Loading Runway...
Loading Runway...
Evidence-backed analysis across 20 specific tasks. Capability claims sourced from peer-reviewed research, independent benchmarks, and industry data. Adoption rates tracked by industry and company size.
At a glance
Early Signal intelligenceTasks tracked
Signals in database
Intelligence confidence
Last updated
AI Exposure
Defensibility
Avg Capability
20/20 tasks with evidence
Avg Deployment
159 evidence sources
What's changing for Product Managers
PM hiring volumes dropped sharply in 2023 and have not fully recovered. The companies growing PM headcount are concentrated in AI-native products and infrastructure, fintech, and healthtech — roles that require domain depth, not just process fluency. Generalist PM roles at consumer internet companies continue to contract. Salary premiums are accruing to PMs who can operate in technical ambiguity: reading code, partnering on model evaluation, and writing precise specs for AI-driven features. Companies building with LLMs increasingly want PMs who can reason about probabilistic outputs and latency tradeoffs, not just user stories. B2B SaaS PM roles remain the largest hiring pool by volume, but competition is high relative to openings. The most durable differentiator in 2024–2025 hiring data is evidence of measurable outcome ownership — PMs who can point to retention, activation, or revenue impact they personally drove. Discovery and strategy skills are being tested more rigorously in hiring loops than two years ago. Execution-only PMs face the most pressure.
Synthesised by claude-sonnet-4-6 · refreshed May 21, 2026
Capability dimensions
How the dimensions of this role are being reshaped by AI · top 8 by weight
Prioritisation & Tradeoffs
Customer & User Understanding
Outcome Ownership
Roadmapping & Sequencing
Problem Framing
Cross-functional Influence
Stakeholder Management
Written Communication
Market Context
Senior PM demand remains strong — AI is amplifying output, not replacing strategic judgment. Entry-level PM/APM hiring is down ~73% (Veritone Q1 2025) as AI handles first-draft PRDs, user stories, and meeting notes. Senior PMs who can direct AI and make strategic product bets are increasingly valuable. Vibe-coding tools now allow PMs to prototype directly, reducing engineering dependency for MVP scoping.
Source: Based on Veritone Q1 2025 Labor Market Analysis, LinkedIn Talent Insights 2025, and analysis of job description shifts in PM roles across US tech companies.
Task breakdown
Top 3 per pressure tier · expand for the full list
Medium automation pressure · 4
Backlog Grooming & Refinement
GitHub's updated impact study shows 46% of all code is now AI-generated among Copilot users, with 82% developer satisfaction. For tasks like Backlog Grooming & Refinement, AI coding assistants demonstrate 69% quality on …
Competitive Analysis
DeepSeek V4 can perform data analysis tasks at a level competitive with leading US AI systems
Release Communication
Public release of agentic AI to general users demonstrates incremental improvement in computer use automation reaching broader adoption.
User Story Creation
GitHub's updated impact study shows 46% of all code is now AI-generated among Copilot users, with 82% developer satisfaction. For tasks like User Story Creation, AI coding assistants demonstrate 69% quality on routine im…
Low automation pressure · 16
Customer Feedback Triage
Fin Apex 1.0 demonstrates superior performance in handling customer inquiries compared to other AI models
Technical Feasibility Assessment
Tesla's development of fifth-generation AI chips for autonomous driving systems requires technical feasibility assessment for implementing advanced AI hardware in automotive applications
Product Experimentation & A/B Testing
Role Defensibility Profile
Higher = harder to automate
Task-Level Analysis — 20 Tasks
Evaluate competing feature requests, technical investments, and strategic bets to sequence the product roadmap based on impact, effort, and alignment with business objectives.
Highest Exposure Areas
Meetings / Coordination / Scheduling
Calendar AI and agentic scheduling tools already handle meeting coordination. The coordination value that remains human is the nuanced political navigation — and that erodes as AI gains organisational context.
Writing / Summarising / Documentation
GPT-5 Deep Research and Claude already produce publication-quality reports, emails, and documentation. By 2027, AI writing assistants will handle first-draft creation for virtually all standard business documents with minimal human input.
Customer / Stakeholder Communication
AI agents are now handling routine customer communication autonomously. The protection in this task comes from novel relationship context and trust — which erodes when your client interactions become standardised or when AI gains sufficient context to replicate the pattern.
