How to make AI better at product 🎨

2025-04-30 · Source: Refactoring · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Project & Product Management, Software Development & Engineering · Depth: Intermediate, medium

Summary

The article highlights a transition in AI development from "prompt engineering" to "loop engineering," advocating for systems thinking to ensure reliability and sustainability. It argues that product development is conceptually more challenging than coding for AI, noting that only 9% of teams use AI for product specifications compared to over 90% for coding. To address this disparity, the author and Doug Peete from Atono propose adapting "Architecture Decision Records" (ADRs) into "Product Decision Records" (PDRs) to document product intent, design, and tradeoffs. Additionally, they suggest creating a "Product Glossary" to define core product abstractions and domain language, providing AI with essential context. Atono's internal tests indicate that glossary-backed AI stories can reduce rework from 60% to 20%, suggesting a path to improve AI's utility in product development workflows.

Key takeaway

For AI Product Managers or Directors of AI/ML aiming to scale product development, recognize that AI's current low adoption for product specs stems from insufficient contextual grounding. You should implement Product Decision Records (PDRs) to capture design intent and tradeoffs, alongside a comprehensive Product Glossary defining core abstractions. This structured knowledge will enable AI to generate more accurate draft specifications, potentially reducing rework from 60% to 20% and making AI a valuable partner in your product workflow.

Key insights

Structured product knowledge (PDRs, Glossaries) enables AI to significantly improve product development, mirroring its success in coding.

Principles

• Product is conceptually harder than coding.
• Systems thinking improves AI reliability.
• Documenting decisions injects judgment into AI.

Method

Capture human intent, then use AI to draft specs grounded in Product Decision Records (PDRs) and a Product Glossary. Review, iterate, and update PDRs/Glossary upon shipping.

In practice

• Implement Product Decision Records (PDRs).
• Create a comprehensive Product Glossary.
• Ground AI-assisted specs in documented knowledge.

Topics

• AI Development
• Product Management
• Loop Engineering
• Product Decision Records
• Product Glossary
• AI-assisted Product Design

Best for: Product Manager, AI Product Manager, AI Engineer, Director of AI/ML

Related on AIssential

• When AI started writing PRDs, I re-understood the value of PM. — AI automates output, elevating the value of human conscious thought, judgment, and contextual understanding in product management.
• A Field Guide to Rapidly Improving AI Products — Successful AI product development prioritizes measurement, iteration, and learning over a "tools-first" approach.
• From Napkin Sketch to Roadmap: 10x Your Product Workflow — Translating product ideas into clear PRDs requires structured thinking to avoid ambiguity and ensure shared understanding.
• Why Businesses Outsource AI Product Development Companies — Outsourcing AI product development mitigates high failure rates by leveraging specialized expertise for market validation and risk reduction.
• Most Companies Don't Have an AI Problem. They Have a Decision-Making Problem. — Companies' AI challenges stem from poor decision-making, not the technology itself.
• The Metric Playbook for AI PMs: Five Layers, Four Practices, One Checklist — AI product success requires managing metric trade-offs and maintaining the chain from technical performance to business outcomes, not isolated metric improvement.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Refactoring.