Ranking 15 PM Skills: What Survives vs. Gets Disrupted by AI | Nan Yu (Linear)

YouTube URL: https://www.youtube.com/watch?v=Dc2lefSwds4

Host(s): Peter Yang (Behind the Craft)

Guest(s): Nan Yu, Head of Product at Linear

Podcast/Video Overview

Overall Summary

This conversation maps the future of product management in the AI era. Nan Yu explains which PM skills stay valuable, which get automated, and which new abilities matter most. The core message is clear: emotional and taste-driven skills will differentiate PMs, while logic-heavy and mechanical tasks become AI-assisted or automated.

You will also see practical demos. Nan shows how Linear uses AI agents and Claude’s MCP to analyze customer feedback, triage issues, find duplicates, and even write code and ship PRs. The episode blends strategy with hands-on workflows you can use today. If you lead product teams, this is a playbook for doing more with fewer PMs while raising the quality bar.

References

• Product taste: The intuitive sense of what feels good, coherent, and delightful in a product.
• 2x2s: A simple decision matrix. You plot options across two axes to compare tradeoffs.
• Golden path: The main, happy-path flow that most users take.
• Triage: The intake and sorting process for new bugs, requests, and tasks.
• IC: Individual contributor. Non-manager role focused on hands-on work.
• PRD: Product Requirements Document. A structured spec for what to build and why.
• OKRs: Objectives and Key Results. A goal-setting framework.
• Evals: Evaluation tests and datasets to measure AI or agent performance.
• System prompt: The hidden instruction that guides how an AI model responds.
• MCP (Model Context Protocol): A standard that lets tools and services plug into AIs (e.g., Claude) to read/write context and perform actions.
• .ics: A calendar file format used for sharing events or subscriptions.
• Vertical rhythm: Consistent spacing and alignment in UI designs for readability.
• Hot code path: A part of code that runs frequently and must be efficient.
• “Ragged edge”: Progress that is uneven across tasks; AI is great at some things, weak at others.
• Context engineering: Curating the right data, instructions, and signals that AIs or agents need to perform well.
• Agent management: Delegating, supervising, and evaluating AI agents in day-to-day workflows.

Key Topics Covered

What PM Skills Still Matter in the AI Era

Under each topic, provide a 100 word summary of the discussion.

• Main Point: Taste, branding, ownership, stakeholder management, and EQ remain core.
• Core Argument: These skills are rooted in emotion, narrative, and judgment, which AI still struggles to replicate.
• Quotes:“Say what you want about AI. They’re not really quite good at the emotional stuff yet.”“You have to care…and be ready for people to look to you for the result.”

Summary: Nan lists the enduring skills that separate strong PMs: product taste, branding (product and personal), ownership with risk appetite, stakeholder management, and emotional intelligence. These skills involve intuition, empathy, and storytelling. They shape how customers feel and how teams align, which AI cannot yet fully replace. Taste helps PMs sense delight or friction before they can explain why. Ownership means embracing risk and standing behind outcomes. Branding frames perception. Stakeholder management and EQ help you earn trust, coordinate resources, and build momentum. These are the human multipliers that compound over time.

Skills Being Disrupted: Strategy and Prioritization

• Main Point: AI will level the playing field on strategy mechanics.
• Core Argument: Logic-heavy analysis, frameworks, and 2x2s are exactly what AI does fast and consistently.
• Quotes:“A lot of strategic thinking…that is something that they are excellent at.”“It’ll be rigorous. It’ll be honest.”

Summary: PMs have long differentiated themselves with frameworks, prioritization rubrics, and strategy memos. AI now helps everyone run this process at speed. You can load docs into a model, ask for scenarios, and refine tradeoffs rapidly. This does not remove your job. It removes the edge you got from manual rigor. Your value shifts toward judgment, narrative, and decision quality, not brute-force analysis. The trope that “AI will automate everyone except my most important work” is wrong. If it’s logical, repeatable, and well-documented, AI will do it, too.

Skills Being Disrupted: Data Analysis and Synthesis

• Main Point: SQL and first-pass insight work are commoditized.
• Core Argument: AI can write queries, scan dashboards, and surface patterns with minimal setup.
• Quotes:“That difference…‘knows SQL’ vs. ‘needs an analyst’…has evaporated.”“I never need to learn regex again.”

