Stonewall: AI workflows for product managers — the 2026 playbook
Published in May 2026, this article from Stonewall addresses a problem most product managers hit after their first months of working with AI tools: the one-off prompt gets things done once but does not compound. The piece argues for treating AI not as a chat interface but as a system of reusable, versioned workflows — and it gives a practical architecture for doing that.
The core distinction: prompts, skills, and integrations
The article separates three levels of AI usage. A prompt is a one-time instruction that lives in a conversation and disappears. A skill is a versioned document — described in a SKILL.md format — that encodes a repeatable process: a competitive analysis method, a PRD structure, a prioritization protocol. The skill runs the same way each time, the way a specific PM would run it. An integration connects the skill to live data via MCP, removing the manual step of copying context into a new conversation each time. As the article puts it: “A skill is encoded judgment — your competitive-analysis method, your PRD structure, your prioritization rules — captured so the agent runs it the way you would.”
The six workflows
The article identifies six workflows worth formalizing:
Feedback synthesis clusters themes by frequency and severity from incoming signals, removing the need to read through all feedback items manually before seeing what to prioritize.
User-research synthesis processes interview transcripts and produces quote-backed theme summaries, reducing the time from data collection to insight by a significant margin.
PRD drafting generates specification documents grounded in actual codebase context, which matters because specs written without that context often describe things engineering has already built differently.
Competitive monitoring automates tracking of pricing, features, and review sentiment across defined sources, replacing the manual process of periodically checking competitor pages and newsletters.
Ticket generation creates Jira tasks automatically from approved specifications, maintaining the connection between what was decided and what was built.
Stakeholder updates drafts status reports from issue trackers, removing the recurring task of translating project state into prose each week.
Prioritization: the underused case
The article identifies prioritization as the most-needed but least-automated workflow in most teams. Standard frameworks like RICE generate consistent scores only if the input data is itself consistent — and it usually is not, because it is collected manually from different sources at different times. The Stonewall approach automates input collection from multiple channels, deduplicates signals, and produces evidence-based scores. Human judgment enters at the final decision step, not the data-collection step.
Implementation sequence
The recommended sequence is narrow by design: audit your week for repeatable tasks, encode the highest-frequency one as a skill, connect it to live data, integrate it into your primary work environment, and maintain human review of all outputs. Starting with one workflow rather than six is explicit advice in the article — each additional workflow adds complexity, and the value of systematizing even one is enough to validate the approach before expanding.
This is most useful for product managers who are past ad-hoc AI use and want consistent, team-shareable processes. The article assumes a working familiarity with AI tools but does not assume any particular platform.