Karthick's Blog

By Karthick

Two weeks ago, I sat in on a hiring panel that had to pick between two Product Manager candidates. Both had 6+ years of experience. Both came from respected companies. Both nailed the case study.

The difference? One had shipped three working prototypes during our interview process using Claude. The other had asked for two weeks to “coordinate with a designer” for the same task.

We hired the first candidate.

That’s not a story about how AI makes people faster (though it does). It’s a story about a shift that’s happening right now in our profession quietly, unevenly, but decisively. And it’s the reason I’m writing this piece.

If you’re a Product Manager who is still treating AI as a “nice-to-have” or a “future concern,” I want to be direct with you: you are watching your career get slowly obsoleted in real time, and most of your peers don’t have the honesty to tell you.

I’m going to.

The Data That Should Terrify Any PM Sitting This Out

Let’s start with facts, because opinions are cheap.

• 65% of product professionals already use AI daily in their workflow. Gartner projects 70% by the end of 2026.
• Meta became the first major tech company to formally tie performance reviews to AI usage in February 2026. High performers can earn bonuses up to 200%. The rest? The compensation curve is bending.
• METR benchmarks show Claude Opus 4.6 now completes software tasks that take humans 12 hours up from 21 minutes just 16 months ago. That’s a 41x jump in a year and a half.
• LinkedIn quietly killed its Associate Product Manager program and replaced it with a “Product Builder” track that trains generalists across product, design, and engineering.
• 80% of top-performing PMs prototype with AI, saving 2–3 weeks per feature by validating concepts before engineering starts.
• Anthropic the company behind Claude now holds 40% of enterprise LLM API spend, up from 12% in 2023. This isn’t hype. It’s revenue.

Read those numbers again. Then ask yourself which side of this shift you want to be on in 18 months.

Why I’m Talking About Claude Specifically

I’ve used them all ChatGPT, Gemini, Perplexity, Copilot. They all have their moments.

But for the actual work of product management, Claude is the tool that changed how I work. Not because I’m sponsored by Anthropic (I’m not). Because Claude does something the other models still struggle with: it produces work that reads like it came from a thoughtful colleague, not a language model in a hurry.

Here’s what makes Claude different for our line of work:

• Long context handling. I can dump 15 customer interview transcripts and get a synthesis that quotes specific participants and finds patterns across all of them.
• Structured output. When I ask for a PRD, I get a PRD not a wall of text pretending to be one.
• Actual judgment. Claude pushes back on bad prompts, asks clarifying questions, and flags edge cases I missed. That alone has saved me from shipping garbage more times than I want to admit.
• A real PM ecosystem. Claude Skills, Cowork, and Code now include purpose-built PM workflows — PRD generators, research synthesis engines, competitive profile builders. These aren’t gimmicks. They’re standardized methodology encoded as tools.

Now let me walk you through the 10 areas where AI (and specifically Claude) has fundamentally changed how I work. Not theoretically. Actually.

1. Building a Complete MVP Without Waiting for Engineering

The old MVP timeline: PRD → design review → sprint planning → 4 weeks of engineering → user testing → learn → iterate.

The new MVP timeline: description → Claude Code → working prototype in a few hours → user testing → learn → iterate.

I’m not talking about pretty prototypes with placeholder text. I’m talking about actually functional MVPs clickable, interactive, with real logic. Last quarter I built a working prototype of a customer onboarding flow for a fintech project in a Saturday afternoon. Real HTML, real state management, real conditional flows. It was shared with 12 test users on Monday. By Wednesday we had validated our core hypothesis and killed two features we had planned to build.

Cost in engineering hours: zero.

Dennis Yang, a PM at Chime, publicly described the same workflow: he writes his PRD in markdown, opens Claude Code in his terminal, and 20 minutes later he has a running prototype to share in Slack. His team then discusses refinements based on something real — not a Figma mockup that has to be interpreted.

If you’re still going PRD → mockup → beg engineering to build, you’re skipping a step your competition isn’t skipping anymore.