Summary: The gap between PMs who can query data and those who can’t is shrinking. Give AI your schema and goals, and it will draft queries, visualizations, and summaries. PMs should spend more time on interpretation and decisions rather than plumbing data. This shift pushes PMs to focus on problem framing, contextual insight, and action. You gain throughput without learning every technical skill. The value now comes from asking better questions and knowing when an answer is good enough to move.

Skills Being Disrupted: Market Research

• Main Point: Desk research is largely automated; high-touch research remains.
• Core Argument: AI excels at broad scans and synthesis; unique insights come from primary, relationship-driven work.
• Quotes:“The low-hanging stuff has been picked.”“If you’re having dinners with executives…that’s still alive.”

Summary: AI can do comprehensive desk research in minutes. It scans reports, competitor sites, and user threads, and returns a coherent brief. The surviving edge is first-party, high-touch research—interviews, advisory councils, and deep domain relationships. That is where you learn motivations, risks, and emotion. Use AI to remove the cost of baseline research. Spend your time building networks, running targeted studies, and validating unique angles. That is where market insight compounds.

Skills Being Disrupted: Project Management and Documentation

• Main Point: Detail diligence and upkeep get automated.
• Core Argument: AI is tireless at reading everything, spotting changes, and keeping docs synced.
• Quotes:“Computers are super duper diligent…always online.”“We didn’t want to hire people…because this stuff is kind of free.”

Summary: The most praised project managers are “machines” who notice every detail. AI now does this at scale. It reads every doc, flags regressions, updates summaries, and prompts people for missing pieces. Linear’s vision assumes doc upkeep and backlog grooming are increasingly AI-first. You still set standards and resolve ambiguity. But you no longer need to hire people to chase status or collate updates. That budget shifts to higher-value roles. The PM’s responsibility rises to orchestration and judgment.

New Skills to Build: Context Engineering and Workflow Design

• Main Point: Feed AIs the right context at the right time.
• Core Argument: Performance depends on relevant inputs, clear instructions, and a process to fetch missing context.
• Quotes:“What information would this model need to have better performance?”“How does it know it needs to gather more?”

Summary: Great outputs start with great inputs. Context engineering means curating the data, instructions, and constraints that raise AI performance. Workflow design turns this into a repeatable system. You decide when and how agents fetch context, what tools they can use, and how they escalate gaps. This is more than prompt engineering. It is an operational layer across your stack. As agents become co-workers, you must design a pipeline that keeps them effective and safe.

New Skills to Build: Agent Management and Evals

• Main Point: Learn how much to specify and how to measure quality.
• Core Argument: Agents vary. Some need detailed instructions; others can run with broad intent. Evals prove what works.
• Quotes:“How under or over specified do you need to make your instructions?”“We start with no evals and build them from real feedback.”

Summary: Treat agents like junior teammates. Some are self-starters. Others need step-by-step. Your job is to set the right level of specificity, then measure outcomes. Nan suggests starting with minimal evals and building them from real incidents and feedback. This keeps you from over-engineering tests early and lets you improve what actually matters. The bar is reliability, not clever prompts. You should also decide who owns outcomes when agents act.

Craft vs. Speed: Short Loops Beat Big-Bang Launches

• Main Point: Ship a working version in 10% of the time, then iterate.
• Core Argument: Software can deploy continuously. You learn faster by shortening feedback loops.
• Quotes:“By the time 10% has elapsed you should have something that gets the job done.”“We use a Russian doll release process.”

Summary: Apple-style polish works for hardware. Software needs speed. Nan’s rule: in the first 10% of time, ship a working core experience. Then refine through constant usage and feedback. Linear scales this with a “Russian doll” release: devs first, internal, select beta, public beta, pre-release, and GA. Each layer hardens the product. This reduces launch risk and accelerates learning. It also builds confidence across engineering, design, and support.

Keep Products Simple: Opinionated Defaults, Flexible Where It Counts

• Main Point: Be opinionated at the workflow level; flexible at higher strategy layers.
• Core Argument: If you make everything configurable, admins become accidental product designers.
• Quotes:“As it gets more specific into individual actions…that’s where we’re more opinionated.”“ICs want a predictable experience.”

Summary: Linear targets the IC experience. The product is flexible for strategy choices (OKRs vs. initiatives), but opinionated for day-to-day workflows. This keeps the experience predictable, fast, and low-friction for the people who use it most. Over-configuring pushes design decisions to admins and leads to inconsistent UX across teams. A better approach: sane defaults, crisp workflows, and targeted flexibility where businesses truly differ.

