find.

Use the free decision record to separate proposals, accepted choices and replacements. Keep the rationale with the decision.

Download the decision log ↓