Most days, for better or worse, I find myself wondering either if it's time to pivot or questioning the integrity of a pivot. Let's unpack the pivot.

What Even Is a Pivot?

At its core, a pivot is a shift in direction based on a fresh understanding of your product, market, or model. It’s not a rebrand, and it’s not panic—it’s a new bet based on an updated signal. Maybe your original assumptions didn’t hold. Perhaps you uncovered something better. But the essence is this: same mission, different motion.

Here are some famous examples:

• Slack started as a gaming company before pivoting to team chat.

• Instagram pivoted from a check-in app (Burbn) to photo sharing.

• YouTube pivoted from a dating site to video sharing for everyone.

• Twitter pivoted from a podcast network to microblogging. These pivots all kept the core mission of connecting people, but found stronger product-market fit in new directions.

Three Failure Points That Trigger Pivots

Product Can't Deliver Compounding Value

Maybe users don’t stick around long enough, or every feature feels like a one-off lift. If each update wins you marginal gains but never momentum, you’re not building a compounding product—you’re building busywork. In an AI era, that also includes infrastructure gaps: if rate limits, hallucinations, or fragile APIs choke your roadmap, the real value may be out of reach for now.

Market Is Inaccessible, Uninterested, or Undefined

Even if the need is real, if you can’t find your audience or they won’t pick up the phone (or click the link), you’ve got a problem. Markets that feel theoretically big often don’t map to a reachable, pain-aware buyer. Early adopters might love the vibe but not the price, timing, or risk.

Business Model Can't Scale Profitably

The engine might start, but it stalls before you hit cruising altitude. If your CAC trends up while LTV flatlines, or if every growth loop breaks under pressure, it’s time to rethink. Great tech doesn’t matter if it dies under margin pressure or operational overhead.

Vibe Founder Updates

Product

Sometimes the product isn't the problem—it's the stack beneath it. For Vibe Founders, product pivots might be driven by platform realities, not just user demand.

• You’ve built something powerful, but OpenAI rate limits make it unusable at scale

• Hallucinations break trust in your AI-powered UX

• The API you rely on gets nerfed or paywalled

• Your “platform vision” spreads too thin without one compelling, repeatable wedge

Market

The best ideas still die if you can’t sell them. Your market may be a myth—or just too early.

• You can’t define your buyer persona in a sentence

• Horizontal use cases create too many messages, not enough traction

• Your TAM looks great in a pitch deck, but no one is converting

• You’re burning cash chasing attention with no clarity on who wants it

Fit

Your business model needs to work before the money runs out. Economics are louder than vibes.

• CAC is high and LTV is theoretical

• Users love the free tier, but ghost when it’s time to pay

• A real demo requires three explainers and a founder walk-through

• Hype masks that the flywheel isn’t turning—yet

Signal Checklist

1. You’ve got a feature trap or infrastructure stall

2. You can't reach or define a buyer persona

3. Your cost-to-profit story has more fiction than finance

Hit two out of three? Don’t wait for the market to tell you. Pivot while you still have dry powder and optimism.

Prompts

Stress-test the Product Surface

“You are a brutally honest PM. List the top 5 workflows our product *must* own to deliver 10× value over current tools. Flag any that drift outside our core promise.”

Red-flag the Market Size

“As a go-to-market strategist, estimate the reachable TAM for <product>. Highlight hidden acquisition costs and how they might scale over 12 months.”

Diagnose Model-Fit Gaps

“You are my CFO. Given our current CAC, churn, and ARPU, simulate three scenarios (status quo, price bump, freemium) and show months to cash-flow break-even.”

TL;DR

• Pivots are proactive course corrections, not panic buttons.

• Watch for product surfaces that are too broad (platform) or too thin (feature).

• Markets that are invisible, microscopic, or amorphous kill momentum; define and target ruthlessly.

• If unit economics refuse to improve, your model—or audience—is wrong.

• Nail two out of three danger signals? Draft your next, sharper hypothesis and pivot before entropy does it for you.