Strongest Defenses
Decision-Making Under Uncertainty
This remains one of the most defensible task categories — AI struggles with genuine novelty and accountability. The erosion condition: as AI decision-support tools become standard, the bar for what counts as 'genuine uncertainty' rises, and roles that mostly execute defined playbooks lose this protection.
Customer / Stakeholder Communication
AI agents are now handling routine customer communication autonomously. The protection in this task comes from novel relationship context and trust — which erodes when your client interactions become standardised or when AI gains sufficient context to replicate the pattern.
Relationship Management / Trust Building
This is the false moat most people rely on. Relationship trust is real protection today — it erodes when: (a) clients become comfortable trusting AI-mediated interactions, (b) your relationship context becomes standardisable, or (c) your firm deploys AI account management tools that clients prefer for speed.
See product managers by industry
Same role, different industry-specific exposure profiles.
Pick another role to see a side-by-side AI disruption comparison. The URL you land on is shareable.
Live signals
Real-time AI signals affecting this role
Compare roles
See how other roles compare
What this means for product managers
The role-average exposure profile above is built on early signals — directionally useful but not yet corroborated across independent sources. Your specific task mix and tooling matter more than the role average here. Get a personal task-level breakdown rather than relying on the headline number.
How we build role intelligence
Runway maintains an atomic task taxonomy (20 tasks tracked for Product Manager) anchored to O*NET occupational data. Per-task signals enter through tier-graded connectors (peer-reviewed papers, statutory labour data, vendor benchmarks, preprints) and pass through the Sentinel auditor — every claim is rubric-scored, cross-checked, and confidence-graded before it can affect a role page. The narrative and task breakdown above are computed from that ledger; nothing is synthesised from first principles. See /methodology for the full pipeline.
Confidence level: Early Signal — based on 0 validated signals for this role across the Sentinel-graded sources we track.
The Stanford HAI AI Index Report 2025 documents AI systems achieving expert-level performance on graduate-level science questions and professional coding tasks. For tasks like Product Experimentation & A/B Testing, curre…
Requirements & Specification Writing
Enhanced language modeling with 4% improvement could produce better structured and clearer technical requirements documentation
User Research Synthesis
Demonstrated autonomous scientific discovery and hypothesis generation in complex biological research indicates meaningful advancement in reasoning and analysis for domain-specific research applications.
Product Analytics Interpretation
Self-service analytics agent enables autonomous query generation and analysis across distributed data sources, improving reasoning capability in data interpretation workflows.
Sprint Planning & Ceremony Facilitation
GitHub's updated impact study shows 46% of all code is now AI-generated among Copilot users, with 82% developer satisfaction. For tasks like Sprint Planning & Ceremony Facilitation, AI coding assistants demonstrate 58% q…
Cross-Functional Dependency Management
Level-4 autonomous optical network manages cross-domain cross-layer dependencies for distributed AI training with 3.2x higher performance than single agents
Metric Definition & Tracking
Tableau AI and Pulse enable natural language data querying and automated insight generation. For tasks like Metric Definition & Tracking, AI tools achieve approximately 48% quality on routine data exploration and reporti…
Pricing & Packaging Decisions
The Claude system card reports near-expert performance on graduate-level reasoning (GPQA), professional coding (SWE-bench), and document analysis tasks. For Pricing & Packaging Decisions, Claude demonstrates approximatel…
Build vs Buy Decisions
AI systems can build other AI agents through agent-driven development
Go-to-Market Coordination
AI agents can reduce coordination overhead in enterprise workflows
Stakeholder Alignment & Buy-In
Fabric IQ's solution to context fragmentation should help ensure stakeholders are working from consistent information and aligned understanding
OKR & Goal Setting
Dell'Acqua et al. found that consultants using GPT-4 completed analytical tasks 25.1% faster with 40% higher quality for tasks inside the AI capability frontier. OKR & Goal Setting contains analytical elements that fall …
Team Mentoring & Development
AI can generate design elements for web development
Roadmap Prioritisation
The Stanford HAI AI Index Report 2025 documents AI systems achieving expert-level performance on graduate-level science questions and professional coding tasks. For tasks like Roadmap Prioritisation, current AI capabilit…
Capability Evidence
The Stanford HAI AI Index Report 2025 documents AI systems achieving expert-level performance on graduate-level science questions and professional coding tasks. For tasks like Roadmap Prioritisation, ...