2. A/B Testing: From Hypothesis to Insight in Hours, Not Weeks

A/B testing was always the discipline that separated intuition-driven PMs from data-driven ones. Claude is compressing every step of that loop.

Here’s how I use it now:

• Hypothesis generation. I feed Claude our top-of-funnel drop-off data and ask it to generate 20 testable hypotheses ranked by expected impact. It surfaces angles I hadn’t considered — often grounded in behavioral psychology research it references directly.
• Test design. I describe the variant I want to test, and Claude drafts the test brief: control vs. variant, sample size calculations, success metrics, guardrail metrics, minimum detectable effect.
• Variant creation. For UI-level tests, I can spin up multiple variants as actual working prototypes and share them with my growth team the same day.
• Result analysis. I paste in raw experiment results, and Claude helps me interpret them — flagging statistical significance concerns, Simpson’s paradox risks, and confounding variables I might have missed.

Last cycle, this took our test velocity from 2 experiments per month to 8. And no, we didn’t sacrifice rigor. Claude is actually stricter about experimental design than most PMs I’ve worked with.

3. UI/UX Design and Improvements, Yes, Even Without a Designer

I’m not going to pretend AI is replacing your designer. If you have a great designer, keep them. Design is craft, and the best designers bring taste and judgment that AI can’t match.

But most of us don’t have unlimited designer time. And every PM has faced that moment where you need to explore five UI directions before deciding what’s worth taking to design.

Here’s where Claude shines:

• Design critiques. I share a screenshot of a mockup and ask Claude to critique it against usability heuristics, WCAG 2.2 accessibility standards, and specific principles like Fitts’ Law or Hick’s Law. The feedback is often better than what I’d get from a design review because it’s exhaustive and consistent.
• Alternative explorations. I ask Claude to generate five different approaches to a UX problem — say, “how should we handle the error state when a user’s payment fails?” — and it returns a range of solutions with pros/cons, complete with rough HTML/CSS renderings.
• Copy improvement. Product copy is design. I paste in every string in a flow and ask Claude to rewrite for clarity, tone consistency, and cognitive load reduction. My designers now start with Claude-refined copy, which has cut our design cycle time by roughly 30%.

The key insight: AI doesn’t replace the designer. It elevates every non-designer’s ability to think in design terms.

4. Early Customer Validation at 10x the Speed

The single most consistent failure mode I’ve seen in product management careers is PMs who ship things they think customers want, without actually validating.

We all know we should validate. We just… don’t, because talking to customers is time-consuming and synthesizing what they say is even more time-consuming.

Claude has demolished that excuse.

My workflow now:

1. Recruit. I still do this human-to-human. AI can’t replace real conversations.
2. Interview. I record every call (with permission).
3. Transcribe. Automatic via existing tools.
4. Synthesize. I dump 5–15 interview transcripts into Claude and ask for thematic analysis. It gives me dominant themes with supporting quotes, contradictions between segments, latent needs the customer didn’t explicitly articulate, and, my favorite a list of follow-up questions I should ask next time.

What used to take me an entire day (or often, honestly, “sometime next week”) now takes about 45 minutes. And the output is better than what I’d do manually, because Claude doesn’t get bored on interview.

The compounding effect: because synthesis is fast, I do more interviews. Because I do more interviews, my product decisions are more grounded. Because my decisions are more grounded, I’m more confident in roadmap trade-offs. That’s a career-defining flywheel and it’s available to every PM who’s willing to build the workflow.

5. Data-Driven Decision Making, Even If You Don’t Know SQL

Some of the best PMs I’ve worked with are surprisingly weak at data. They rely on analysts. They wait for dashboards to update. They defer to data science.

That was a viable career strategy until 2024. It isn’t anymore.

Claude changed my relationship with data in three ways:

• Query writing. I describe the question in plain English, and Claude writes the SQL. I paste it into our warehouse. Answer in minutes. No more “let me file a ticket with the data team.”
• Analysis interpretation. I paste in a dataset or query output and ask Claude what stands out. It flags anomalies, correlations, and cohort behaviors I would have missed.
• Metric definition. Half the arguments in product meetings are about how you define a metric. I now bring Claude into these conversations to help me think through activation, retention, and monetization definitions before the meeting, so I show up with clarity instead of confusion.

