Two Years Ago I Said PM Skills Are 'Find the Right Problem, Design the Right Solution, Make Change Happen.' Has AI Changed That?
Two years ago I made a video about what product managers actually do. I said the core of PM work isn’t meetings, writing docs, or arguing with engineers — that’s all surface. The real value comes down to three things:
Find the right problem. Design the right solution. Make change happen.
Two years later, AI has gone from a novelty to a default tool on everyone’s desk. When ChatGPT first launched, people were amazed it could write poetry. Now AI can help you run user research, draft PRDs, generate prototypes, analyze data, and write production code.
So people keep asking me: Are PMs going to be replaced by AI?
My answer: No. But if you’re still working the way you did two years ago, you’ll be replaced by a PM who knows how to use AI.
The core framework hasn’t changed — it’s still find the right problem, design the right solution, make change happen. But what each layer means has shifted fundamentally. And there’s an entirely new dimension that didn’t exist two years ago: Can you build an AI-powered system around yourself and become a “super-individual”?
Let me break it down.
Finding the Right Problem: From Gut Feel to Human-AI Perception
What I said two years ago
Finding the right problem means understanding business objectives and breaking them down layer by layer. If Xiaohongshu wants profitability, decompose that into user growth, then into content creator volume, community engagement quality, and so on. You need data literacy and user empathy — the ability to read what’s hiding behind the numbers.
All of that is still true. But if you’re still doing it manually, you’re already slow.
What’s changed
AI has fundamentally shifted “finding the right problem” in two ways.
First, the ceiling on information gathering has been shattered.
Two years ago, a competitive analysis meant a full day of downloading competitor apps, testing features, taking screenshots, writing docs. Now you can have AI scrape public competitor data, summarize top pain points from user reviews, compare feature sets, and generate a structured report — in about 30 minutes.
Second, the dimensions of what you can “see” have expanded.
You used to analyze user behavior through dashboards and interviews. Now you can feed thousands of user reviews to AI for topic clustering, sentiment analysis, and discovery of long-tail needs you might have missed entirely. AI doesn’t replace your judgment, but it lets you see things you couldn’t before.
The new skill requirement
| Good enough two years ago | Must add now |
|---|---|
| Read data dashboards | Use AI for deep analysis and non-obvious trend detection |
| Conduct user interviews | Use AI to process qualitative data at scale (reviews, tickets, feedback) |
| Do competitive analysis | Build an “AI radar” that continuously monitors market signals |
| Have business intuition | Have systems thinking — don’t just watch yourself, build a system that watches for you |
The key shift: from “I go find problems” to “I build a system that surfaces problems for me.”
Here’s a concrete example: set up a simple automation where AI scrapes competitor app store reviews, social media discussions, and industry news daily, then summarizes it into a “daily market pulse” report pushed to you each morning. Five minutes of scanning versus an hour of manual research — that’s a 10x improvement at minimum.
But here’s the critical “but” —
AI helps you see more. But the judgment of what constitutes a good problem still depends entirely on you. More information actually makes it easier to get lost. The ability to identify the one problem worth committing resources to, amid all that noise — that skill is rarer and more valuable than it was two years ago.
Designing the Right Solution: From “I Think of Solutions” to “I Curate and Validate Solutions”
What I said two years ago
Designing good solutions requires data thinking, user thinking, and competitive research skills — the ability to evaluate many options and pick the one that fits your product and your users.
That’s still foundational. But the production process for solutions has been completely rewritten.
What’s changed
First, the cost of generating solutions has dropped to near zero.
Two years ago, coming up with three options meant multiple brainstorming sessions, several wireframe iterations, and a stack of documents. Now you can give AI a requirements description and get 10 proposals in 5 minutes — conservative to radical, tech-driven to ops-driven, every angle covered.
What does this mean? “Coming up with solutions” is no longer the scarce skill. “Evaluating and selecting solutions” is.
You used to be valuable because you could think of what others couldn’t. Now AI can generate 100 options. You’re valuable because you can pick the one that fits this stage, these resources, these users. That’s taste. That’s judgment. That’s experience. AI can’t replicate those.