Every startup talks about ICPs—short for Ideal Customer Profiles—but let’s be real, most of us wing it. We think we know our customer, and we might even have a Notion doc to prove it. But knowing of your customer isn’t the same as knowing how to find, target, and win them.

An Ideal Customer Profile (ICP) is a clear, data-informed picture of the kind of customer who is most likely to buy, stay, and spread the word. It’s not just a pretty persona slide for the deck. It’s your operating system. It decides what you build, how you talk, and where you grow. And when it's vague? You burn cash, chase bad leads, and lose your edge.

Need a deeper dive? Read our previous post:
https://paragraph.com/@vibefounders/just-in-time-the-ideal-customer-profile

Get Your ICP Planner

We've made the ICP Planner available as a Google Sheet for you to copy and use on your own.

ICP Planner

This browser version is no longer supported. Please upgrade to a supported browser.

http://docs.google.com

How to Use the ICP Planner

The ICP Planner is a worksheet that helps you organize, research, and prioritize your Ideal Customer Profile. It’s been tested across decades and dozens of startups.

Sheet 1: Niches

This is where you zoom out. It helps you map the landscape of possible customer groups. You might feel like you're building for the masses, but that's a trap. You win by starting small and loud: identify a niche that deeply cares, win them completely, then use that momentum to expand.

Market

Track all the niche markets that might benefit from your product. Be specific—lots of tightly-defined niches are better than a few broad groups.

Opportunity

These columns help quantify your opportunity from different angles.

Value

For TAM, a bottom-up approach paints a clearer, more grounded picture. Here’s how the rest of the model flows:

Next Steps

Once you complete the worksheet, you’ll have a strong picture of your most promising markets. Consider:

• Small customer numbers = fewer at-bats

• Big TAMs don’t mean real demand

• Lower complexity is easier to sell, but may bring less value

• Focus on 2–4 niches and collect real-world signals

Sheet 2: Signals

This sheet organizes insights from interviews or community observation to validate assumptions.

Your job is to collect as many real signals from real users as quickly as possible.

Sheet 3: ICP

This tab pulls together all your research to reveal your most aligned, high-value customer segments.

Key Caveats

• Directional, not definitive—validate with real users

• TAM ≠ demand

• ACV can be inflated by optimism—check against reality

• Observed behavior > claimed preferences

• Your ICP will evolve—check back often

Vibe Founder Prompts

Modern founders can go even faster by pairing the planner with AI. Here’s how:

Find niche audiences + TAM

What are 10 potential niche audiences for a [product or feature]? For each, estimate their size (TAM) and suggest a method to reach them.

Identify relevant spending behavior and budgets

For this niche audience [insert niche], what is their typical budget range for tools like [product type]? What do they currently spend on comparable solutions?

Generate interview questions tailored to niche segments

Write 5 high-signal customer interview questions for [niche audience] that will uncover their pains, motivations, and decision-making process.

Curate pain points from Reddit and other communities

Find top Reddit threads and community posts where [niche audience] discusses their pain points with [problem area]. Summarize the top 5 unmet needs mentioned.

These prompts turn AI into your fastest researcher, so you can spend less time guessing and more time building.

TL;DR

• Most founders treat ICPs like a branding exercise—it’s a growth engine

• The ICP Planner helps turn vague ideas into signal-rich customer profiles

• It models TAM, ACV, complexity, conversion, and revenue potential for each niche

• Use interviews and observation to validate real customer pain and intent

• Layer in AI to accelerate research and uncover insights from the web

“Speed is the only moat” - Varun, Co-founder Windsurf

What Are Moats?

A moat is what protects your business from competitors once you start winning. It’s the thing that makes it hard to copy you—or at least, not worth the effort. Just like a castle’s moat kept enemies out, a startup’s moat keeps disruption out.

What Are Good Moats?

Classic moats include network effects (your product improves the more people use it), proprietary technology or intellectual property (you've built something others can’t legally replicate), brand equity (people trust you more than a new player), and economies of scale (you can offer more for less). These all take time, traction, and capital to build—and most of them show up after you have product-market fit, not before.

