Run an autoresearch loop to optimize any product artifact

The Problem

Product teams guess more than they admit. A PM writes a PRD, reviews it once, ships it to engineering, and moves on. A designer creates a user flow, gets feedback in one round, and calls it final. The artifact improves once, maybe twice, then freezes. But the best version of any product artifact is rarely the second draft. It is the sixth.

The problem is not that teams lack standards. It is that iteration is expensive. Each review cycle requires scheduling meetings, waiting for feedback, resolving conflicting opinions, and manually applying changes. According to Atlassian, knowledge workers spend an average of 31 hours per month in unproductive meetings, many of which are review sessions that could be replaced by asynchronous, automated feedback loops.

Meanwhile, the artifacts that drive product decisions, PRDs, design briefs, competitive analyses, launch plans, get fewer iterations than the code they inform. A GitClear analysis found that the average pull request receives 2.4 rounds of review before merging, but the product documents that define what gets built receive an average of 1.1 rounds. The document that matters most gets reviewed the least.

The Case for Looping

AI changes the economics of iteration. When a review cycle costs five minutes instead of five days, you can iterate six times before lunch. The constraint shifts from "How many reviews can we afford?" to "How good do we want this to be?"

How This Prompt Works

This prompt sets up an autonomous optimization loop that takes any product artifact, reviews it against defined quality criteria, generates specific improvement recommendations, applies them, and repeats until the artifact converges on a quality threshold.

The loop follows a four-step cycle:

• Evaluate: The AI reviews the current artifact against a rubric you define (clarity, completeness, specificity, actionability, or custom criteria)
• Diagnose: It identifies the weakest dimension and explains why it is weak, with specific evidence from the text
• Improve: It generates a revised version that addresses the diagnosed weakness while preserving the strengths
• Compare: It scores the revised version against the original and decides whether to loop again or stop

The loop terminates when the improvement delta drops below a threshold you set, or after a maximum number of iterations. The output includes the final artifact plus a changelog showing what improved in each round.

When to Use It

• You have a first draft of any product document and want to reach "review-ready" quality before involving stakeholders
• You want to stress-test a PRD, strategy memo, or launch plan against specific quality criteria
• Your team ships artifacts too fast and needs a quality gate that does not require a meeting
• You want to see how your writing improves under systematic pressure

Common Pitfalls

• Running too many loops without human checkpoints: After 3-4 iterations, AI optimization can start to overfit. It polishes language at the expense of voice, or adds precision at the expense of readability. Review the output every 3 loops.
• Using vague evaluation criteria: "Make it better" is not a rubric. Define specific dimensions: Is every user story testable? Does every requirement have an acceptance criterion? Are all assumptions stated explicitly?
• Optimizing the wrong artifact: Looping is most valuable on high-leverage documents (PRDs, strategy memos, launch plans). Do not spend six iterations optimizing a meeting agenda.

Further Reading

• You Waste a Lot of Time at Work - Atlassian
• Software Development Velocity Trends - GitClear
• The Art of Iteration in Product Development - Silicon Valley Product Group