Second, prototyping and validation speed has compressed by 10x.
Two years ago, you had an idea, drew wireframes, synced with design, waited for high-fidelity mockups, queued for engineering — maybe one or two months before an experiment went live. Now you can use AI tools (Cursor, v0.dev, Bolt) to generate an interactive prototype, even a working MVP. The time from “I have an idea” to “users can try it” has shrunk from months to days, sometimes hours.
This isn’t a nice-to-have improvement. This is a paradigm shift in how PMs work. When the cost of validating an idea is low enough, you can run more experiments, fail faster, and find product-market fit sooner.
Third, PMs can now cross boundaries.
Two years ago, there was an invisible wall between non-technical PMs and engineering. You could only write PRDs and wait. Now you can use AI-assisted coding to write a data analysis script, build an internal tool, or even develop a simple product prototype.
This doesn’t mean replacing engineers. It means doors that were closed because “I don’t know how to code” are now open.
The new skill requirement
| Good enough two years ago | Must add now |
|---|---|
| Come up with good solutions | ”Orchestrate” AI to generate many options, then make precise selections |
| Draw wireframes | Use AI tools to rapidly generate interactive prototypes |
| Write PRDs | Write prompts — give AI precise instructions to produce the output you need |
| Depend on engineering to validate ideas | Build quick MVPs yourself with AI for validation |
| Understand UX design | Understand “probabilistic interaction” — AI outputs vary, how do you design for graceful uncertainty? |
The key shift: from “creator of solutions” to “curator of solutions + rapid validator.”
One trap to watch for: because AI makes generating solutions so easy, some PMs fall into “solution inflation” — producing lots of complete-looking proposals that lack deep thinking. AI makes 60-point solutions effortless. But the road from 60 to 90 still requires your judgment, your empathy for users, and your understanding of business logic.
Making Change Happen: From “Pushing Projects” to “Human + AI System-Driven”
What I said two years ago
Making change happen requires communication skills and execution ability — working with engineering, design, QA, and operations to push projects forward.
These “soft skills” haven’t become obsolete. In the AI era, they’re more important than ever.
What’s changed
First, AI accelerates execution, but “people” are still the bottleneck.
AI can help write code, generate design assets, automate testing. Execution speed has genuinely improved. But whether a project moves forward has never been about “how fast the code gets written” — it’s about “whether people can reach consensus.”
You need to convince your manager the direction is worth investing in. You need engineers to understand why this requirement has the highest priority. You need to coordinate the rhythm between designers and developers. You need to make tradeoffs when the project hits roadblocks. AI can’t do any of that for you.
Second, the boundaries of “making change happen” have expanded.
A PM’s project scope used to be limited by team resources. You had X engineers, so you could do X amount of work. Now one person plus AI can accomplish what used to require a small team.
What does this mean? PMs can become genuine “super-individuals.” You don’t have to wait for company resources. You can use AI to build a side project, validate an idea, accumulate a portfolio — or even create a real product.
“Making change happen” is no longer limited to “pushing projects inside a company.” It extends to “using AI to empower yourself so that any idea can potentially become reality.”
Third, communication skills have a new dimension.
Beyond communicating with people, you now need to learn to “communicate with AI.” This isn’t a joke — Prompt Engineering is fundamentally a communication skill. Can you describe your needs in clear, precise language to get the output you want? Can you set appropriate roles, constraints, and evaluation criteria for AI?
And as more team members start using AI, whether you can help the team establish shared AI workflows and standards becomes a new PM responsibility. Which steps benefit from AI acceleration? Which require human oversight? How do you ensure output quality? Someone needs to think through and drive all of this — and that someone should be the PM.