If you’re a PM who has been hiding behind “I’ll ask the data team,” 2026 is the year that excuse expires.

6. PRDs and Product Documentation, My Favorite Time-Save

I used to spend 4–6 hours writing a good PRD. I now spend 30–45 minutes.

Here’s the honest breakdown:

• Claude drafts the structure: problem statement, user stories, acceptance criteria, success metrics, edge cases, open questions.
• I bring the strategic input: why this matters, how it aligns with company goals, non-obvious trade-offs, my judgment on scope.
• We iterate. I flag weak sections. Claude rewrites. I refine.
• The final PRD is often better than my solo work because Claude catches edge cases I would have shipped past.

The rule I follow: Claude drafts. I decide. Strategic judgment is mine. Mechanical work is not.

The same principle applies to every document PMs write launch plans, retrospectives, stakeholder updates, one-pagers, executive summaries. If it has structure, Claude can accelerate it.

Marcus Moretti, a general manager at a well-known AI company, said publicly: “The only product document I’ve written in my new role is the roadmap. Everything else every PRD and every ticket — has been written by Claude.”

Let that sink in. Not as a threat, but as an invitation.

7. User Story Creation That’s Actually INVEST-Compliant

Bad user stories are the silent killer of product velocity. “As a user, I want to log in so that I can access my account” is not a user story. It’s a shopping list item.

Claude writes user stories in Given/When/Then acceptance criteria format, checks them against INVEST criteria (Independent, Negotiable, Valuable, Estimable, Small, Testable), and flags dependencies I missed.

More importantly, Claude uses the customer’s actual language if you feed it your research transcripts. That’s a subtle but massive shift. Your stories stop sounding like PM jargon and start sounding like the words your users actually said.

For a checkout redesign I ran last quarter, Claude generated 47 user stories from a single well-written feature brief, complete with acceptance criteria and edge cases. My engineers told me it was the cleanest backlog they’d worked from in three years.

8. Market Research Without Consultants

Market research reports used to cost tens of thousands of dollars. Now Claude — combined with tools like Perplexity for cited sources gives me a defensible market analysis in an afternoon.

My typical workflow:

1. Define the question narrowly. Bad: “Analyze the CRM market.” Good: “Which segments of SMB CRM users are underserved by HubSpot’s current feature set, and what would it take to peel them off?”
2. Ask Claude to build a research plan what to investigate, what data sources to check, what frameworks to apply (Porter’s Five Forces? Jobs to be Done? Category creation vs. displacement?).
3. Execute the research. Some via Claude with web search, some via Perplexity for cited sources, some via manual interviews.
4. Synthesize with Claude. I paste all findings and ask for a structured market brief.

The output isn’t consulting firm polished. But it’s 80% as good, at maybe 5% of the cost and I own the analysis, which means I can refresh it every quarter instead of every three years.

9. Competitive Analysis That Actually Stays Current

Traditional competitive analysis dies in a Google Doc that no one updates. Claude fixed this for me.

I have Claude generate a structured competitive profile for each of our top competitors same dimensions every time: positioning, pricing, feature set, GTM motion, org signals from LinkedIn, recent press. Because it takes 15 minutes per competitor, I actually refresh them every quarter.

More importantly, I use Claude for competitive war-gaming: “Assume you’re the head of product at Competitor X. What would you launch in the next 6 months to attack our position?” The moves Claude suggests are often eerily close to what those competitors actually make. It sharpens my strategic thinking in a way that no static competitive matrix ever did.

10. Product Strategy and Roadmap Planning

This is the one where I want to be most careful in what I claim.

AI does not do product strategy for you. Strategy is the part of PM work that requires judgment, context, taste, and the willingness to be wrong. No LLM has that. Anyone claiming otherwise is selling you something.

