Welcome

The Agentic PM

Product Management in the Age of AI

Asaf Nakash  |  Principal Product Manager, Microsoft Defender

Use → to navigate  |  Press Esc for overview

2026

A PM wrote a spec last week.
An AI agent wrote a better one in 4 minutes.

So what does a product manager actually do now?

PM
Part 1

The Foundations

What product management is — and why it matters more than ever

Let's Start With You

What do you think a
Product Manager does?

A
Manages a team of engineers
B
Decides what to build and why
C
Designs the user interface
D
Tells an AI agent what to build
B is closest — for now. A PM doesn't manage people or design UI directly — they figure out what problems are worth solving and why. But D is where things are heading fast.
The Role

So what is a Product Manager?

"A product manager is the person who identifies the customer need and the larger business objectives that a product or feature will fulfill, articulates what success looks like, and rallies a team to turn that vision into reality."

— Martin Eriksson, founder of ProductTank

Think of it as the CEO of the product — without the authority to tell anyone what to do.

The Sweet Spot

Where PM Lives

User
Desirability

What do people actually need? What problems keep them up at night?

Technology
Feasibility

Can we build it? What are the technical constraints and trade-offs?

Business
Viability

Does it make sense for the business? Will it drive growth or revenue?

  PM sits at the intersection of all three
Why It Matters

Why do we need
Product Managers?

Direction

Without a PM, teams build features nobody asked for. PMs make sure we solve the right problems.

Alignment

Engineering, design, marketing, sales — they all have different goals. PMs keep everyone rowing in the same direction.

Prioritization

There are always more ideas than time. PMs decide what to do first and what to say no to.

Building Products

The Classic Product Cycle

Discover

Talk to users.
Understand their
problems deeply.

Define

Pick the problem
worth solving.
Set clear goals.

Design

Sketch solutions.
Create prototypes.
Test with users.

Develop

Engineers build it.
Start small, iterate
quickly.

Deliver

Ship to users.
Measure results.
Learn & repeat.

This cycle hasn't changed. But who does each step is changing fast.

Start Small

The MVP Mindset

Minimum Viable Product

Not an MVP

  • A half-finished product
  • Something ugly & broken
  • A prototype with no users
  • A checklist of features
VS

A Real MVP

  • Solves ONE core problem well
  • Real users can try it today
  • Generates real feedback
  • Teaches you what to do next
Feedback Loops

The Engine of
Great Products

Hypothesize

"We believe X will
solve problem Y"

Build

Create the smallest
thing to test it

Measure

Collect data from
real users

Learn

Pivot or persevere
based on evidence

  Then repeat — every cycle gets you closer to product-market fit
AI
Part 2

The Shift

Everything you just learned is being rewritten — right now

Here's what changed

AI Can Build

An AI agent can scaffold a feature in an afternoon. The "develop" step just got 10x faster.

AI Can Research

Analyzing 10,000 support tickets used to take weeks. AI does it in minutes and surfaces patterns humans miss.

AI Can Design

Generate 20 UI variations from a text description. Prototype in hours, not sprints.

So if AI can do all of this... what's left for the PM?

Evolution

The PM Role is Evolving

2010s

Traditional PM

Write detailed specs. Long release cycles. Waterfall to Agile transition. Manual analysis of user data. The PM owned the roadmap doc.

Now

AI-Augmented PM

AI helps summarize user research, draft specs, analyze data patterns, and generate prototypes in minutes. PM still drives decisions.

Emerging

Agentic PM

AI agents autonomously run experiments, analyze results, and propose product changes. PM becomes the strategic orchestrator.

Agentic Teams

What are Agentic Teams?

Traditional Team

Humans do all the work — research, design, code, test, ship. Bottleneck is always people's time.

Agentic Team

AI agents handle routine execution — coding, testing, data analysis. Humans focus on strategy, creativity, and judgment.

"The best PMs of the future won't write more specs — they'll orchestrate AI agents, ask better questions, and make the judgment calls machines can't."

Real World

What this looks like
in practice

 Before (2023)

PM writes a 15-page spec over 2 weeks. Reviews with 3 teams. Engineering estimates 6 sprints. Ships in 3 months.

 Now (2026)

PM describes the problem to an AI agent. Gets a draft spec + prototype in a day. Tests with users by Thursday. Ships validated in 2 weeks.

The cycle didn't change. The speed changed. And that changes everything about what PMs should spend time on.

The New Job

What an Agentic PM
actually does

Problem Selection

AI surfaces options. You decide what's worth solving.

Agent Orchestration

Direct agents to research, prototype, test. Then review critically.

Judgment Calls

AI gives you 10 options. Knowing which one to ship is a human skill.

Guardrails

Keep what gets built safe, ethical, and mission-aligned.

Stakeholder Trust

AI drafts the strategy. You're the one who sells it.

The Trap

The biggest mistake PMs
make with AI

"We can ship 10x faster now!"

— Every PM after discovering AI coding agents

Faster shipping without better problem selection just means you build the wrong thing 10x faster.

Speed is the easy part. Knowing what to build is the hard part.

The Human Edge

What can't AI replace?

Empathy

Truly understanding human pain and joy.

Vision

Imagining a future that doesn't exist yet.

Ethics

Should we build it? That question belongs to humans.

Trust

Building real relationships with teams and users.

Judgment

Making tough calls with incomplete information.

AI is a power tool, not a replacement. The best PMs will be those who learn to wield it.

Framework

The Agentic PM Framework

Orchestrate

Direct AI agents
with clear intent.
Set the mission.

Review

Treat AI output
as untrusted.
Verify everything.

Decide

Make the calls
AI can't make.
Own the outcome.

Ship

Get it to users
fast. Measure
real impact.

Learn

Feed results back.
Improve the agents.
Improve yourself.

Superpowers

AI Superpowers for PMs

Research at Scale

Analyze thousands of tickets and reviews in seconds to surface patterns humans miss.

Rapid Prototyping

Generate working prototypes from descriptions. Test 5 ideas in the time it took to debate 1.

Predictive Analytics

Predict which features drive retention before you build them. Fail cheaper, learn faster.

Always-On Feedback

AI agents continuously monitor user behavior and surface opportunities. Never stops.

Quick Check

In 5 years, the most valuable
PM skill will be...

A
Writing detailed specs and PRDs
B
Orchestrating AI agents effectively
C
Deep customer empathy and judgment
D
Coding and technical depth
B and C together. Specs will be auto-generated. Coding is increasingly commoditized. But knowing which problem matters (C) and how to get machines to solve it well (B) — that combination is the new PM superpower. The PMs who master both will be unstoppable.
Recap

Key Takeaways

PM = Problem Finder

Not a feature factory. PMs find the right problems — more important now than ever.

Build → Measure → Learn

Same loop, 10x faster. AI compresses cycles from months to days.

Orchestrate, Don't Abdicate

AI agents are powerful, but the PM still owns the outcome. Trust nothing blindly.

The Best Time is Now

Every PM who learns AI agents today has a 2-year head start. Start experimenting.

Keep Learning

Want to dive deeper?

"Inspired"

Marty Cagan · The PM bible

"The Lean Startup"

Eric Ries · Build, measure, learn

Lenny's Podcast

Top PMs sharing what actually works

Build with AI agents

Ship something small. Best way to learn.

"Agentic Engineering"

Andrej Karpathy · How builders work with AI

The future of PM isn't
AI vs. humans

It's humans who know how to use AI vs. humans who don't.

Questions? Let's chat!

Asaf Nakash  |  LinkedIn  |  X  |  GitHub

Press Esc for slide overview