How I built Next Step PM

Courses teach product management in the abstract. Most of us learn it on a real product, under a real deadline, picking up advice as we go. I wanted a plan that starts from the work in front of you: your role, your product, where it is now, your biggest challenge and how long you have.

You answer a few questions and get a step-by-step plan to follow on that work, pitched at your level. There's no sign-up, and your answers live in the web address, so you can bookmark or share your plan.

No one could write a curriculum for every mix of role, product, stage and level, so I didn't try. Instead, the site has 171 learning blocks. Each one is a small, reusable activity, such as writing a problem statement. It has steps for three levels of experience, who in a team can help, what “done” looks like, common pitfalls and the sources its advice comes from.

A rules engine picks the blocks that fit your answers, puts them in a sensible order, moves the ones that help with your challenge to the front and spreads them across your timeline. It's plain code with no AI and no database, so the same answers always give the same plan, and every block can say why it's in yours.

• Every block is checked against the content rules. A validator checks each block file: only allowed values, steps for each level, real links between blocks and at least one source. A block that breaks a rule fails the build, so it can't reach the live site.
• Automated tests build real plans. They try every route through the questions, at every level and way of working, and check that each plan follows the rules: phases in order, earlier work first, sensible timings and clean wording.
• Plans are read as well. Tests show that a plan is valid. Only reading it shows whether it's useful. A usefulness check on 30 realistic profiles found every plan valid, but some too long and some with advice that didn't fit, such as team rituals for a PM with no team. That led to the next round of fixes.
• Every change gets a preview. Each change is a pull request with its own preview link. I test it there, and it only goes live when I merge it.

• Written rules beat clever prompts. The working rules, design system and content model live in files that Claude Code reads at the start of every session. Getting those right did more for quality than any single instruction.
• Small steps I can check. I work in short phases. Claude Code builds one, tells me exactly what to click, and nothing is committed until I've seen it working.
• Ask it to disagree. I asked for a partner who challenges ideas. A review of the live site found that a Director got the same hands-on plan as a PM. That led to a separate view for leaders and a path for people exploring the role.
• Keep a decision log. Every decision is written down with its date and the reason for it, so later sessions stay consistent and I can explain any choice.