Ninety-eight percent of product managers now use AI at work—an average of eleven times a day—yet only 39 percent have received comprehensive, job-specific training, according to a survey from tech education company General Assembly. The survey collected responses from 117 product managers in the United States, United Kingdom, Canada, and Singapore, all working at companies with at least 100 employees. Nearly half taught themselves the tools they use.

What PMs Are Actually Doing With AI

Adoption has moved well past chatbot queries. Seventy-eight percent of respondents use AI agents, and 31 percent have created or adapted custom language models, specialized agents, or domain-specific GPTs. PMs reported relying on AI for managing development cycles, coordinating cross-functional teams, building roadmaps, running customer interviews, and analyzing feedback data.

The sharpest skill gap shows up in "vibe coding"—prototyping and validating product concepts without engineering support. Forty-seven percent of PMs want to learn it; only 38 percent currently do. Beatrice Partain, director of product marketing at General Assembly, argues the practice is valuable precisely because it collapses the distance between idea and testable artifact.

"It's helping you get to the prototype stage faster. In the past, you might have drawn out an idea on paper or a whiteboard," she says. A working prototype customers can interact with generates feedback before you spend engineering hours.

Partain recommends starting with experimentation: explore no-code or AI prototyping platforms to learn their basic capabilities, then learn to evaluate output quality. Define a personal framework—say, a repeatable process for setting up a prototype. And use peer learning: group training sessions, workshops, or informal knowledge-sharing with other PMs.

Productivity Without Headcount Cuts

Ninety-seven percent of respondents said AI helps their departments make faster decisions, and 98 percent said it improves their product lifecycle. Two-thirds reported productivity gains without headcount reductions; more than a quarter said their teams have grown since adopting AI. Only one percent reported shrinking teams.

The anxiety persists anyway. Twenty-six percent said their biggest concern is that AI could eventually replace their roles. Another 25 percent worry it will make it harder for entry-level PMs to build foundational skills, and 22 percent fear broader workforce displacement among their peers.

Vishal Sood, chief product officer at Typeface, sees the winners taking a hands-on approach. "They're not waiting for engineering to validate their ideas, they're using prompt engineering and no-code tools to create compelling proof-of-concepts that get stakeholders excited and customers engaged," he says. By the time these PMs hand work to engineering, they have de-risked the core assumptions and built organizational buy-in around a proven concept.

Why Generic Upskilling Fails

The training demand is specific. Nearly two-thirds of PMs want regular training updates as AI tools evolve, just over half value peer learning sessions, and nearly half favor self-paced programs with product-specific examples. Others cited technical support and hands-on workshops built around practical use cases.

Partain is blunt about what doesn't work. "One thing we have found is that generic AI upskilling is largely ineffective," she says. "People need to learn how they can use AI as part of their specific workflows and responsibilities." Telling a team "we want everyone to use AI" is useless without defined outcomes. Her prescription: targeted programs mapped to specific PM workflows, peer learning built in, and continuous updates as the technology moves—so usage supports business goals, not just individual productivity.

Governance Through Platforms, Not Rulebooks

Sood argues governance should be solved through platforms and configurations, not policy documents. "Embed AI into the systems people already use rather than hoping they'll follow compliance guidelines," he says. The right platform choices handle governance at the infrastructure level and prevent the productivity islands that emerge when everyone adopts their own tools.

Partain agrees an AI usage policy matters: once you have clear goals and guardrails, you can map training to those outcomes. Ryan Gialames, principal product manager, innovation at Western Governors University, recommends sandbox environments plus a clear set of Gen AI Design Principles. "Providing approved, internal agentic tools that are pre-wired for security and data governance allows teams to innovate boldly while ensuring data provenance, compliance, and a seamless, user-centered experience," he says.

What Hiring Managers Actually Screen For

Sood hires for speed and iteration: breaking down roadmaps, turning ideas into AI-powered experiments. "Experience with prompt engineering, rapid prototyping, and integrating AI into broader product and marketing ecosystems really stands out," he says. "We also want to see impact." The strongest candidates show shipped MVPs delivered faster, automated grunt work, and measurable results—with portfolio examples that make the stories concrete.

Partain's advice for interviews: skip the tool inventory. "It's ultimately all about the outcomes you achieved, regardless of what tools or technology you used to do it," she says. Curiosity about AI helps, but the differentiator is how you used it to create value for the business or end users.

The survey's core tension—near-universal adoption against a 39 percent training rate—won't resolve itself. As AI absorbs more of the product craft, the PMs who compound their judgment with prototype-speed iteration, inside governed platforms, will set the bar everyone else is hired against.