On September 23, 1997, Steve Jobs stood in front of Apple employees and did three unfashionable things: he simplified, he shortened the supply chain, and he reminded everyone that marketing is about values, not speeds and feeds.

Apple Internal - Introducing the Think Different Campaign : Apple Inc. : Free Download, Borrow, and Streaming : Internet Archive

Five days later, the Think Different spot debuted during The Wonderful World of Disney’s TV premiere of Toy Story. Apple didn’t pitch megahertz; it honored misfits. The rest is corporate mythology — and operating discipline.

For context: Jobs had just been made interim CEO on September 16, 1997 (“iCEO,” if you like your titles as lowercase as your ego). In that internal talk, he said he’d “been back about 8–10 weeks,” which tracks with the board shake-up over the summer. The point isn’t nostalgia; it’s that his playbook is weirdly more relevant amid today’s AI gold rush.

The five fundamentals (that beat FOMO)

1) Ruthless subtraction → coherent product.

Jobs killed ~70% of the roadmap so customers could actually understand the line-up. Focus created legibility; legibility created demand. In AI terms: fewer “labs” features, more end-to-end workflows people finish on a Tuesday.

2) Shrink latency, not just ship models.

Apple pulled months of inventory out of its pipeline so the customer could “tell us what they want,” and we could respond fast. Replace “inventory” with “backlog and MLOps latency,” and you’ve got your GTM: shorter model-to-market loops beat bigger parameter counts.

3) Distribution is a product.

Jobs talked about catching up — and then innovating — in distribution and manufacturing. In 2025 that means: integrations that install in minutes, procurement-friendly pricing, and security reviews that don’t eat Q4. The fastest route to revenue isn’t a benchmark; it’s a purchase order.

4) “Marketing is about values.”

Nike doesn’t sell air soles; it honors athletes. Jobs refused to sell “nips and megahertz” (his words), and reframed Apple around creators and progress. In AI, swap spec-sheets for outcomes: saved hours, fewer incidents, higher win-rates. If your homepage reads like a dissertation, your pipeline will too.

5) Brand = promise you keep when no one is watching.

The campaign worked because the company then shipped iMacs and G3s that embodied the promise — fast, simple, colorful, connected. A brand that outkicks its product is performance art; a product that outruns its brand becomes a movement.

10× vs. 10i: how “the crazy ones” actually execute

Jobs loved “the crazy ones,” but his “crazy” was methodical — reduce scope, tighten feedback, and amplify taste. If you want 10× outcomes, you need 10i — ten tight iterations — per cycle. Ten user-observed releases beat one keynote. Ten hour-to-insight loops beat a quarter of “strategic exploration.” Ten serious customer references beat a thousand retweets. (Yes, this is the part where your growth deck quietly closes itself.)

Think of it as physics: mass (adoption) × velocity (iteration) = momentum. The only deadline is physics — frictionless systems win. Jobs’ 1997 week looked like: simplify the SKU tree, collapse lead times, launch a story people remember, then ship products that prove the story. The airing on Sept 28, 1997 wasn’t the finish line; it was the ignition key.

If Jobs were here, what would he say about his own “craziness”?

Probably something like: “Crazy is just focus that looks impolite at first.” The internal talk makes it plain: he cut, he tightened systems, and he told a true story. He’d tell AI founders today to:

• Delete 70% of the roadmap you can’t explain in one slide. (You’ll ship the remaining 30% in half the time and actually support it.)
• Measure pipeline half-life, not vanity metrics. If a feature sits six months between demo and deployment, the problem isn’t “market education”; it’s entropy. (Jobs’ fix was to remove months of guesswork from the chain.)
• Honor your user, not your model weights. Nike honors athletes; you should honor operators — analysts, PMs, finance teams, sales reps — who will stake their quarter on your product. Tell their story.
• Ship a bill of materials for trust. Security, privacy, data-residency, audit trails — if you can’t pass procurement, you can’t pass go.

A quick SJ reality check

Yes, everyone is “going to the moon.” Also: you’ve brought beige spreadsheets to a rocket launch. Jobs literally mocked beige. The antidote is not chaos-monkey “innovation theater”; it’s taste + throughput. If your team can’t name your customer, can’t map a 7-click flow to value, and can’t release weekly without drama, your 10× talk is just karaoke with better lighting.

The operating system of “Think Different,” updated for 2025

1. Clarity is a growth hack. Simple catalog, sharp ICP, one pricing page.
2. Latency kills love. Collapse time from request → usable feature.
3. Distribution compounds. Land where users live (GSuite, Slack, Salesforce, Snowflake), not where you wish they lived.
4. Story first, specs second. But never story instead of specs — earn the slogan.
5. 10i → 10×. Ten real iterations beat one mythical leap. Physics is still the only deadline.

The 1997 talk wasn’t poetry; it was process wrapped in poetry. That’s the part to steal.

References (unsorted)

• Apple Internal — Introducing the Think Different Campaign (internal talk; meeting dated Sept 23, 1997). Internet Archive.
• Steve Jobs named interim CEO (Sept 16, 1997). Los Angeles Times; also summarized by Wired and Poynter.
• “Marketing is about values” transcript & video excerpts. Speakola
• Think Different campaign (full versions and Jobs-narrated cut). Internet Archive / YouTube.
• Broadcast context: Toy Story on ABC’s Wonderful World of Disney, Sept 28, 1997, with period commercial breaks. IMDB / archived TV uploads.

Imagine the Tyrannosaurus Rex. Yes, the 12,000-pound apex predator with tiny arms that look like it’s auditioning for the role of “most awkward gym bro.” But beyond the jokes, paleontologists have shown that T-Rex was a masterpiece of efficiency. Its ligaments and tendons were tuned to store and release energy with remarkable frugality, much like kangaroos hopping today (Alexander, 1996). In physics, this is described by the Cost of Transport (COT): the energy expended per unit of body weight per unit of distance traveled. Lower COT means greater efficiency.