The new skill requirement
| Good enough two years ago | Must add now |
|---|---|
| Communicate with people (listen, articulate) | Communicate with AI (Prompt Engineering, instruction design) |
| Coordinate teams to push projects | Build “human + AI” collaborative workflows |
| Drive change within the company | Independently ship side projects with AI, validate ideas |
| English communication skills | English still important + AI can help bridge language gaps |
| Design experiments to test hypotheses | Validation speed exponentially faster — run a complete experiment loop in a weekend |
The key shift: from “project driver” to “system builder” — you’re not just pushing one project, you’re building a system that keeps you consistently effective.
The New Dimension That Didn’t Exist Two Years Ago: Build Your AI System
This is the point I most want to make today.
Two years ago, a PM’s capability model was: the stronger you are, the bigger the things you can do.
Now the model is: the stronger the system you build, the bigger the things you can do.
What do I mean?
You used to do user research by scheduling interviews yourself, conducting them yourself, organizing notes yourself. Now you can build a system: AI automatically pulls user voices from support tickets and social media → AI does sentiment analysis and topic clustering → generates a weekly summary pushed to you → you spend just 20% of your time on the highest-value question: “what do these signals mean?”
You used to write PRDs start to finish. Now you build a system: give AI your product context, user personas, and technical constraints as a “knowledge base.” Each time a new requirement comes in, AI generates a PRD draft based on that context. You refine, add judgment, fill in edge cases. The efficiency gain isn’t linear — it’s exponential.
You used to track project progress through daily standups and endless meetings. Now AI can aggregate status from Jira, Slack, and GitHub into a daily “project health report.” Your attention is no longer wasted on collecting information — it’s focused on making decisions.
This is what I mean by “building an AI system around yourself.” At its core, you’re doing one thing: freeing your time and attention from low-value “information collection and processing” to focus on high-value “judgment and decision-making.”
Lots of people use AI tools. But very few can orchestrate AI into a system that continuously works for them. This is the biggest differentiating capability for PMs in the AI era.
What a concrete “AI system” looks like
If I were a PM building a community product, I’d set up a system like this:
🔍 Perceive Layer
- AI monitors competitor updates daily (app store changes, social media discussions)
- AI summarizes the top 10 user pain points and sentiment trends weekly
- AI tracks industry reports and key opinion leaders
🔨 Create Layer
- AI generates requirement docs and solution drafts based on my product knowledge base
- AI rapidly produces interactive prototypes for user testing
- AI assists with data analysis to validate hypotheses
🚀 Amplify Layer
- AI helps draft product launch copy and user communications
- AI auto-generates weekly and monthly product data reports
- AI transforms meeting notes into action items and follow-up checklists
The key: this isn’t “occasionally using ChatGPT.” It’s a continuously running system. You decompose your workflow into steps, identify which ones AI can do faster and better, then systematize and automate those steps.
Three Constants, Three Transformations
Back to the two-year-old framework. The core logic hasn’t changed:
- Find the right problem — still the most important starting point
- Design the right solution — still requires judgment and taste
- Make change happen — still requires communication and execution
But the “how” at every layer has transformed:
- Finding problems: from manual research to AI-assisted systematic perception — you’re not watching alone, you have a system watching with you
- Designing solutions: from ideation to curation + rapid validation — generating solutions is cheap now; filtering and validating them is what’s valuable
- Making change: from pushing projects to building systems — you’re not just driving one initiative, you’re building a flywheel that keeps you consistently effective
And there’s an entirely new capability dimension:
Can you build an AI-powered system around yourself, transforming from a “solo PM fighting alone” into a “super-individual backed by an AI system”?
Two years ago I said “academic background, internship experience, English skills, and technical background aren’t required skills.” That’s still true. But I want to add one thing:
AI capability isn’t a required “technical skill” either — it’s a way of thinking. Just as I said you need “user thinking” and “data thinking” two years ago, now you also need “systems thinking” — learning to orchestrate AI as an extension of yourself, so your capabilities are no longer limited by your own time and energy.
And the line I ended with two years ago still holds. I’ll say it again:
Stay confident. Stay curious. Stay learning. If you want to become one, you absolutely can.
Only now, you’ve got a powerful partner by your side. Use it well. 🚀