Write for AI First Readers

• Main Point: Your main reader is often an AI, then a human.
• Core Argument: AI will route, enrich, and act on your docs. Include details it needs to trigger automations.
• Quotes:“Everything you write… the main audience is not really people anymore.”“If I don’t put enough detail then all the automations won’t work.”

Summary: Linear’s AI reads issues before people do. It links related work, suggests assignees, and starts workflows. That means your issues and docs should include repro steps, intent, and context. Humans benefit too, because the same details improve clarity. Keep main docs concise (e.g., half a page) and move depth to appendices. Assume someone will use AI to summarize your writing. Make that summary accurate by writing with AI in mind.

Live Demos: Claude MCP, Product Intelligence, and Coding Agents

• Main Point: AI can analyze feedback, triage, find duplicates, read code, and ship PRs.
• Core Argument: With MCP and agents, PMs can scale work without adding headcount.
• Quotes:“You can just spawn off a hundred of these simultaneously.”“Some human being has responsibility for the outcome.”

Summary: Nan demos three workflows. First, Claude + MCP pulls Linear issues and customer requests, then summarizes themes. Second, Product Intelligence suggests teams, assignees, tags, and duplicates during triage, with explainer “why” notes. Third, coding agents like Charlie and Codegen read the codebase to confirm behavior and even implement features via PRs. You can assign agents issues in bulk while a human remains responsible for outcomes. It’s a glimpse of an operating system where humans and agents collaborate inside your product stack.

Key Themes and Insights

List down all themes discussed in the Video

Emotion-led skills are your edge

These include product taste, EQ, and brand. They shape perception and adoption. Use them to frame value before features exist.Quotes:

• “You’re building the emotional resonance before you even have a value offering.”
• “You’ll have the feeling before you know why.”

AI levels logic-heavy work

Strategy papers, 2x2s, and first-pass research are now AI-accelerated for everyone. Differentiate on judgment and narrative.Quotes:

• “It’ll do the mental process you wish you could do…really fast.”
• “There’s no physics about why [AI] can’t touch the important parts.”

Write for AI-first workflows

Docs fuel automations. If they lack details, the system breaks. Treat AI as the first reader.Quotes:

• “The main audience is not really people anymore.”
• “If I don’t put enough detail then all the automations won’t work.”

Short feedback loops beat polish

Ship in 10% of the time, then iterate through concentric releases.Quotes:

• “Get as many feedback loops as possible.”
• “Russian doll release process.”

Build context and workflows for agents

Agents need the right inputs and tools. Design how they fetch missing context and when to escalate.Quotes:

• “What information would this model need?”
• “How does it know it needs to gather more?”

Human accountability for agent work

Keep a named owner for every agent-driven task. Celebrate wins and own failures.Quotes:

• “Some human being has responsibility for the outcome.”
• “You can’t just be like, ‘the robot did it.’”

Opinionated where it matters

Give ICs fast, consistent workflows. Offer flexibility at higher levels, not in every UI toggle.Quotes:

• “As it gets more specific…that’s where we’re going to put a lot more guidance.”
• “ICs want a predictable experience.”

Actionable Advice and Takeaways

Immediate Actions You Can Take:

• Audit your PM tasks: mark what’s emotional vs. logical. Automate the logical with AI tools.
• Rewrite one PRD for AI-first reading. Add context, repro steps, and expected outcomes.
• Set up Claude MCP (or similar) to analyze customer feedback in your tracker.
• Pilot an AI triage workflow to suggest assignees, teams, and duplicates.
• Delegate one low-risk feature to a coding agent. Keep a human owner.

Long-term Strategies:

• Invest in product taste. Review products weekly and explain what feels right or wrong.
• Build an agent management playbook: context patterns, escalation rules, and evals from real incidents.
• Adopt a Russian doll release cadence to harden features through layers of users.
• Be opinionated in IC workflows; stay flexible in strategy layers.

Questions for Reflection:

• Which parts of my job are actually emotional, not analytical?
• If AI did 80% of my analysis, how would I raise the bar on decisions?
• Where do my documents fail AI readers today?
• What evals matter for my agents in production?

Noteworthy Observations and Unique Perspective

List down any observations that is important for an entrepreneur or business person and Unique perspective shared by the guys. List down the quote that provides context

Hire for the future, not to maintain docs

• Linear avoided hiring PMs for work they expected AI to automate.
• Quotes:“We saw that type of work being largely overtaken by AI.”“We didn’t want to be stuck…now this stuff is kind of free.”

Docs are operational primitives now

• Writing is for AI routing and action first, humans second.
• Quotes:“The main audience is not really people anymore.”“If I don’t put enough detail then all the automations won’t work.”