Why Investors Care About Moats

Investors are betting on your ability to win and keep winning. A good moat means lower risk and higher upside. It signals that you’re not just a flash in the pan. If you don’t have a moat yet, they want to at least see a plan for one.

But here’s the truth: in early-stage companies, your moat is usually a mirage. So what investors are really looking for is: can you move faster than the next ten clones?

Do Customers Care About Moats?

Nope. Not consciously. But they do care about what moats produce: better products, lower costs, more value, and faster updates. Customers don’t say, “I use this because of their strong defensibility strategy.” They just say, “This works better for me than anything else.”

Speed Is The Only Moat

Here's the real shift: speed is your moat. Not later—right now.

You're not trying to out-patent or out-brand anyone in the early days. You're trying to out-learn, out-ship, and out-adjust. AI just cranks that dial up.

Anyone with a weekend and GPT-4 can clone the basics. Your edge? You iterate faster than they can even organize a Figma file.

This isn’t about breaking stuff. It's about learning in public. Every ship is a rep. Every tweak builds compound insight. That's your wall. That's your moat.

And let’s be honest—if you’re using AI and still slow? That’s not a tech problem. That’s your problem. Push. Post. Pivot. Repeat.

Prompts To Try

Use these prompts to operationalize your speed-as-moat advantage—and get smarter about how your competitors are defending their turf:

Identify your laggards

Find slow spots in your startup and speed them up.

Audit my current workflows. Where am I wasting time that could be saved using AI, automation, or delegation?

Design a speed sprint

Create a 3-day velocity experiment using AI.

Help me plan a 72-hour sprint where I ship 3 versions of my idea using AI tools. Include launch, feedback, and iteration steps.

Test your defensibility

Pressure-test your moat by simulating a copycat.

Act as a fast-moving competitor. Rebuild my startup’s core features using free tools and AI. What would you launch in a week?

Understand competitor moats

Break down what makes your competitors hard to beat.

Give me a breakdown of the likely moats my top 3 competitors have. What are they doing that keeps others out?

Analyze industry defensibility

Explore how moats work (or don’t) across your whole market.

What types of moats are common in my industry? How do they typically form, and where are they weakest?

TL;DR

• A moat protects your business, but in the early days, speed is the only one you’ve got.

• You’re already building in public. AI just makes the loop faster.

• Speed up your work, study your rivals, and stress-test what’s defensible with these prompts

If your startup is a vehicle for delivering customer value, then the Ideal Customer Profile (ICP) represents the customer, and the product is the value. The ICP is a sketch of the people who buy quickly, stick around, and tell their friends.

Most importantly, not every customer is your ideal customer. Without an ICP, you waste time chasing bad fits, overbuilding features, and confusing your team. With one, you know who to target, what to build, and how to grow with intent.

You don’t discover your ICP in a workshop — you earn it. You learn by shipping, selling, and listening. You start with a guess, then tighten it through patterns in usage, retention, and feedback. This is where AI shines: spotting the signals, clustering the cohorts, and showing you who’s really in your corner.

Let's explore more about your ideal customer.

“Every customer request falls into one of three buckets: Yes, No, or Maybe. The trick is getting good at sorting quickly.”

Timeless

Founders have always needed to triage customer feedback to maintain focus and momentum.

Timely

AI tools can now assist in categorizing and prioritizing customer requests based on alignment with product vision and potential impact.

Try This

SYSTEM
You are “FeedbackSorter,” an AI assistant for product managers.

USER
Here is a list of recent customer feature requests. Categorize each as 'Yes', 'No', or 'Maybe' based on our product roadmap and strategic goals. Provide a brief justification for each categorization.

[PASTE CUSTOMER REQUESTS]

“Saying yes to the wrong things can slow you down more than saying no ever will. You burn cycles. You confuse your team.”

Timeless

Misaligned feature development can derail progress and dilute product focus.

Timely

AI analytics can predict the potential ROI and resource requirements of proposed features, aiding in informed decision-making.