Now, replace the T-Rex with Anthropic, OpenAI, Grok, or ElevenLabs. Suddenly, the question becomes: which AI company is learning to “walk” most efficiently, and which is burning through capital like a rocket strapped to a refrigerator?

The Physics of Efficiency, Applied to AI

In mechanics, there’s a principle called the Principle of Least Action (Hamilton, 1834). Put bluntly, nature finds the path that minimizes wasted energy. Economists later reformulated similar ideas into transaction cost economics (Coase, 1937; Williamson, 1985). In business, we call this efficiency ratios.

For AI startups, the analogy is:

• Energy input = capital, compute, and talent.

• Distance traveled = validated progress (deployments, revenue, retained users).

• COT in AI = resources consumed per unit of real-world adoption.

So when OpenAI spends billions training frontier models, their COT might be high, but the “distance” is global market leadership. Meanwhile, Hugging Face’s “model zoo” lowers COT for thousands of smaller players by democratizing access to models, much like bicycles lowering human transport costs compared to sprinting (Vogel, 2013).

The Big Labs: T-Rex Scale, Rocket Fuel Burn

OpenAI, Anthropic, and Google DeepMind operate like T-Rexes on steroids. Their strategy: pour vast amounts of capital into compute-intensive models with the hope of extracting network effects and enterprise contracts later.

• OpenAI leverages Microsoft’s infrastructure like a dinosaur using a conveniently placed volcano — risky, but spectacular.

• Anthropic markets safety and alignment, but its Claude models remain compute-hungry.

• Grok/XAI positions itself as “irreverent efficiency,” but in practice, training costs remain astronomical.

Their COT resembles early aerospace: breathtaking burn multiples, justified only if they achieve orbit. As Wright’s Law (1936) shows, costs per unit should fall with cumulative production — but only if you survive long enough to accumulate scale.

The Niche Sprinters: Efficiency as Strategy

Contrast this with ElevenLabs (voice synthesis) or Perplexity (AI search). These “AI-native applications” (let’s call them AINA for brevity: AI-Native Applications) pick a narrow task, optimize ruthlessly, and deliver value faster than the giants can generalize.

• ElevenLabs achieves ~$0.8M revenue per employee (20VC, 2025) by focusing exclusively on voice. That’s tendon-level efficiency.

• Hugging Face functions as infrastructure scaffolding for the AI ecosystem. Instead of building the biggest dinosaur, they build the forest in which others evolve.

The lesson: in early phases, efficiency often beats brute force. Just as birds — descendants of dinosaurs — evolved into lighter, faster forms, so too do AI startups find survival in niches.

Avoiding FOMO: Lessons for Incumbents

Corporate executives see AI headlines and feel the gravitational pull of FOMO: “Quick! Build a foundation model before our board asks why we don’t have one!” This is akin to buying a Ferrari for a 200-meter school run.

Physics reminds us: efficiency is not speed alone — it’s optimal energy per distance. Business leaders should:

1. Define distance: is success measured in ARR, retained customers, or validated experiments?

2. Measure COT: track burn multiple, CAC payback, and operational leverage.

3. Resist FOMO: efficiency metrics protect against irrational capital expenditure.

As in biology, survival is not about who runs fastest, but who burns energy wisely over the long distance.

So, What Can AI Learn from T-Rex?

• Frontier labs: accept higher COT, but only if orbit (dominance) is plausible.

• Niche players: exploit tendon-like efficiency to outpace lumbering giants.

• Incumbents: define your distance, then minimize capital per unit progress.

In short, the physics of efficiency applies as much to companies as to kangaroos. Or to put it in John Oliver’s words: “If you’re spending billions training a model that still hallucinates like a Victorian poet on laudanum… maybe, just maybe, you’ve got a Cost of Transport problem.”

References

• Alexander, R.M. (1996). Dynamics of Dinosaurs and Other Extinct Giants. Columbia University Press.

• Coase, R.H. (1937). “The Nature of the Firm.” Economica.

• Hamilton, W.R. (1834). “On a General Method in Dynamics.” Philosophical Transactions of the Royal Society.

• Vogel, S. (2013). Comparative Biomechanics: Life’s Physical World. Princeton University Press.

• Williamson, O.E. (1985). The Economic Institutions of Capitalism. Free Press.

• Wright, T.P. (1936). “Factors Affecting the Cost of Airplanes.” Journal of the Aeronautical Sciences.

“at least half of our ideas are just not going to work”

because customers often don’t share the same enthusiasm as product teams. This reality underscores the importance of product discovery — the process of validating what to build before investing in development.

Today, Artificial Intelligence (AI) is transforming this process, enabling product teams to make smarter, faster, and more customer-centric decisions.

The Burning Gap: Why Product Discovery Needs AI

As companies scale, the complexity of customer interactions increases. Sales, support, and marketing teams gather feedback across various touchpoints, but this data often becomes fragmented and biased. The result is a widening gap between product managers (PMs) and their users.

AI helps bridge this gap by automating the collection and analysis of both structured data (e.g., CRM metrics, product analytics) and unstructured data (e.g., support tickets, interviews, social media). Up to 90% of valuable customer insights lie in unstructured formats — an area where AI excels.

What Is Product Discovery?

Product discovery is about reducing three core risks:

• Desirability: Do customers want this?
• Feasibility: Can we build it?
• Viability: Will it deliver business value?

AI enhances each of these dimensions by surfacing patterns in user behavior, predicting outcomes, and enabling rapid experimentation.

How AI Enhances Product Discovery

1. Understanding the Problem Space
   AI tools like Dovetail and Maze automate user research, helping teams extract insights from interviews, surveys, and support logs [1].
2. Validating the “Why”
   AI models can test hypotheses by analyzing sentiment, predicting churn, and identifying revenue opportunities. This enables faster iteration through MVPs and A/B testing [2].
3. Discovery-Led Planning
   AI supports roadmap planning by modeling “what-if” scenarios and prioritizing features based on impact and feasibility [3].
4. Closing the Feedback Loop
   Real-time AI systems can notify users of changes made based on their feedback, fostering loyalty and trust.