But AI is a phenomenal strategic thinking partner. Here’s what I do:

• Assumption testing. I write out my strategic bets and ask Claude to steelman the counter-argument. Every single time, it finds at least one hole in my logic.
• Trade-off framing. I describe a prioritization conflict and ask Claude to structure the trade-off explicitly — what I’m gaining, what I’m giving up, what the second-order effects are.
• Roadmap stress-testing. I share my quarterly roadmap and ask Claude to poke holes: what’s the biggest risk to shipping on time, which initiative is most likely to fail, where am I underweighting technical debt.
• Now/Next/Later drafting. From a strategy doc, Claude can propose a Now/Next/Later structure aligned to it — a great starting point, though the final call is always mine.

The metaphor I keep coming back to: Claude is the smartest analyst on my team who I can talk to at 11 p.m. They don’t make the decision. But they make sure I’ve thought about the decision from every angle before I make it.

“But I’m Worried About…” The Objections I Keep Hearing

Let me address the objections I hear most often from PMs who are resisting the shift.

“AI outputs are unreliable.”
Correct. That’s why the PM’s job is now more about judgment and evaluation, not less. Your value increased, not decreased. AI creates the draft. You decide if it’s right.

“My company doesn’t allow AI tools.”
For sensitive data, that’s fair. For everything else, it’s usually a policy that hasn’t caught up to reality. Push. Advocate. Show ROI. Be the PM who brings the vetted-tool proposal to security — not the one who complains about the ban.

“I don’t want my work to sound generic.”
Then don’t let it. Feed Claude your own writing samples, your product’s voice, your competitive positioning, your customer research. Generic in equals generic out. Specific in equals specific out.

“AI will replace PMs.”
It won’t. But AI-skilled PMs will replace AI-illiterate ones. Meta is already ranking employees on AI usage. LinkedIn is redesigning the entry-level PM role around AI fluency. This is happening whether you engage or not.

“I’m too senior to start over.”
Then you’re too senior to fall behind. Senior PMs who master AI compound their existing product judgment with 10x leverage. Senior PMs who don’t become the “why isn’t this shipped yet?” bottleneck.

How to Actually Start This Week

If you’ve read this far, I don’t want to leave you with generic advice. Here’s the exact sequence I recommend to any PM ready to close the gap:

Week 1: Set up.

• Get Claude Pro ($20/month). Not the free tier the free tier won’t do the heavy lifting.
• Install Claude Desktop.
• Try Claude Cowork if you want a desktop agent that works with your local files.

Week 2: One workflow at a time.

• Pick your biggest time sink. For most PMs, that’s either PRDs or customer research synthesis.
• Do it in Claude for two straight weeks. Save the prompts that work. Iterate.
• Don’t try to change everything at once that’s the fastest way to abandon the practice.

Week 3: Build a context file.

• Create a document that describes your product, your users, your competitors, your voice, your metrics, and your constraints.
• Feed it to Claude at the start of every session. This is the single highest-leverage move you can make.

Week 4: Add a second workflow.

• If PRDs worked, add competitive analysis. Or A/B test design. Or user story writing.
• Repeat.

Within 90 days, this will feel as native as opening Google Docs.

The Train Is Leaving. Get On.

I opened this piece with a hiring story. I’ll close with the truth that story implies.

Every quarter for the next 3–5 years, the gap between AI-fluent PMs and AI-illiterate PMs is going to widen. Meta ties bonuses to AI. LinkedIn redesigns the entry role. Anthropic ships Claude Cowork. METR benchmarks show AI doing 12-hour engineering tasks. The pyramid inverts. The train accelerates.

If you get on that train now even clumsily, even with mistakes, even feeling behind — you have 12–18 months to catch up before the compounding curve makes the gap unbridgeable.

If you don’t, one day soon you’ll interview for a role and lose it to someone who did.

That’s not a threat. It’s just the arithmetic of exponential technology change meeting a profession that lives at the intersection of technology and human judgment.

I’m not asking you to abandon what makes you good at product management. I’m asking you to add a new capability to the ones you already have. The PMs who do this will have the careers I’d want. The ones who don’t will spend the next five years asking, “Wait, when did this happen?”

The train is leaving the station. I’m already on it.

Come sit with me.