The IMF finds that approximately 40% of global employment is exposed to AI, with up to 60% in advanced economies. For knowledge work tasks like Roadmap Prioritisation, the study estimates 10% of task ...
Dell'Acqua et al. found that consultants using GPT-4 completed analytical tasks 25.1% faster with 40% higher quality for tasks inside the AI capability frontier. Roadmap Prioritisation contains analyt...
Deployment by Industry
Navigate competing priorities across engineering, design, sales, marketing, and executive leadership to build consensus on product direction and trade-off decisions.
Capability Evidence
Anthropic's study of real-world Claude usage across millions of professional conversations found that tasks related to Stakeholder Alignment & Buy-In represent a significant category of AI-augmented w...
The Anthropic Economic Index analysis of real-world Claude usage patterns found that tasks related to Stakeholder Alignment & Buy-In represent a meaningful share of professional LLM usage. The study i...
Dell'Acqua et al. found that consultants using GPT-4 completed analytical tasks 25.1% faster with 40% higher quality for tasks inside the AI capability frontier. Stakeholder Alignment & Buy-In contain...
Deployment by Industry
Translate business needs and user problems into detailed product requirements documents, functional specifications, and acceptance criteria that engineering can build against.
Capability Evidence
AI can automatically extract software requirements from raw specification documents
The Anthropic Economic Index analysis of real-world Claude usage patterns found that tasks related to Requirements & Specification Writing represent a meaningful share of professional LLM usage. The s...
Noy & Zhang found in a controlled experiment that AI assistance reduced professional writing task completion time by 40% and improved output quality by 18%. Tasks similar to Requirements & Specificati...
Deployment by Industry
Define success metrics and KPIs for product features and initiatives, set up tracking instrumentation, and monitor metric performance against targets.
Capability Evidence
The IMF finds that approximately 40% of global employment is exposed to AI, with up to 60% in advanced economies. For knowledge work tasks like Metric Definition & Tracking, the study estimates 27% of...
Tableau AI and Pulse enable natural language data querying and automated insight generation. For tasks like Metric Definition & Tracking, AI tools achieve approximately 48% quality on routine data exp...
Cognizant and Oxford Economics analysed 18,000+ tasks across industries and found that Gen AI will impact 90% of jobs but fully displace very few. For tasks like Metric Definition & Tracking, the stud...
Deployment by Industry
Analyse qualitative and quantitative user research — interviews, surveys, usability tests, behavioural data — and distil findings into actionable product insights.
Capability Evidence
Demonstrated autonomous scientific discovery and hypothesis generation in complex biological research indicates meaningful advancement in reasoning and analysis for domain-specific research applicatio...
AI systems can adapt research workflows and toolsets to changing scientific tasks
Automated verification of document-centric responses can assist researchers in validating citations and document references during research synthesis
Deployment by Industry
Write user stories with clear acceptance criteria, edge cases, and context that enable engineering teams to implement features without ambiguity.
Capability Evidence
The Anthropic Economic Index analysis of real-world Claude usage patterns found that tasks related to User Story Creation represent a meaningful share of professional LLM usage. The study indicates 51...
Noy & Zhang found in a controlled experiment that AI assistance reduced professional writing task completion time by 40% and improved output quality by 18%. Tasks similar to User Story Creation fall w...
Peng et al. found that developers using GitHub Copilot completed coding tasks 55.8% faster in a controlled experiment. For tasks like User Story Creation, AI coding assistants demonstrate 56% quality ...
Deployment by Industry
Facilitate sprint planning sessions, define sprint goals, negotiate scope with engineering leads, and ensure the team commits to a realistic and valuable set of work.
Capability Evidence
The Anthropic Economic Index analysis of real-world Claude usage patterns found that tasks related to Sprint Planning & Ceremony Facilitation represent a meaningful share of professional LLM usage. Th...
Dell'Acqua et al. found that consultants using GPT-4 completed analytical tasks 25.1% faster with 40% higher quality for tasks inside the AI capability frontier. Sprint Planning & Ceremony Facilitatio...