Metrics That Matter

To measure the success of AI-driven discovery, teams should track:

• Business Impact: Revenue growth, cost savings
• Customer Satisfaction: CSAT, NPS, onboarding success
• Engagement: Retention, churn, usage patterns

These metrics help determine whether the product is achieving product-market fit — a critical milestone in any product’s lifecycle [2].

Challenges in Adopting AI

Despite its promise, AI adoption faces hurdles:

• Organizational Readiness: Teams need training and infrastructure.
• Data Quality: Incomplete or siloed data limits AI effectiveness.
• Human-AI Balance: AI should augment — not replace — human judgment [4].

The Future of Product Discovery

The future lies in continuous discovery, where AI and machine learning are embedded across the product lifecycle. From ideation to iteration, AI will enable:

• Deeper customer insights
• More accurate prioritization
• Predictive user behavior modeling
• Faster time-to-market

As AI matures, it will not only accelerate innovation but also democratize it — empowering teams of all sizes to build products that truly resonate.

References

• Douglas, C. (2024). The Role of AI in Roadmap Planning for Product Managers. Visily. https://www.visily.ai/blog/ai-in-roadmap-planning/
• Mahajan, P. (2025). Product Discovery with AI [PDF slides]. Productside. https://productside.com/webinar/how-to-do-product-discovery-with-ai/
• McKinsey & Company. (2024). AI-enabled software development fuels innovation. https://www.mckinsey.com/industries/technology-media-and-telecommunications/our-insights/how-an-ai-enabled-software-product-development-life-cycle-will-fuel-innovation
• Naeem, R., Kohtamäki, M., & Parida, V. (2025). Artificial intelligence enabled product–service innovation: past achievements and future directions. Review of Managerial Science, 19, 2149–2192. https://link.springer.com/article/10.1007/s11846-024-00757-x
• Productboard. (2025). Using AI for Product Roadmap Prioritization. https://www.productboard.com/blog/using-ai-for-product-roadmap-prioritization/
• Skala, A. (2024). 25 Best AI Tools for UX Research. UXtweak. https://blog.uxtweak.com/ai-tools-for-ux-research/
• Stevens, E. (2024). The Top 5 AI Tools for User Research (and How To Use Them). UX Design Institute. https://www.uxdesigninstitute.com/blog/top-ai-tools-for-user-research/
• Toner-Rodgers, A. (2025). Artificial Intelligence, Scientific Discovery, and Product Innovation. arXiv. https://arxiv.org/abs/2412.17866
• Zeda.io. (2024). Build Products Customers Love: AI-Powered Customer-Centric Development. https://zeda.io/blog/build-products-customers-love-with-ai

Artificial intelligence can now assist with more than ideation or summarization — it can support the end-to-end development of a product roadmap. This guide walks through how to do exactly that using just ChatGPT. No Notion, no spreadsheets, no specialized product software.

All you need is structured input, a clear set of prompts, and a basic understanding of your product’s context.

Step 1: Cluster and Analyze User Feedback

Start with unstructured data: customer complaints, support tickets, interviews, surveys, or product reviews. Your goal is to extract patterns without spending hours manually reviewing individual entries.

Prompt:
Here’s a batch of user feedback.
Cluster the content into 3–6 themes. For each:
- Give the theme a clear label
- Rate its pain level (1–10)
- Estimate frequency (High, Medium, Low)
- Include 1–2 representative user quotes

This gives you a structured foundation to work from, based on real user input

Step 2: Translate Problems Into Product Ideas

Once themes are identified, the next step is to translate those into candidate features or improvements. Focus on clear connections between user pain and the proposed solution.

Prompt:
Given these problem themes, suggest 5 product ideas.
Each should:
- Solve a specific user pain
- Include a 1-sentence description
- Mention how AI or automation might improve the experience (optional)

These ideas can now serve as your roadmap candidates.

Step 3: Prioritize Using RICE

Use the RICE framework (Reach, Impact, Confidence, Effort) to score and compare feature ideas in a consistent way. ChatGPT can generate reasonable estimates and produce a sortable table.

Prompt:
Evaluate the following product ideas using the RICE framework.
Provide a table with Reach, Impact, Confidence, Effort, and a final RICE score.
Assume this is a mid-sized B2B SaaS product.

This gives you a ranked list of features, with scoring rationale you can refine further.

Step 4: Create a Roadmap Draft

With prioritized items, the next step is to draft a timeline. Use GPT to group items by delivery phase, team function, or quarter.

Prompt:
Turn these prioritized features into a 2-quarter product roadmap.
For each feature, include:
- Name
- Objective
- Expected outcome
Group by Q3 and Q4.

This roadmap can be easily reviewed, adjusted, or copied into your internal tooling.

Step 5: Align With Strategy

A roadmap should reflect company goals. Whether you’re prioritizing enterprise features, reducing churn, or improving margins, ChatGPT can help you assess whether your proposed features are aligned.

Prompt:
Here is our company strategy:
- Expand into enterprise accounts
- Focus on retention and LTV
- Leverage AI for differentiation

Given this roadmap, evaluate:
- Which features align best
- Which are weak fits
- Any features to reconsider or reframe

This keeps the roadmap focused and defensible during reviews.

Step 6: Pressure-Test from a Competitor’s Viewpoint

As a final step, it’s useful to step outside your own frame and evaluate your roadmap from a competitor’s perspective.

Prompt:
Act as the Head of Product at our top competitor.
Given this roadmap:
- Identify potential weaknesses or gaps
- Suggest what your team would build to counter it
Then switch back and suggest how to reinforce our roadmap

This helps you uncover potential blind spots and strengthen your competitive differentiation.

Conclusion

Using ChatGPT in this structured way allows you to build a grounded, flexible, and strategy-aligned product roadmap — without needing spreadsheets, tools, or large teams. This process is especially useful for early-stage startups, lean product teams, or solo builders.