Try This

plaintextCopyEditSYSTEM
You are “FeatureEvaluator,” an AI tool for assessing feature requests.

USER
Analyze the following feature requests and estimate their potential ROI, development time, and alignment with our core product objectives. Recommend whether to proceed, defer, or decline each.

[PASTE FEATURE REQUESTS]

“The ‘maybe’ pile is where pivots are born.”

Timeless
Ambiguous feedback often holds insights that can lead to significant product evolution or pivot opportunities.

Timely
AI clustering algorithms can identify patterns in 'maybe' feedback, revealing emerging customer needs or market trends.

Try This

plaintextCopyEditSYSTEM
You are “InsightMiner,” an AI assistant for uncovering patterns in customer feedback.

USER
Analyze the following ambiguous customer comments and identify common themes or unmet needs that could inform potential product pivots.

[PASTE CUSTOMER COMMENTS]

“Not every customer is your customer. That’s not arrogance — it’s survival.”

Timeless
Recognizing and focusing on your Ideal Customer Profile (ICP) is crucial for sustainable growth.

Timely
AI-driven segmentation can help identify and target customers who align closely with your ICP, optimizing marketing and product development efforts.

Try This

plaintextCopyEditSYSTEM
You are “ICPAnalyzer,” an AI tool for customer segmentation.

USER
Based on our existing customer data, identify the characteristics of our Ideal Customer Profile and suggest strategies to attract and retain similar customers.

[PASTE CUSTOMER DATA]

This post is a part of the “Just in Time” series, where I revisit and adapt my experiences and advice for founders from the past 20 years. It is based on “On Customers: Just Say No, Or Yes, Or Maybe”, originally published on June 29, 2017.

“A head start is rarely large enough to matter, and time spent in stealth mode-away from customers-is unlikely to provide a head start. The only way to win is to learn faster than anyone else.”

— Eric Ries

Five years ago, if you told me you were going into stealth mode, I’d assume you were either overestimating your idea or underestimating your market. Most people simply don’t care enough to steal what you haven’t proven works. And yet, the myth persists: build in silence, launch with fireworks, and watch the world gasp in admiration.

That narrative is seductive, but misleading. It conflates secrecy with strategy and overestimates the power of surprise in a world where iteration, not inspiration, tends to win.

A lot of Vibe Founders build in public—but that's correlation, not definition. The vibe isn’t about being loud, it’s about being in tune. Sometimes that means working in the open; other times, it means staying heads-down until the work is too sharp to ignore. In an AI-first world, though, the bar for originality has dropped. A thin idea has gotten even thinner. If you’re not careful, your vibe can get scraped, cloned, and commoditized before it ever compels. So yes, expose yourself to feedback. But not before you've built something worth reacting to.

Modes of Building

1. Stealth Mode

No website. No social. You're effectively off the grid.

Why founders choose it:
They're scared of being copied. They think they have a secret insight. They want to look legit before showing anything.

Why it fails:
You’re not testing anything. You’re not learning. You’re not talking to users. So, your big secret? It’s probably wrong, and no one’s around to correct you.

2. Open Mode

You’re sharing, but quietly. A few trusted users. Maybe some early testers. A Notion doc in a DM.

Why it works:
You’re validating without broadcasting. It’s controlled exposure. You’re getting smarter without giving up the whole story.

Where it breaks:
You can get stuck. Not visible enough to attract momentum, not focused enough to move quickly. Half-visible = half-effective.

3. Public Mode

You’re showing your work. Screenshots. Threads. Changelogs. Community.

Why it works:
You’re building in public, and people rally around the journey—that momentum compounds distribution, feedback, and even hiring.

Where it breaks:
It’s noisy. You might overfit too early hype. Worse, competitors could fast-follow your positioning before you’ve nailed the product.

AI Changed the Equation

AI didn’t just speed up development. It collapsed the cost of copying. Your idea, once surfaced, can be replicated, remixed, and rebranded in minutes. The marginal effort to duplicate a startup has never been lower.