Backlog pruning is an AI job

• AI can find duplicates, suggest owners, and merge context.
• Quotes:“That’s probably the main one…mark this as a duplicate.”“Why was [person] suggested? He’s the lead…and assignee on related issues.”

Agents can read code so engineers don’t have to stop

• Ask an agent to confirm behavior from the codebase before pinging engineers.
• Quotes:“You don’t have to bother an engineer…you can just ask an agent.”“It’s a hot code path. We’re good.”

Parallelize work with agents at scale

• Bulk-assign tickets to coding agents and supervise outcomes.
• Quotes:“You can spawn off a hundred of these simultaneously.”“Some human being has responsibility for the outcome.”

Companies, Tool and Entities Mentioned

• Linear
• OpenAI, Ramp, Mercury (Linear customers)
• Anthropic Claude (via MCP)
• MCP (Model Context Protocol)
• Codegen (coding agent provider)
• Cursor (AI coding environment)
• Git/version control and .ics calendar format
• “Clue/Cluey” (example of brand-led growth)
• Behind the Craft (podcast)

Final Thoughts

AI is reshaping product management. The parts driven by logic and diligence will become AI-first. Your edge will be product taste, EQ, narrative, and judgment. The best PMs will orchestrate humans and agents, write AI-ready docs, and design workflows where context flows to the right place at the right time. Start small: automate triage, run MCP research, and delegate a safe feature to an agent. Keep a human owner. Learn fast through short loops. The future PM is less paper-pusher, more creative director of systems.

Linkedin Ideas

The PM Skills AI Won’t Disrupt

• Core Argument: Emotional skills like product taste and EQ will define PM excellence.
• Key Point: Logic-heavy work is now a level playing field with AI.
• Quotes: “They’re not really quite good at the emotional stuff yet.” “You’ll have the feeling before you know why.”

Write for AI First, Humans Second

• Core Argument: AI reads your docs before people do; structure them to trigger automations.
• Key Point: Add repro steps, context, and crisp outcomes.
• Quotes: “The main audience is not really people anymore.” “If I don’t put enough detail then all the automations won’t work.”

From Strategy Hero to Orchestrator

• Core Argument: AI handles the frameworks; PMs must focus on judgment and narrative.
• Key Point: Use AI to draft; you decide what matters and why.
• Quotes: “It’ll do the mental process you wish you could do…really fast.”

Short Loops > Big Bangs

• Core Argument: Ship a working version in 10% of the time and iterate.
• Key Point: Concentric releases harden quality without drama.
• Quotes: “Get as many feedback loops as possible.” “Russian doll release process.”

Agents in Production Need Human Owners

• Core Argument: Keep a named human responsible for agent outcomes.
• Key Point: Delegate widely, but never abdicate accountability.
• Quotes: “Some human being has responsibility for the outcome.”

Blog Ideas

How to Build an AI-Ready PM Org

• Core Argument(s): Automate logic work; upskill on taste and EQ; design agent workflows.
• Key Themes to explore: Context engineering, evals, triage automation, accountability.
• Key Point (s): Docs for AI first; human owners for agent tasks.
• Quotes: “Everything you write…the main audience is not really people anymore.”

Product Taste in the Age of Claude

• Core Argument(s): Taste is still a moat; AI lifts analysis but not emotion.
• Key Themes to explore: How to practice taste; pairing taste with AI rigor.
• Key Point (s): Use AI for breadth; apply taste for depth.
• Quotes: “You’ll have the feeling before you know why.”

The Russian Doll Release Playbook

• Core Argument(s): Short loops beat polish-heavy cycles in software.
• Key Themes to explore: Dev-only, internal, select beta, public beta, GA.
• Key Point (s): Risk reduction and faster learning.
• Quotes: “By the time 10% has elapsed you should have something that gets the job done.”

Taming the Backlog with AI

• Core Argument(s): Let AI find duplicates, assign owners, and merge context.
• Key Themes to explore: Product Intelligence, MCP research, agent triage.
• Key Point (s): Free PM time for decisions, not diligence.
• Quotes: “That’s probably the main one…mark this as a duplicate.”

Agent Management 101 for PMs

• Core Argument(s): Agents are teammates; give them context, tools, and clear owners.
• Key Themes to explore: Under/over-specifying, evals from real feedback, escalation rules.
• Key Point (s): Parallelize with agents; maintain accountability.
• Quotes: “How under or over specified do you need to make your instructions?”