For those who want to implement it right away, you can start with just:
1. 10–20 user feedback points
2. 20–30 minutes of focused prompting
3. A running document for logging your GPT outputs

Sources

• https://www.hustlebadger.com/what-do-product-teams-do/product-roadmap-examples/
• https://notion.notion.site/29d47fbeefc84054a98a3f8005c377ad?v=fcf868b8a42543b1bccf53a609fd2d1b
• https://github.com/orgs/github/projects/4247/views/1?sliceBy%5Bvalue%5D=%F0%9F%9A%80+Accelerate+with+AI
• https://www.reforge.com/artifacts/strategy-roadmap-template-at-stripe
• https://drive.google.com/file/d/1nNxRei3wZje2G7mhnnfA9p1p5g0bTh93/view
• https://docs.google.com/spreadsheets/d/1zlx3RuidNOW40Zf7gh07p2SqoR53Ungv9JFT-PhHwxI/edit?gid=184965050#gid=184965050

The Hard Truth You Need to Hear

So you think shipping features on time and hitting your targets means promotion is inevitable? Cute. Reality check: It’s not what you do, but what others remember and can repeat about what you do. Great work is invisible without a story (Gupta & Gupta, 2024).

Roshan Gupta, a six-year Google Group Product Manager, confirms that product leadership hinges on the ability to “show your impact in a way that resonates and sticks.” This is not about spinning fairy tales, but about strategically crafting narratives that align teams and decision-makers behind your results.

1. The Dilemma of Impact: Delivering Without Resonance

Imagination: “If I ship features and meet deadlines, my work speaks for itself.”

Reality: Nope. Sociologist Erving Goffman (1974) taught us all social interactions are performances — your work needs a stage and a script.

AI-powered narrative builders synthesize raw data and business outcomes, crafting clear, persuasive stories that show how your contributions move the needle. Instead of drowning in tasks, your story shows why your work matters strategically (Gupta & Gupta, 2024).

Pro Tip: Use AI tools to analyze KPIs and convert numbers into narratives. This is storytelling as business strategy, not fluff.

2. Failure to Articulate: When Great Work Is Silent

Imagination: “Senior leaders will recognize my greatness even if I say nothing.”

Reality: Daniel Kahneman’s (2011) research on cognitive overload means your impact disappears unless you make it visible. Promotions are often decided in your absence.

AI writing assistants craft sharp, customized impact statements that frame your results in terms execs care about: revenue, retention, growth, efficiency. AI adapts tone and style for your company’s culture, so your story doesn’t just land — it sticks (Gupta & Gupta, 2024).

3. The Peer Problem: Admiration Without Advocacy

Imagination: “My peers get what I do; that’s enough for promotion.”

Reality: Social capital guru Pierre Bourdieu (1986) reminds us that admiration is not advocacy. Your peers must activate their support, which requires shareable, easy-to-understand narratives.

AI-powered platforms generate shareable impact summaries, micro-videos, and badges. These give your colleagues the ammo they need to champion your cause authentically. Without this, you’re in a lonely fight for recognition — and losing (Gupta & Gupta, 2024).

How AI Supercharges Your Storytelling — Practical Steps

• Synthesize and Align: Use AI to turn disparate project data into a coherent story linked to business outcomes.
• Automate Real-Time Documentation: Let AI track wins and lessons during your workday, so your narrative never misses a beat.
• Personalize Communication: AI crafts different versions of your story tailored for execs, peers, and cross-functional teams.
• Empower Peer Advocacy: Generate shareable content and track engagement to mobilize your internal network.
• Practice Your Story: Use AI coaching bots to rehearse and refine your pitch before promotion meetings.

Final Challenge

Stop assuming your work will speak for itself. It won’t. Start telling your story so well that it can’t be ignored — today.

Use AI tools to craft your impact story, share it with your allies, and push them to become your loudest advocates. Then come back in 90 days and tell me you’re still invisible.

References

Bourdieu, P. (1986). The Forms of Capital. In J. Richardson (Ed.) Handbook of Theory and Research for the Sociology of Education.

Goffman, E. (1974). Frame Analysis: An Essay on the Organization of Experience. Harper & Row.

Gupta, A., & Gupta, R. (2024). Storytelling: The Product Manager’s Superpower. Product Growth. Retrieved from https://www.news.aakashg.com/p/storytelling-pm

Kahneman, D. (2011). Thinking, Fast and Slow. Farrar, Straus and Giroux.

Let’s face it: most PRDs are broken before they even ship.
They’re too long, too vague, or way too confident about ideas that haven’t been tested yet.

The usual PRD template tries to do everything at once — from vague problem exploration to final dev specs. That’s the problem. You end up forcing structure on ideas that aren’t ready, or leaving out important thinking when you are.

So here’s a better way.

Use Different PRD Prompts for Different Stages

Writing a PRD should be a process — not a one-time document. Your first version shouldn’t look anything like your final one. That’s why I created a set of AI prompts that match the actual product development process — from early discovery to build-ready spec.

Each version maps to a different stage:

• PRD 0.1 — You’re exploring a problem. You’re still gathering input.
• PRD 0.3 — You’ve found a real pain point and want to define it clearly.
• PRD 0.5 — You’re sketching out solution ideas and thinking about what to test.
• PRD 0.7 — You’re preparing for MVP. You’ve got something to build and validate.
• PRD 1.0 — You’re ready to ship. Specs, metrics, risks, edge cases — all covered.

Instead of filling out a massive doc from the start, you just use the right prompt for where you are — and let it evolve as you learn.

PRD 0.1 Prompt — Problem Discovery (Exploration)

You are an early-stage product strategist. Based on the input below, write PRD version 0.1. Focus on uncovering the real user pain and systemic gaps. The goal is not to suggest solutions but to frame and document the **problem space** clearly.

Include these sections:

0. TL;DR Summary – A short statement of what this is about and why it might matter
1. Problem Statement – Raw, exploratory problem description
2. User Observations – Include anecdotes, quotes, research notes
3. Jobs to Be Done – What are people trying to get done (even if badly)?
4. Open Questions – What don’t we know yet?
5. Assumptions – What are we assuming about users, market, tech?
6. Scope – What’s clearly out of scope?
7. Next Steps – What research or exploration do we need?

