From Skillet to Product Atlas

Context-building, not prompt-writing

When I joined Artium in January 2026, I didn't bring a skill library. I brought a deep understanding of how to build context for AI tools. Months of work with Claude and ChatGPT had taught me that the bottleneck was never the model. It was the structure around it: how you organize knowledge so an AI agent can actually use it.

Artium had a consulting problem I recognized immediately. Smart people reinventing wheels across engagements. The institutional knowledge existed, but it lived in people's heads. It didn't compound.

The fix wasn't better prompts. It was reusable, structured workflows that encoded methodology. I started calling them skills: markdown files with trigger conditions, numbered steps, knowledge references, output templates, and cross-references. Designed for Claude Code and Codex. Each one turns a repeatable consulting workflow into something an AI agent can execute with a PM in the loop.

The synthesis problem

Everyone assumed the hard part of AI adoption was getting people to use the tools. It wasn't. The hard part was synthesis.

Artium had five AI initiatives running in parallel. Ugo's PM Workflow, Rashard's Arty Slack bot, Jason's OMP, John's AI PM baselines, Mark's AI skills framework. Each team solving a real problem. None of them talking to each other in a structured way.

My role became central synthesizer: the person who saw patterns across workstreams and turned them into reusable components. Same PM discipline, just applied to internal infrastructure. Discovery, prioritization, architecture. The methodology got pressure-tested on client engagements at Fortrea and Thinkific. Different domains, same underlying system.

How it's built

Three pieces make the system work.

Skills follow a strict convention. Title, status badge, "use this when" opener, numbered steps, output format. Once you've seen one, you know how to read all of them. So does an AI agent.

The knowledge canon means skills don't duplicate reference material. They point to it. One canonical file per domain. Update it once, every skill gets the current version.

The intake pipeline is the piece most people skip. New content gets triaged through a queue, processed, and integrated with cross-references. Without it, any skill library becomes a junk drawer within a month.

The system grew from zero to 340+ skills in five months at Artium without collapsing. The architecture held.

From skills to recipes

Individual skills are useful. Chained skills are transformative.

The system evolved into three levels. A Skill is a single job. A Skillset chains skills for a phase of work. A Recipe chains them into a complete methodology.

The product-strategy recipe chains 16 skills across four phases: Foundation, Strategy, Prioritize, and Deliver. It starts with JTBD analysis and ends with a coherence audit across every artifact produced. The key design: bridge questions between every step where the PM validates before the next skill runs. The system doesn't automate judgment. It structures it.

Already have personas? Start at step 4. Every skill still works standalone. The recipe is a map, not a mandate. That's what makes it a Product Atlas.

What happened after Artium

I left Artium in May 2026. Some of what I built there is probably theirs. But a system like this doesn't sit still, and the version that exists today isn't what existed at Artium.

Since going independent, the system has nearly doubled to 584 skills. More importantly, it's changed shape. The recipe layer didn't exist at Artium. There are now 18 recipes covering everything from fractional launch playbooks (signed contract to demonstrated value in 14 days) to hard conversation prep kits to full UX practice cycles. Skills I originally wrote for Artium's consulting context have been rewritten for my own: different voice, different knowledge references, different output formats tuned to how I actually work with clients now.

The biggest shift is that the system powers a live product. At Artium, Skillet was internal tooling. Now it runs k8mak.com. The site's tools, content workflows, and audit templates are all skills I use with clients. It isn't a brochure for my work. It's an artifact of it.

Independence removed the distance between seeing a need and shipping the solution. No approval layer, no "whose initiative does this belong to." Just: notice a pattern, build the skill, use it on the next engagement.

The meta-lesson

Product managers build systems that help people do work. That's what Skillet is. The material changed: prompts instead of features, skills instead of user stories. The discipline is identical.

The interesting question isn't whether AI replaces product managers. It's whether product managers recognize that building AI infrastructure is product work, and that the skills they already have are exactly the ones it requires.