That doesn’t make stealth more valuable.

It makes speed and defensibility more valuable. It raises the bar for how clearly you understand the problem. It forces you to create moats early, through data, distribution, community, or workflows that can’t be easily cloned.

And it amplifies the cost of delay. Because while you were hiding, someone else was learning. And learning, not secrecy, is what wins.

When (and Why) Stealth Makes Sense

Stealth isn’t always a mistake. But it should never be the default.

Use it if:

• You’re in a regulated space and need time to navigate compliance.

• Your insight is deeply technical, and the build cycle is long.

• You’re pursuing IP and need the legal clock to tick before the press release drops.

• Your main competition is other nimble startups

• You’re building a product with minimal initial moat and want time to harden the edges before the copycats show up.

In the AI era, even the thinnest ideas can look like viable startups. If you haven’t tested the problem deeply or figured out what makes you different, stealth might buy you time to do that — just don’t mistake it for a shield.

Even then, be surgical. Stay stealthy with the press, not with your users.

AI Prompts to Build Your Moat

If you are operating in stealth—or preparing to go public—use AI to identify and reinforce your unfair advantage.

Moat Mapping

List the defensible components of my startup idea, and suggest strategies to strengthen them before launch. My idea: [your 1-paragraph description]

Replication Risk Audit

What parts of this startup are most vulnerable to AI replication? Suggest ways to mitigate or obscure them before public reveal.

Pre-Public Moat Builder

Given this concept, how can I create data, workflows, or user feedback loops that are hard to clone, even with access to the public site?

TL;DR

Stealth mode is a luxury most startups can't afford.

You don’t need to be secret. You need to be fast, exposed, and learning.

Ship fast. Talk to users. Build in public, if you can. If you can’t, know exactly why, and have a plan to exit stealth before it becomes your default state.

Because in the AI era, the real threat isn’t being copied. It’s being too slow to matter.

Vibe Founders have many different goals:

• Making enough to support yourself and your passions

• Building a small team and business to support your lifestyle

• Shooting for the moon to create a unicorn business

However, our goals don't change our need to understand the fundamentals of our business. Fortunately, there are well-established methods for thinking about your market, aka who you're selling to. Here's a quick primer.

What is TAM / SAM / SOM?

Let’s start with the basics — because way too many founders still wing this.

• TAM is Total Addressable Market: the big, top-line number. If every possible customer bought your product, how much money could you make?

• SAM is Serviceable Available Market: a slice of the TAM that your current business model, features, or geography can actually serve.

• SOM is Serviceable Obtainable Market: your real near-term opportunity — based on go-to-market strategy, resources, and traction.

TAM is about vision. SOM is about execution. You need both. But you especially need TAM if you ever want to raise money, enter new markets, or just to sanity-check your roadmap.

Computing TAM

There are two main ways to calculate TAM, and both have their pros and cons.

Top-Down TAM

This is the classic “analyst report” approach. Start with the entire industry’s spend — then narrow it.

• Step 1: Find a big, reputable number — industry size, annual spend, etc.
  Example: “$200B spent annually on health & wellness.”

• Step 2: Apply filters — geography, demographics, behavior.
  Example: “We’re targeting U.S. men aged 25–45 who use supplements = $10B.”

• Step 3: Apply assumptions — e.g., what % will switch, what % we can reach.
  Example: “Assume 10% adoption → $1B TAM.”

This is great for fundraising. It shows scale, narrative, and optionality if you need to pivot. But it’s easy to BS — and investors know it.

Bottoms-Up TAM

This is the builder’s approach. Start with what you sell and who you’re selling to.

• Step 1: Identify your ideal customer and their spend.
  Example: “Our product costs $50/mo. Our early users are freelance designers.”

• Step 2: Estimate how many of those customers exist.
  Example: “There are ~5M freelance designers globally.”

• Step 3: Multiply: $50/mo × 12 × 5M = $3B TAM

This method forces clarity: you need to define your customer, your pricing, and your assumptions. It’s grounded. It’s honest. It’s usually smaller — but that’s a good thing. It helps you make better decisions faster.

