Product Managers Are All In on AI, but Training Gaps and Shadow AI Put Teams at Risk

Product Managers Are All In on AI - But Skills Gaps and Shadow Tools Put Outcomes at Risk

AI is now embedded in product work. According to a new survey from General Assembly, 98% of product managers use AI on the job - yet only 39% received comprehensive, role-specific training. Meanwhile, 66% admit they're using unapproved tools. That mix fuels speed, but also risk.

Adoption Is High. Training Isn't.

Nearly half (45%) said they taught themselves AI. They use it about 11 times per day on average. Advanced usage is already common: 78% use AI agents, and 31% are building or adapting custom language models, specialized agents, or CustomGPTs.

Where PMs Use AI Today

• Managing product development cycles, sprint planning, and delivery - 54%
• Cross-functional collaboration - 52%
• Creating product strategies and roadmaps - 48%
• Developing customer interviews or role-playing interviews - 46%
• Backlog grooming, ticket creation, or QA support - 44%
• Analyzing customer feedback - 42%

The Skills Gap You Can't Ignore

There's a clear gap between what PMs want to do with AI and what they're doing now. For example, 47% want to learn "vibe coding" (prototyping and validating product concepts without heavy eng support), but only 38% do it today.

What PMs say would help them stay sharp:

• Regular training updates as tools evolve - 64%
• Peer learning sessions - 51%
• Self-paced trainings with product-specific examples - 49%
• Ongoing support and troubleshooting - 40%
• Interactive workshops focused on real product use cases - 37%

Impact on Teams

• 97% say AI helps their department make decisions faster
• 98% report improved product lifecycle outcomes
• 66% see productivity gains without headcount growth
• 26% say their team has expanded
• Only 1% report fewer people on their team since adopting AI

Career Concerns Are Real

• 26% worry AI could eventually replace them
• 25% worry it could make it harder for entry-level PMs to learn
• 22% worry it could replace colleagues

Shadow AI: A Management Problem, Not a User Problem

Two-thirds of PMs are using unapproved tools. That's a governance gap. The solution isn't to clamp down - it's to set clear guardrails and give people safe, capable options that match how they work.

• Publish an approved tool stack with clear data-handling rules.
• Classify data and restrict sensitive inputs by default.
• Adopt an AI risk framework and log model/tool usage for audits. See the NIST AI Risk Management Framework for a simple starting point: NIST AI RMF.
• Create minimum viable AI reviews for product changes that touch users or data.
• Stand up a help channel for prompt patterns, failure modes, and red flags.

A 30-60-90 Day Plan for Product Leaders

• First 30 days: Audit AI usage across squads. Identify top 3 high-value workflows. Publish "safe use" guidelines and an approved tool list. Pick 2 pilot use cases per team.
• Next 30 days: Build squad-level agents for routine PM tasks (grooming, interview prep, QA triage). Standardize prompt libraries. Instrument outcomes (cycle time, defect rates, decision latency).
• Final 30 days: Scale what works. Add role-specific training paths. Tie AI practice to onboarding and performance. Run quarterly model/tool reviews and refresh guardrails.

What This Means for Management

The upside is clear: faster decisions, smoother lifecycles, and more output without adding headcount. The risk is also clear: inconsistent quality, data exposure, and knowledge that lives in ad hoc prompts. Close the gap with structured, job-specific training and lightweight governance that keeps speed intact.

Survey Snapshot

General Assembly surveyed 117 product managers from Oct 2-13, 2025 across the United States, United Kingdom, Canada, and Singapore. Respondents work at companies with 100+ employees and manage software and digital products.

Where to Skill Up

If your PMs are self-teaching while shipping, you're leaving results on the table. Give them role-specific paths and ongoing refreshers so experimentation becomes repeatable practice.

• Explore role-based AI learning tracks: Complete AI Training - Courses by Job

Bottom line: AI is already in your product workflow. Make it safe, measurable, and teachable - then scale it.