Here is the raw input:
[Paste user feedback, data, observations, etc.]

PRD 0.3 Prompt — Problem Framing (Definition)

You are a mid-stage product manager validating and framing a clearly defined user problem. Write PRD version 0.3 with sharper structure and early impact framing. Do not yet include full solutions.

Use this structure:

0. TL;DR Summary – Clear problem summary and why it matters now
1. Problem Statement – Validated and scoped user pain point
2. User Segments – Who is affected most and why?
3. JTBD – Specific jobs this user group wants to solve
4. Business Impact – Why solving this is valuable to the company
5. Scope – What’s in/out of our initial MVP scope
6. Success Hypothesis – If we solve this, what will improve?
7. Open Questions – What must we still research or confirm?
8. Risks – Misframing, invalid assumptions, etc.

Input data:
[Insert insights, pain points, validation interviews]

PRD 0.5 Prompt — Ideation & Opportunity Framing

You are a cross-functional product team lead shaping solution directions. Write PRD version 0.5 with structured early concepts and testable hypotheses. This version should guide early prototyping and lo-fi tests.

Structure:

0. TL;DR Summary – Problem + high-level idea + outcome goal
1. Problem Statement – Refined problem definition
2. User Stories – Sample real-life usage contexts
3. JTBD – Actionable job statements
4. Business Impact – What shifts if we get this right?
5. Scope – What’s tentatively in/out
6. Assumptions – What must be true for this to work?
7. Success Criteria – What indicators should improve? (directional)
8. Prototype Prompt – Include Figma / code prompt suggestion
9. Validation Plan – What do we test first, with whom, and how?
10. Risks – UX failures, false positives, competitor moves

User input:
[Paste early solution ideas, hypotheses, research]

PRD 0.7 Prompt — MVP Execution Spec (Ideate)

You are a senior product owner preparing an MVP for build and learning. Write PRD version 0.7 optimized for design and engineering alignment. Include all tactical components and test plan.

Structure:

0. TL;DR Summary – What we're building and why
1. Problem Statement – Root cause and user motivation
2. User Stories – Prioritized flows
3. JTBD – What users are really trying to solve
4. Goal & Business Impact – KPIs, growth lever, revenue, retention
5. Scope – Define MVP line clearly
6. Success Metrics – Add ranges (e.g., Retention Day 7 > 30%)
7. Assumptions – What still needs testing
8. Risks & Edge Cases – Technical, UX, data integrity issues
9. Technical Tasks – Split by frontend/backend, labeled P1–P3
10. Prototype Prompt – Detailed, for design/code handoff
11. Post-Launch Plan – What success looks like in 2 weeks

Feed:
[Paste MVP concept, UX plan, user feedback, business goals]

PRD 1.0 Prompt — Ready-to-Build (Test)

You are a senior product delivery manager preparing for final build. Write PRD version 1.0. This is the fully validated, engineering-ready spec. It should contain no major unknowns.

Use this format:

0. TL;DR Summary – Concise one-liner + KPI targets
1. Problem Statement – Final statement linked to outcomes
2. User Stories – Realistic and confirmed
3. JTBD – Fully synthesized
4. Goal & Business Impact – Growth, retention, strategic alignment
5. Scope – MVP + fast-follower backlog
6. Success Metrics – Firm KPIs + targets (e.g. CTR ≥ 12%)
7. Assumptions – Note remaining soft spots
8. Key Risks and Edge Cases – Full spectrum
9. Technical Tasks – FE/BE split, P1 (critical), P2 (stretch), P3 (future)
10. Post-Launch Validation Questions – What are we watching?
11. Prototype Prompt – Ready for design/dev
12. Release Plan – Timeline, launch strategy, ownership

Input:
[Paste polished spec, validation results, user flows, tech context]

🛠️ How to Use These Prompts

1. Identify Your Current Stage: Are you still learning about the problem? Do you have a clear direction? Are you ready to build?
2. Paste the Right Prompt into your AI assistant: Each version includes the right sections, questions, and structure. Just drop it into ChatGPT, Claude, or your tool of choice.
3. Add Your Input: Include customer interviews, feedback, sketches, goals, or even just a loose idea.
4. Review the Output Critically: Use the draft to align with your team, surface missing insights, and refine thinking — not to “finish the doc.”
5. Iterate Forward: As you validate assumptions and sharpen hypotheses, upgrade to the next PRD version prompt. Each version builds on the last. It’s a living doc — not a final deliverable.

⚔️ The Hard Truth

Most startups claim to be user-centric, but few actually have the systems to back it up — especially small teams. They’re too strapped for time, people, or resources to run traditional interviews, synthesize findings, and make data-driven product calls at speed.

Over the past months, I’ve consulted multiple product teams — early-stage startups, growth-phase orgs, and AI-first builders.

Despite their differences, one pattern keeps repeating:

⚠️ Most teams say they’re user-centric, but don’t have the systems to make that real.

They skip discovery. Delay synthesis. Or treat feedback like noise.

It’s not that they don’t care — they lack time, structure, or the right tools.

So I started recommending a repeatable, AI-powered user research loop that works under pressure. Lightweight. High-signal. LLM-native.

Here’s the system I share with every product team I advise.

AI-Powered User Research Stack: A Systematic Toolkit for Small Teams

🧬 Start with Synthetic Personas, Not Hypotheses

Problem: Early-stage teams often guess personas or use outdated templates.

My AI approach:

• I use GPT-4 or Claude to create high-fidelity proto-personas by prompting it with real market data: review scrapes, Reddit threads, App Store comments.
• I enrich this with semantic clustering from tools like Glean, [ChatGPT Advanced Data Analysis], or Klu.

Prompt sample:

“Analyze 300 app reviews from competitors. Identify 4 core user personas and their emotional drivers. Output: motivations, behaviors, preferred outcomes, and jobs-to-be-done.”

✅ Result: Personas based on actual voice-of-customer, not guesswork.

2. 🔦 Find the Pain, Don’t Assume It

Problem: Startups often solve the wrong problem — or solve a real one too late.