Defining TAM

Now that we have the fundamentals in place, let’s walk through a couple of examples.

Example 1: Top-Down (Consumer Fitness App)

• Start with: $150B global fitness market

• Focus on: smartphone users aged 20–40

• Narrow to: digital fitness subscriptions ($15B)

• Target: English-speaking markets = $5B

• Assume: 5% market penetration = $250M TAM

Example 2: Bottoms-Up (Freelancer CRM)

• Product: $30/month tool for solo consultants

• Target audience: ~10M globally

• Estimate TAM: 10M × $30 × 12 = $3.6B

• SAM (realistically reachable users): maybe 1M = $360M

• SOM (first 12 months): 10K users = $3.6M

Prompts to Help You Do This With AI

Top-down TAM prompt:

scssCopyEdit"Help me estimate the top-down TAM for [industry]. Start with global market size, then narrow by geography, demographics, and known spend."

Bottoms-up TAM prompt:

cssCopyEdit"Based on [my product], who is my most likely initial customer segment? Estimate size, CAC, and potential ARPU to calculate bottoms-up TAM."

Comparison prompt:

arduinoCopyEdit"Compare top-down and bottoms-up TAM for my startup idea: [insert idea]. Highlight risks and blind spots in each."

Deck-ready prompt:

cssCopyEdit"Create a TAM slide for a pitch deck using both top-down and bottoms-up methods. Assume I’m targeting [X niche in Y market]."

TL;DR

• TAM = Total Addressable Market (big picture),
  SAM = Serviceable Available Market (reachable now),
  SOM = Serviceable Obtainable Market (realistic near-term win).

• Top-down TAM starts with industry size and narrows from there. Great for pitching. Easy to fudge.

• Bottoms-up TAM starts with your product, pricing, and customer count. Great for planning. Harder to fake.

• Use both: top-down for vision, bottoms-up for reality.

• Top-down is best if you're fundraising, bottroms-up is always useful

• Let AI help — it can size markets, suggest segments, and pressure-test your assumptions in minutes. Just prompt it smartly.

Over the past 30 years, I’ve founded a dozen startups, from scrappy bootstraps to venture-backed moonshots. I’ve worn every hat—designer, engineer, product lead. I’ve had wins and more than a few losses. Each one left me with lessons worth sharing.

For the last 20 years, I’ve tried to do just that—turning my experiences into insight through blogs, podcasts, live streams, and talks. I’ve always believed: if I can’t compress it into something useful, I haven’t really learned it yet.

"Vibe Founders" is what I call this new wave of founders. More intuitive. More experimental. More AI-native. This publication is where I’ll explore what it means to build in this moment, where everything is being rewritten, but remains strangely familiar.

About Me

Gregarious Narain is a serial entrepreneur and founder, having founded more than a dozen companies over the past 25 years.  His most recent company, Chute (YC W12), raised $16MM, grew to millions in revenue with 50+ employees spanning 6 cities, before being acquired in 2018.  Prior to Chute, Greg was the first employee and V.P. Product at Klout.

Greg’s diverse background gives him a unique point of view on the startup experience.  A designer and developer by trade, Greg straddles the technical, product, and business realms to help align businesses around a common, customer-driven vision.  Greg’s no-nonsense, outside-in approach to entrepreneurship has helped startups and the Fortune 500 alike understand and achieve the success they need. 

Greg has been a creator for more than a decade, starting with his blog Social Twister in 2003.  He was an early podcaster, launching the first syndicated, group podcast - Beercasting in 2004.  Although he writes frequently, his true passion is live content.  He has produced a number of live streaming shows, including his 2 most recent, Get Unstuck, a daily office hour show for founders and entrepreneurs on Clubhouse and The Created Economy, a weekly audio/video live stream covering the creator economy.

Greg has spoken internationally at many leading industry events, including Cannes Lions Innovation, DreamForce, NAB, SXSW, AWS Re:Invent and mentored for programs including YC Startup School, Google Launchpad, TechStars and Founder Institute.

