AI for product managers: what I learned training ~20 Salesforce PMs for six weeks

The setup

This spring HFI Institute brought me in as the AI expert instructor on a programme they were building for Salesforce: six weeks of live online sessions for roughly twenty senior product managers and UX designers. I co-designed the curriculum with HFI's Abdul Suleiman and led the hands-on Claude and Claude Code sessions. The whole engagement is documented in the Salesforce case study.

The brief was specific. These PMs wanted to use AI to build the evidence and ROI case for UX decisions, and to hold real technical conversations with their engineers. AI knowledge in the room ran from beginner to expert. What follows is what actually worked, written for any product team considering the same journey.

PMs don't need to build AI. They need to work with it

The instinct on a product team is to treat AI as a feature question: what should we build into the product? We deliberately flipped that. The programme focused on how PMs use AI to improve their own workflow, analysing PRDs, making sense of survey data, preparing presentations, structuring user feedback.

That flip is what made the training land. A PM who has personally taken a messy pile of user interviews and turned it into structured findings with AI understands the technology at a level no feature briefing can teach. The product intuition follows from the personal fluency, not the other way round.

Foundations first, tools second

With a cohort spanning beginner to expert, we opened with foundations: what separates a model from an agent, how context shapes output, and the judgement to know when to trust what AI gives you. Every later session stood on that shared vocabulary.

This is the part most tool-led training skips, and it shows. If half the room thinks an agent is magic and the other half thinks it's a cron job, no amount of prompt technique will get them collaborating. The levels of AI autonomy framing is a good example of the kind of shared map that pays for itself all programme long.

The step change is project-wide context

The single biggest jump in capability came in week three, when the cohort moved from chat windows into Claude Code in VS Code, working with full project context. Suddenly the AI wasn't answering questions in a vacuum. It was reading the same documents, data, and prototypes the PM was working on.

Participants went from first prompts to working HTML prototypes they could put in front of users. For PMs specifically, quick disposable prototypes turned out to be the killer use: cheap to make, cheap to throw away, and far better than a slide at getting a real reaction from a stakeholder or a test participant.

Make the work outlive the session

One-off prompting is a treadmill. The habits that survived the programme were the ones that produced reusable assets: custom commands for repeated tasks, shared skills the whole team could pull from, and git-based collaboration so nothing lived in one person's chat history.

The same principle applied to research. We built workflows that turn raw user feedback into structured findings and an evidence trail, with a human making the judgement calls at every step. The aim was always to prepare for human research, not to replace it. AI compresses the drudgery; the PM keeps the decisions.

What I'd tell any product team

If you're putting AI training in front of PMs, the shape that worked here:

• Survey the cohort before session one. We mapped AI knowledge, pain points, and tool access first, and rebuilt the curriculum around what we found.
• Run it on real work. Every session used the cohort's actual PRDs, feedback, and prototypes rather than canned exercises.
• Keep rebuilding. We reshaped sessions weekly based on what the cohort struggled with. Dense sessions were split, abstract material was cut for demos.
• Work inside the approved stack, and fight for the tools that matter. Getting proper editor access mid-programme changed what participants could do.
• End with evidence discipline. The lasting skill isn't prompting. It's using AI to build a case a CFO or an engineering lead will accept.

That shape isn't specific to Salesforce or to product managers. It's just what tailored training looks like when the goal is behaviour change rather than a nice feedback score. If you want the fuller picture of how the programme ran week by week, it's in the case study, and the general buying questions are covered in how to choose an AI training provider.