Alloy: AI workflows for product managers — what changes and what to measure

Christian Iacullo published this guide on Alloy’s blog in June 2026 as a ground-level account of how AI has changed four core product management workflows. What distinguishes it from overview articles is that it goes beyond the claim that AI helps PMs work faster — each section describes what the AI actually does, names the tools most teams are using, and specifies what measurable change looks like.

Research and discovery

Tools like Grain and Dovetail convert interview transcripts into structured themes. The process that previously took two to three days of manual affinity mapping now takes a few hours. The article notes that the PM skill shift this produces is meaningful: the system finds clusters, but the PM decides what matters. Pattern recognition gives way to interpretation as the dominant skill in this workflow.

Roadmap prioritization

Jira Product Discovery and Productboard are the platforms cited for AI-assisted feature scoring. Both surface customer feedback signals and score features against strategic objectives without requiring custom data pipelines. The framing is that AI reduces the cost of being systematic — when scoring is automated, it becomes practical to apply the same criteria consistently rather than letting the loudest stakeholder drive the outcome.

Writing and documentation

ChatGPT, Claude, and Notion are the drafting tools named. PRDs, release notes, and user stories that once took hours now take minutes. The more consequential shift, the article argues, is what this does to the PM’s actual job: when writing is cheap, the constraint becomes deciding what to write, not writing it.

Prototyping

This is the newest category in the PM stack and also the most variable in quality across tools. Alloy’s own platform, along with similar products, allows PMs to capture a live page and hand off a GitHub pull request without engineering involvement. The ability to generate something working from a description has changed the economics of exploring ideas — finding out whether something looks right has become substantially cheaper than it was two years ago.

How to know if it is working

The piece closes with a three-layer measurement framework: model performance metrics (accuracy), user experience metrics (task completion rates), and business outcome metrics (retention and support deflection). The test the author states directly: “A model with 94% accuracy that users routinely abandon is still a failed product.” The framework applies not just to AI features but to AI-assisted PM workflows — if the tooling saves time but the outputs are worse, time saved is the wrong metric.

Relevant for product managers who want a practical update on which tools have become reliable enough to build into their regular workflow and how to judge whether the adoption is worth the change.