My AI approach:

• I mine qualitative insights from forums, chats, Discords, and community threads.
• Use AI to synthesize and categorize complaints, questions, and dreams.

Prompt sample:

“Summarize 1,000 support tickets and online reviews. Group pain points by urgency and emotional tone. Prioritize by frequency and revenue impact.”

✅ Result: A heatmap of real frustrations, by persona and context.

3. ✍️ Write Stories, Not Features

Problem: PMs jump too quickly to features. I slow down to write stories that sell.

My AI approach:

• I use LLMs to rewrite insights as Jobs-To-Be-Done, user stories, or day-in-the-life narratives.
• Then test those against real customer voice data to confirm alignment.

Prompt sample:

“Turn these 20 review quotes into structured Jobs-To-Be-Done statements. Then rewrite them as first-person user stories in customer language.”

✅ Result: Alignment between customer mental model and product strategy.

4. 🧠 Use Vector Search to Find Repeated Patterns

Problem: Qualitative research is messy and expensive to analyze.

My AI approach:

• I embed all user research data — interviews, feedback, tickets — into a vector database like Pinecone, Weaviate, or OpenAI’s file search.
• Then I query the corpus for recurring patterns.

Prompt sample (run against vector index):

“What are the top 5 workflow friction points mentioned by users in the last 3 months?”

✅ Result: High-signal insight without re-reading everything manually.

5. 📈 Run Continuous AI-Based Simulations

Problem: Most research is one-off. I create continuous loops.

My AI approach:

• I simulate user conversations based on collected data and updated personas.
• Use AI chatbots trained on user pain points to test new features or messages.

Prompt sample:

“Based on this persona and recent complaints, simulate a conversation where they react to our new onboarding flow. What questions do they have? What might they drop off from?”

✅ Result: Preemptive UX testing, before wasting dev hours.

🧭 Bonus: My “5x5x5 Research Cadence

To keep things practical, I apply a lightweight cadence in team’s calendar:

• 5 hours/month of raw input gathering (reviews, support, community)
• 5 prompts/week to synthesize and test user insights
• 5 minutes/day to log findings into a rolling Notion or Coda doc

This is manageable solo — and compoundingly valuable.

🧠 Systems, Not Just Tools

Tools come and go. The system stays.

Here’s the mental model behind everything I do:

Most PMs stay at the signal layer. My approach moves through all three — fast.

The best PMs aren’t just user-centric. They’re system-centric about users. This is mine. Now show me yours.

👊 The Challenge

If you’re building a product with limited bandwidth, zero dedicated researchers, and high uncertainty — this method is designed for you.

It’s not perfect. But it’s fast, repeatable, and grounded in actual user voice. And it lets you ship products that land closer to real human needs — not assumptions.

📎 TL;DR: My AI-Driven Research Stack

Tool / Resource: Purpose

ChatGPT / Claude: Analysis, synthesis, persona creation

Klu / Glean / Notion AI: Highlighting and summarizing user data

Pinecone / Weaviate: Vector search across qualitative content

Typeform + GPT: Smart survey summarization

Notion / Coda : Continuous knowledge base of user insights

📬 Want to Collaborate?

I’m sharing this to start better conversations. If you’re a startup founder, product leader, or researcher who wants to jam on AI x product discovery, I’d love to chat.

Let’s build smarter, faster, and closer to real human needs.

If you want the prompts, dashboards, or tools I use — happy to share. Just message me.

This guide maps the three stages of user research based on your AI fluency: Capable, Adoptive, and Transformative. For each level, we’ll explore:

• Best-practice setups
• Automation examples
• Agent-based workflows
• Strategic risks

Stage 1: Capable — Basic AI Adoption

Goal: Use AI to reduce repetitive, low-leverage work.

Recommended Tools

• Otter.ai
• Notion AI
• Calendly
• Typeform
• ChatGPT
• Zapier

Automation Examples

• User Interview Notes: Transcribe interviews with Otter.ai, then summarize and extract insights using ChatGPT:

Prompt (Interview Synthesis):
“Here’s an interview transcript. Extract: 1) main user goals, 2) pain points, 3) emotional tone, 4) any quotes worth adding to a slide deck.”

• Survey Theme Detection: Export Typeform to Google Sheets, use AI to cluster answers into sentiment and feature themes.

Prompt (Survey Theme Clustering)
“Analyze this sheet of open-ended survey responses and group them into common themes, giving each a title and description.”

Best Practices

• Start with 1–2 bottlenecks (e.g., transcription, synthesis).
• Use no-code automation like Zapier to connect tools.
• Always review AI output manually to catch hallucinations.
• Prioritize time-saving, not “perfect” accuracy.

Hard Truth:

If you’re still summarizing interviews manually, you’re burning time better spent talking to more users.

Stage 2: Adoptive — Integrated Multi-Tool Workflows

Goal: Create automated, multi-step research pipelines.

Recommended Tools

• Dovetail (Prompt (Insight Extraction) “From these tagged highlights, create a 1-page research report with: 1) top themes, 2) example quotes, and 3) product suggestions.”
• Maze
• Hotjar
• Amplitude
• UserTesting
• Make.com

AI Agent Workflows

• Behavioral Analysis: Hotjar sessions → AI agent detects rage clicks/dead zones → auto-generates hypotheses → sends to Maze for A/B testing.

Tool: Hotjar + ChatGPT Prompt (Session Behavior Analysis)
“Review this list of session behaviors (rage clicks, scrolls, exits). Identify common usability issues and suggest a hypothesis for A/B testing.”

Tool: Maze Prompt (Usability Test Debrief)
“Summarize the main issues found in this usability test and recommend changes to improve task completion.”

• Research Pipeline: Schedule participants → conduct & transcribe interviews → cluster insights → auto-generate insights deck in Notion.

Tool: Make.com (Orchestrator) Prompt (Trigger Setup):
“When a new insight is created in Dovetail with the tag ‘frustration,’ auto-generate a Slack alert and add the theme to Notion’s research log.”