Let’s examine the original thoughts and how to adapt them to today.

“Product Market Fit. That mythical unicorn all startups spend tireless nights and weekends toiling over greasy keyboards and greasier food.”

I first wrote that on March 30, 2017. Has much changed in how founders seek out and attain Product-Market Fit in the age of AI?

Timeless

While it’s become easier than ever to discover what people think and care about (we share everything with immense volume on multiple platforms), it’s still, arguably, pinning down what people will pay for is as slippery as ever.

Timely

Make use of LLMs to scour Reddit, sites, and more to do an early pass on user research before committing more of your time to design and code.

Try This

SYSTEM
You are “ResearchCopilot,” a UX strategist.

USER
Here is a raw scrape from r/[Subreddit] discussing problems with [domain] (≈3,000 words).  

1. Extract the 12 most common pain points as bullet phrases (≤10 words each).  

2. Group those pains into up to 4 Jobs-to-Be-Done buckets.  

3. For each bucket, list one verbatim quote (<120 chars) that exemplifies it.  

[[PASTE THREAD TEXT]]

“The goal is not to get to the minimum version that checks some magical box; it’s to deliver a quantum of value to your customer.”

Timeless

An MVP remains a moving target, unlocking value only momentarily until that value has been consumed, internalized, and normalized.

Timely

One-shot prompts and AI IDEs empower founders, product managers, and others to engage with customers more frequently without incurring the cost or risk.

Try This

SYSTEM  
You are “MVP-Smith,” an expert in lean TypeScript + React prototypes.

USER  
Draft a minimal **React + Express** project that proves this value hypothesis:  

“Users will upload a PDF and receive a one-sentence summary.”
  
Constraints:  
• Front end = one drag-and-drop zone + spinner + result box.  
• Back end = single `/summarize` endpoint calling OpenAI API (gpt-4o).  
• Include `package.json`, env variable placeholders, and comments for TODOs.  

Respond with the full folder tree and inline code blocks.

“A better approach to market sizing is to see it not from the outside but from the inside… It’s no coincidence investors like founders with domain expertise—it gives them an unfair advantage.”

Timeless

Your unfair advantage is still worth playing whenever you can. Much like a bottom-up TAM is preferred to a top-down one, insider knowledge lets you leverage not just the problem but the players.

Timely

AI gives everyone their local expert to learn from. The real advantage will come from building up your own fine-tuned data to yield even better results.

Try This

SYSTEM  
You are “FineTune-Maker,” skilled at converting domain docs into Q-A pairs.

USER  
Here is an internal FAQ for our niche industry (≈2,000 words).  
Turn it into **50 JSONL** training lines.  
• `prompt`: a likely question from a newbie operator (≤120 chars)  
• `completion`: the expert answer (≤250 chars, no code fences)  
Output JSONL only—one line per pair—ready for OpenAI fine-tuning CLI.  
  
[[PASTE FAQ]]

“Fit is not something that can be easily measured… we understand fit through approximation and proxies.”

Timeless

PMF is still measured through metrics like revenue, engagement, and more, still proxies of the value and sentiment being derived.

Timely

AI lets us look at data with new eyes, spotting the patterns that were previously difficult or expensive to surface. Further, AI can engage our customers directly via chat and features to capture new sets of data on sentiment, NPS, and more.

Try This

SYSTEM  
You are “RetentionRadar,” a data-savvy product analyst.

USER  
I’ve pasted a CSV with daily active users, sign-up date, plan tier, and churn flag.  
1. Identify the three cohorts with the **highest early churn (≤30 days)** and quantify churn %.  
2. Suggest one hypothesis per cohort for why they leave.  
3. Recommend one experiment each to improve retention.  
  
[[PASTE CSV]]

This post is a part of the “Just in Time” series, where I revisit and adapt my experiences and advice for founders from the past 20 years. It is based on “Product-Market Squint”, originally published on March 30, 2017.