Peng et al. found that developers using GitHub Copilot completed coding tasks 55.8% faster in a controlled experiment. For tasks like Sprint Planning & Ceremony Facilitation, AI coding assistants demo...
Deployment by Industry
Define quarterly and annual product objectives and key results aligned to company strategy, negotiate targets with leadership, and track progress throughout the period.
Capability Evidence
MIT Sloan Management Review's annual survey of 3,000+ managers found that only 10% of organizations report significant financial value from AI deployment, despite widespread experimentation. For tasks...
Dell'Acqua et al. found that consultants using GPT-4 completed analytical tasks 25.1% faster with 40% higher quality for tasks inside the AI capability frontier. OKR & Goal Setting contains analytical...
OpenAI's o1 system card demonstrates significant advancement in complex reasoning tasks, achieving 83rd percentile on Codeforces and 93rd percentile on AMC math competitions. For analytical aspects of...
Deployment by Industry
Identify, track, and resolve dependencies between product, engineering, design, data, and infrastructure teams to prevent blockers and ensure coordinated delivery.
Capability Evidence
Extract and reconcile design data for nuclear engineering applications
The Stanford HAI AI Index Report 2025 documents AI systems achieving expert-level performance on graduate-level science questions and professional coding tasks. For tasks like Cross-Functional Depende...
Anthropic's study of real-world Claude usage across millions of professional conversations found that tasks related to Cross-Functional Dependency Management represent a significant category of AI-aug...
Deployment by Industry
Analyse product usage data, funnel metrics, retention cohorts, and feature adoption patterns to identify opportunities, diagnose problems, and validate hypotheses.
Capability Evidence
Self-service analytics agent enables autonomous query generation and analysis across distributed data sources, improving reasoning capability in data interpretation workflows.
The Stanford HAI AI Index Report 2025 documents AI systems achieving expert-level performance on graduate-level science questions and professional coding tasks. For tasks like Product Analytics Interp...
Anthropic's study of real-world Claude usage across millions of professional conversations found that tasks related to Product Analytics Interpretation represent a significant category of AI-augmented...
Deployment by Industry
Coordinate product launches across marketing, sales, support, and documentation teams — defining launch tiers, messaging, enablement materials, and rollout timelines.
Capability Evidence
Dell'Acqua et al. found that consultants using GPT-4 completed analytical tasks 25.1% faster with 40% higher quality for tasks inside the AI capability frontier. Go-to-Market Coordination contains ana...
Noy & Zhang found in a controlled experiment that AI assistance reduced professional writing task completion time by 40% and improved output quality by 18%. Tasks similar to Go-to-Market Coordination ...
Brynjolfsson, Li & Raymond found that AI assistance increased customer service worker productivity by 14% on average, with 34% gains for novice workers, in a study of 5,179 agents. For tasks like Go-t...
Deployment by Industry
Continuously refine the product backlog — re-prioritise items, split large epics into implementable stories, remove stale items, and ensure the top of the backlog is always ready for engineering.
Capability Evidence
The Anthropic Economic Index analysis of real-world Claude usage patterns found that tasks related to Backlog Grooming & Refinement represent a meaningful share of professional LLM usage. The study in...
Google Research found that LLMs show strong performance on isolated code generation but struggle with large-scale codebase understanding and architectural decisions. For Backlog Grooming & Refinement,...
Peng et al. found that developers using GitHub Copilot completed coding tasks 55.8% faster in a controlled experiment. For tasks like Backlog Grooming & Refinement, AI coding assistants demonstrate 54...
Deployment by Industry
Monitor competitor products, features, pricing, and positioning to identify market gaps, inform differentiation strategy, and anticipate competitive threats.
Capability Evidence
Perform geospatial analysis beyond vector-only limitations
The Stanford HAI AI Index Report 2025 documents AI systems achieving expert-level performance on graduate-level science questions and professional coding tasks. For tasks like Competitive Analysis, cu...
Dell'Acqua et al. found that consultants using GPT-4 completed analytical tasks 25.1% faster with 40% higher quality for tasks inside the AI capability frontier. Competitive Analysis contains analytic...
Deployment by Industry
Write release notes, changelog entries, and internal announcements that clearly communicate what shipped, why it matters, and what users or teams need to know.