Agent Setup Guidance

• Build role-based agents: Interview Assistant, Theme Synthesizer, Insight Communicator.
• Set up data validation checkpoints after each major synthesis step.
• Use Make.com or Zapier to create robust workflows with fallback alerts.

Strategic Advice

• Layer human verification into every major step.
• Prioritize auditability — understand what your AI is doing and why.

Hard Truth:

AI without orchestration is a bunch of smart tools waiting to break in unpredictable ways. Systems thinking is your multiplier.

Stage 3: Transformative — Autonomous Research Ecosystems

Goal: Let AI continuously discover, test, and validate hypotheses — with humans steering strategy.

Advanced Tool Stack

• Claude or GPT-4
• Synthetic Users
• Custom APIs
• Quantum Metric
• DataRobot
• n8n

Example: Autonomous Discovery System

• Quantum Metric detects user frustration.
• Research Planner agent designs a micro-study.
• Synthetic Users simulate usability tests.
• GPT-4 generates insights & recommends changes.
• Slack alerts product team with action-ready report.

Tool: GPT-4 or Claude Prompt (Autonomous Discovery Agent)
“Monitor usage logs for anomalies or frustration signals. When 3+ signals match, design a micro-study to validate a root cause.”

Tool: n8n Prompt (Orchestrated Feedback Loop)
“Build a loop: app data → usage drop → insight agent triggers → study results → product change tracked → loop restarts with outcome tracking.”

Agent Ecosystem Roles

• Research Planner: defines scope, goals, population.
• Data Collector: auto-fetches user behavior and interviews.
• Synthesis Agent: clusters data and identifies key findings.

Tool: Synthetic Users Prompt (Scenario Simulation)
“As a synthetic user persona (new to finance apps, age 34, low tech fluency), evaluate this onboarding flow. Highlight friction and emotions.”

• Communication Agent: formats insights into decks, emails, Notion.

Tool: Custom Multi-Agent Setup
Prompt (Research Planner Agent)
“Given this product roadmap and recent NPS dip, suggest three user research studies to uncover causes and opportunities.”
Prompt (Insight Communicator Agent)
“Take the research synthesis document and convert it into a 5-slide exec briefing. Include 1 quote per slide and a TL;DR per insight.”

Strategic Risks

• Tooling complexity → brittle workflows unless modularized.
• Over-reliance → insight without context or emotional nuance.
• Data governance → privacy and consent must be built-in.

Hard Truth:

Without governance, you’re not building AI research — you’re building a liability machine.

User Research Prompting Tips

• Use roles and personas. Example: “You are a senior UX researcher…” or “Act as a frustrated new user of a budgeting app…”
• Structure your outputs. Ask for: bullet points, tables, executive summaries.
• Iterate. One prompt rarely solves everything. Use refinement prompts like: “Now condense this into a one-pager” or “What are the top 3 takeaways?”
• Always validate AI output with human review. Even at the transformative level.

🧭 Next Step: Build Your Prompt Playbook

Create a shared Notion or Google Doc titled:
🧠 Prompt Playbook — [Your Company/Product Name]

Structure it like this:

Final Thoughts: Don’t Just Adopt Tools — Design Intelligence

Most teams adopt AI as another SaaS tool. That’s a dead-end. AI isn’t a product — it’s a capability. And like any capability, it must be designed, managed, and improved.

Start with a system. Think in agents. Build in loops. Insert oversight. Move fast, but verify.

Call to Action

In the next 30 days, audit your research stack. Identify your top 2 time sinks. Build a Capable-level AI workflow. Then start designing your Adoptive-level agent chain. When you’re ready, upgrade to Transformative orchestration.

Let’s build the future of product research — faster, deeper, and smarter.

References & Further Reading

• Accelerating Research with AI (Nielsen Norman Group)
• Fast-track your research with Hotjar AI
• What is automated user testing (Dovetail)
• n8n Agent Workflows
• Hallucinations (Wikipedia)

In the competitive arena of modern software development, time has become the principal currency of innovation. Product managers (PMs), historically regarded as bridge-builders between engineering and users, are increasingly under pressure to deliver strategic insight — not merely feature checklists. At the heart of this transformation lies a shift in how user research is conducted: from manual, anecdote-driven practice to automated, AI-assisted systems of pattern recognition and validation.

In the past, user research was synonymous with qualitative interviews, endless transcription, laborious clustering of sticky notes, and often a gap between insight and execution. In 2025, AI has begun to redraw this map.

The Changing Imperatives of Product Management

Where once product decisions could be based on intuition supplemented by quarterly surveys or occasional customer interviews, the modern PM operates in an environment of continuous discovery. For businesses — particularly in SaaS, fintech, and consumer platforms — the velocity of insight now determines whether they lead or follow.

This is especially true in the prototyping phase, where product ideas are still fluid and risk of misalignment with user needs is high. To mitigate this, PMs must leverage AI-enabled research workflows capable of integrating qualitative depth with quantitative breadth.

A Five-Phase Framework for AI-Enabled User Research

1. Strategic Planning

The foundation of any research initiative must be a clearly articulated hypothesis. AI tools such as ChatGPT-4.5 and Claude 3 can assist in refining hypotheses, crafting interview guides, or simulating stakeholder reactions. Modern PMs are encouraged to anchor research around frameworks like Jobs-to-Be-Done (JTBD), which frame product functionality in the context of user progress, rather than demographics or personas alone.

Tools: Notion AI, Miro, Airtable.

2. Participant Recruitment and Sampling

In B2B contexts, research velocity is constrained by access to niche user groups. Tools such as User Interviews, Respondent, or Ethnio offer CRM-integrated recruitment platforms that can target participants by role, company size, and digital behavior.

For B2C, platforms such as Prolific, PlaybookUX, and Maze Panels automate screening, outreach, and scheduling, often within hours.

Best practice: Maintain a research CRM (e.g. Rally UX, ConsentKit) to track past participation and prevent repeat sampling.

3. Data Collection Across Modalities

Modern user research is no longer a binary choice between surveys and interviews. The most robust insight pipelines integrate:

• In-depth interviews (moderated or AI-led)
• Surveys (for attitudinal clustering and prioritization)
• Intercepts (real-time behavioral questioning inside the product)
• Mobile ethnography (diary studies)
• Session replays and heatmaps (passive behavioral analysis)

AI tools like Outset.ai now conduct and synthesize interviews autonomously, generating not only transcripts but also sentiment maps and quote clustering. For B2C, tools like Sprig, Hotjar, and dscout provide behavioral validation at scale.

4. Synthesis and Sensemaking

Traditionally, synthesis was a manual bottleneck in product research. The cognitive tax of pattern extraction — clustering quotes, surfacing contradictions, identifying priority themes — meant that insight lagged far behind data collection.

Today, Dovetail, Condens, Notably, and Aurelius provide AI-assisted tagging and clustering capabilities. When layered with LLMs like Claude or GPT, they allow PMs to prompt:

“Identify the three primary user frustrations from these interviews and recommend product opportunities backed by evidence.”

This makes it possible to transform raw data into directionally useful insights within hours, rather than days.

5. Insight Activation

Perhaps the most overlooked element in traditional user research is the activation loop: how insights inform decisions. AI-enabled PMs are not content with dashboards — they integrate insight outputs directly into product decision-making artifacts:

• PRDs generated from tagged insight cards
• Roadmaps scored by insight confidence
• User quotes embedded in Jira or Productboard

Productboard, Coda, Notion AI, and Airtable + GPT enable this transition, creating traceability from quote → insight → requirement → feature.

The Rise of the “Insight Flywheel”

Among elite product organizations, the research function is no longer episodic — it is continuous and self-improving. This insight flywheel can be described as:

1. Frame (Hypothesis, audience, scope)
2. Collect (AI-enhanced interviews, surveys, passive data)
3. Synthesize (LLM pattern detection, quote extraction)
4. Activate (PRD creation, roadmap scoring)
5. Adapt (Feedback from shipped features loops back to Frame)

Time to insight, not time to deliverable, is now the key performance metric.

Strategic Implications for PMs and Organizations

The distinction is stark. Companies still relying on outdated research methods will find themselves outpaced — not by larger competitors, but by faster ones.

Cautionary Notes

• Over-reliance on automation can introduce bias if qualitative nuance is ignored. Human review of AI-generated themes is essential.
• Ethical considerations around AI-led interviews, data consent, and interpretability must be embedded in research design.
• Tool fragmentation must be managed via strategic selection — integrations between research, synthesis, and roadmap tools are critical.

A Sample Stack (B2B vs B2C)

Conclusion: AI as Strategic Amplifier

AI does not replace the product manager’s judgment, but amplifies it. It accelerates feedback cycles, deepens evidence, and allows for higher-quality decisions made at speed. The modern PM must be as fluent in user signals as they are in product metrics, and AI is no longer optional — it is foundational.

Those who fail to embrace this shift risk building in the dark. Those who master it will outlearn, outrun, and outmaneuver their competitors.

And that, in the world of product, is how empires are built.

<🎤 Narrator Voice>

Ladies and gentlemen of the modern #workforce,

Use large language models aka #LLMs

If I could offer you only one tip for the #future, #AI would be it.

The long-term benefits of LLMs have been proven by #engineers, #researchers, and every SpaceX Tesla. #DOGE intern with a #MacBook.

Whereas the rest of your productivity advice has no basis more reliable than a TikTok. hustle bro’s vibe check.

<🧠 Verse 1>

Use #AI

Even if your job is “mostly meetings,”

Even if your team still uses Microsoft. #Excel to manage roadmaps.

Trust me, in 5 years,

You’ll look back at 2025 and say,

“Why the hell didn’t I have Anthropic. #Claude write that email?”

<💼 Verse 2>

Don’t waste time arguing with middle managers who think #AI is “just a phase.”

The productivity wave will sweep them aside like #Blockbuster in a Netflix. world.

Don’t feel guilty if you don’t know exactly what a #Transformer model is.

Most #VPs don’t either.

But #Claude does.

<⚙️ Verse 3>

Work hard. But not too hard.

Outsource the thinking that doesn’t make you smarter.

Summarise. Rewrite. Draft.

#Claude doesn’t sleep, and doesn’t complain about Atlassian. #Jira tickets.

<🗣️ Verse 4>

Maybe you’re a #Product #Manager.

Maybe you’re a content #writer.

Maybe you’re a #CTO who secretly Google. searches “what does LLM stand for.”

It doesn’t matter.

If you touch words, you should touch #AI.

<🌍 Verse 5>

Don’t envy the 22-year-old with five Notion. workflows and a productivity Youtube. channel.

He’s probably using #ChatGPT to script it anyway.

Instead:

Use the #AI.

Write better. Think clearer.

And for the love of Turing,

Never manually rewrite a product spec again.

<☀️ Outro>

Use #Claude.

Use OpenAI. #GPT.

Use whatever doesn’t make you sound like a corporate caveman.

Because in the end, we’re all just trying to keep up…

…and the only thing worse than being replaced by #AI

is being replaced by someone who knows how to use it.

Use #AI. Or Stay #Dumb.

🎹🧴🎶 https://music.youtube.com/watch?v=2CwxfaSFzRs&si=oW4iXK8PzcGtZqeu

— -// — -(Pianississimo) — -// — -

Dear #professionals over 35 still clacking away at keyboards like it’s 1997:

Anthropic. #Claude isn’t just an #AI — it’s the coworker you wish you were.

If you’re a #PM, #analyst, or just someone with a pulse and a #deadline, and you’re not using an #LLM to think better, faster, and <insert_own>…

That’s not “old school.” That’s just “old.”

It’s not a tool. It’s a sin not to use it.

🫡 Kudos, Anthropic for showcasing the raw, radiant beauty of intelligence — digital and otherwise.☀️

#ProductManagement #ModernWork #LLMsAreTheNewExcel

©️ https://youtu.be/oqUclC3gqKs?si=tg6PzqxquonYRYnb