Capability Evidence
Public release of agentic AI to general users demonstrates incremental improvement in computer use automation reaching broader adoption.
Anthropic's study of real-world Claude usage across millions of professional conversations found that tasks related to Release Communication represent a significant category of AI-augmented work. The ...
The Anthropic Economic Index analysis of real-world Claude usage patterns found that tasks related to Release Communication represent a meaningful share of professional LLM usage. The study indicates ...
Deployment by Industry
Evaluate whether to build capabilities in-house, integrate third-party tools, or partner — weighing cost, time-to-market, strategic control, and long-term maintenance burden.
Capability Evidence
The Stanford HAI AI Index Report 2025 documents AI systems achieving expert-level performance on graduate-level science questions and professional coding tasks. For tasks like Build vs Buy Decisions, ...
The Anthropic Economic Index analysis of real-world Claude usage patterns found that tasks related to Build vs Buy Decisions represent a meaningful share of professional LLM usage. The study indicates...
MIT Sloan Management Review's annual survey of 3,000+ managers found that only 10% of organizations report significant financial value from AI deployment, despite widespread experimentation. For tasks...
Deployment by Industry
Collect, categorise, and prioritise customer feedback from support tickets, sales calls, NPS surveys, and user interviews to inform product decisions.
Capability Evidence
LLM-based code generation agents can perform multi-round debugging of GUI code when provided with visual feedback (screenshots), achieving more reliable results than with text-only feedback
Anthropic's study of real-world Claude usage across millions of professional conversations found that tasks related to Customer Feedback Triage represent a significant category of AI-augmented work. T...
MIT Sloan Management Review's annual survey of 3,000+ managers found that only 10% of organizations report significant financial value from AI deployment, despite widespread experimentation. For tasks...
Deployment by Industry
Work with engineering to assess technical complexity, architectural implications, and implementation risks of proposed features before committing them to the roadmap.
Capability Evidence
The Anthropic Economic Index analysis of real-world Claude usage patterns found that tasks related to Technical Feasibility Assessment represent a meaningful share of professional LLM usage. The study...
Dell'Acqua et al. found that consultants using GPT-4 completed analytical tasks 25.1% faster with 40% higher quality for tasks inside the AI capability frontier. Technical Feasibility Assessment conta...
AI can accelerate technical support functions, enabling teams to handle more work without additional staff
Deployment by Industry
Define product tiers, packaging structure, and pricing strategy based on value analysis, competitive positioning, and willingness-to-pay research.
Capability Evidence
The Stanford HAI AI Index Report 2025 documents AI systems achieving expert-level performance on graduate-level science questions and professional coding tasks. For tasks like Pricing & Packaging Deci...
Dell'Acqua et al. found that consultants using GPT-4 completed analytical tasks 25.1% faster with 40% higher quality for tasks inside the AI capability frontier. Pricing & Packaging Decisions contains...
On the MMLU benchmark, which tests knowledge across 57 professional and academic domains, frontier AI models achieve 85-90%+ accuracy. For knowledge-retrieval aspects of Pricing & Packaging Decisions,...
Deployment by Industry
Coach junior product managers, provide feedback on their work, and help develop product thinking skills across the organisation.
Capability Evidence
The IMF finds that approximately 40% of global employment is exposed to AI, with up to 60% in advanced economies. For knowledge work tasks like Team Mentoring & Development, the study estimates 5% of ...
AI can generate design elements for web development
The WEF Future of Jobs Report 2025 projects that employers expect 83 million jobs displaced and 69 million created by 2030, with analytical thinking and creative thinking remaining the most valued hum...
Deployment by Industry
Design and run product experiments — A/B tests, feature flags, beta programmes — to validate hypotheses with data before committing to full rollouts.
Capability Evidence
The Stanford HAI AI Index Report 2025 documents AI systems achieving expert-level performance on graduate-level science questions and professional coding tasks. For tasks like Product Experimentation ...
Anthropic's study of real-world Claude usage across millions of professional conversations found that tasks related to Product Experimentation & A/B Testing represent a significant category of AI-augm...
The Anthropic Economic Index analysis of real-world Claude usage patterns found that tasks related to Product Experimentation & A/B Testing represent a meaningful share of professional LLM usage. The ...
Deployment by Industry