Building something now is easy. It’s getting anyone to use the thing that’s so hard.

But there’s a rare, generational, distribution opportunity that can change that (for a limited time). A beachhead if you will to finding customers. The most analogous event in our lifetime starts with a quick story from the summer of 2008, with a guy named Steve Demeter who was writing software for ATMs.

Nights and weekends, he built Trism, a little puzzle game. When Apple opened the App Store that July, he put it on the shelf for $5. Two months later he'd pocketed $250k and quit his ATM job. Then Ethan Nicholas did it with a tiny artillery game. Built it in six weeks; made $800k in five months.

Steve and Ethan are the modern vibe-coders dream. Make something that people find and use really fast + pocket some cash.

And what made Steve and Ethan’s stories possible is obvious. The App Store was brand spanking new; the most fertile ground ever for acquisition. Millions of people standing in front of a nearly empty shelf, downloading literally whatever crap was on it. The shelf had just 500 apps on it, and those apps got 10M downloads in 3 days.

That ratio (enormous demand, almost no supply) was a goldilocks period for builders, and lasted maybe two years. By the time everyone's cousin and fart app was building an app, the shelf was full. The gold rush just became a red rush of competition.

Two weeks ago, we looked at the demand map we got handed from Anthropic's free 1M free user interviews, where the study told us exactly what and how people ask their AI for help on. Wild data across health, careers, relationships and money. The play we landed on was to verticalize a sub-topic for the consumer, because Anthropic was leaving that layer to us and only showing competitive interest at the business/enterprise layer.

At the end, I flagged the one thing the map doesn't hand you. Distribution.

About that..

The place where all that demand shows up, inside the Claude/ChatGPT window, is now a dynamic app shelf you can put your product on without people opening your app.

And for context. OpenAI’s app store has only 1,600 apps, sitting in front of 800 million weekly users.

Apple never released the numbers for how many people were on the App Store in 2008, but we know they sold 13.7M phones that year, which is the max it could have been. That means the goldilocks ratio today is 1:500,000, vs the 2008 1:28,000.

Are your ears ringing yet?

I hear those opportunity bells. That per-user math against 2008 means today's empty shelf is more than 18X emptier than the one Steve and Ethan got rich on.

The demand map told you what to build. The chat window is now where you can go ship it to access that demand. And the people are waiting.

Platforms giveth, and platforms taketh away. The 2008 devs who built real businesses weren't the ones who won a lottery placement. They were the ones who turned downloads into an audience they owned.

So whatever you build from today's idea (or are just building today), my die-on-a-hill advice is run a newsletter next to it. When the algorithm changes (it will), your list is the one channel no platform can reroute. beehiiv is where I'd build it: free up to 2,500 subs, with everything you need to grow and engage after that.

Want the full creator kit I have?

🥷 Take the sub-topic you picked off the demand map and ship it as an app inside ChatGPT (and Claude), so when someone describes your exact problem mid-conversation, the AI hands them your product at the moment of peak intent.

If you haven’t seen it yet from 2 weeks ago, open this tab for later with Anthropic’s study, How people ask Claude for personal guidance.

Think about what actually happens in those conversations. Someone is telling their AI, in their own words, that they can't make sense of their macros, or their job offer, or their lease. That is the most opportune moment where a pain/problem is real for someone.

In our unbundling AI idea, the move was making the consumer product for that person because Claude wouldn’t.

Today's move is where you put it. Because since last October, apps can live inside ChatGPT. Not linked from it. But in it, rendering their own UI right in the reply.

And this really matters because people’s chat app is becoming a fundamental platform surface. We used to need standalone apps on our phone because the phone was the main surface, but as AI does more and more, the phone is just a way to get into our AI platform. So, you need to be in that AI. Just like how everyone rushed to build an app back in the day, now it’s the same thing for AI apps.

Plus, there’s a distribution mechanic that’s very useful → ChatGPT will suggest apps contextually. The user doesn't need to search for you. As they discuss their problem, and the model can decide your app fits the moment. It's intent-matched distribution, the thing every ad platform on earth is a rough proxy for.

One quick example and it’s perfect. Cal AI was a hero of the demand-map post. Mediocre-but-smart calorie app that rode one sub-topic to $50M+ ARR and got acquired by MyFitnessPal.

MyFitnessPal is now in the ChatGPT app directory. Check this out.

The company that bought a demand-map winner looked at where that demand comes from (people literally asking about their macros or meals or fitness) and put themself there.

This new shelf is not empty because the opportunity is bad. It's empty for a few reasons we'll break down now, and they're sort of the same reason most devs sat out 2008 and watched.

Let's break it down so you get to be like Steve and Ethan.

Today's move is that products win by living inside conversations instead of next to them. Granola figured this out early. They’re my AI notepad I use multiple times a day without me remembering it’s there half the time. Granola just sits inside the conversations I’m already having.

It works quietly off device audio, and by the time you’re done, the notes are done and filed. Great example of a product that went to where the conversation was, instead of asking the conversation to come to it.

Notes, actions and memory. Without a meeting bot.

All distribution channels in history have separated the moment of need from the moment of discovery. The problem is prime on Tuesday, you see the ad on Wednesday. SEO closed that gap a little, but this closes it completely—the conversation where the need is primed is the surface where the product is primed.

There are exactly three ways people can find apps in ChatGPT:

→ Named call of the app by the user. Like a slash command almost, the user just starts a message with the app's name: "Spotify, make a playlist for this weekend's party." Means people need to know your name.

→ A contextual suggestion by the AI. ChatGPT proposes an app unprompted when the convo matches the use case. OpenAI has said the ranking signals are conversational context, usage patterns, user preferences and quality. Apps users return to earn more prominent placement over time. This is like early App Store top charts around a relevance model.

→ The directory a user goes to. Browsable at chatgpt.com/apps, like any other App Store where you find and add.

And apps show up in three formats and it’s good to know as we approach the “ok how do I make this” part

Inline cards that render in the reply (the default for listings, results, etc)

Fullscreen takeovers for maps and dashboards, like when I ask about places for sale in Greenpoint and Zillow tells me no…😭

Picture-in-picture, which is a small floating window that keeps running while the user chats. Not my favorite UX tbh.

And the SEO parallel goes deeper than discovery. The more users associate "apartment search in ChatGPT" with Zillow, the harder that habit becomes to break. Early rank compounded in Google for a decade. Early habit will compound here the same way.

Now imagine the same mechanic pointed at "help me understand this job offer" or "what should I do about my knee." Every sub-topic on the demand map is a conversation like these. If you didn’t see the demand map, here’s the full Notion I put together with all the data

The App Store launched July 10th ‘08 with ~500 apps and did 10 million downloads in 72 hours. 100 million by September. 1 billion by April 2009. The entry fee was $99 and still is for developers, which put solo bedroom devs like Steve on equal footing with the companies they worked for.

The ChatGPT directory, per the mid-2026 census, has 1,624 apps across 44 categories against 800M+ WAU. The SDK is free and built on MCP, and a working prototype is something you can do with your Claude in a day or two.

And the composition of the current app store tells you where the white space is. Mostly productivity and business tools like Canva, Zoom, Adobe and the enterprise crowd. The consumer guidance verticals from the demand map are nearly unoccupied.

I’d have thought way more apps would be here as we approach a 1 year old app store. But empty shelves always have a reason. And here are some I can think of, and why I think they're the same vibe as 2008. AKA the kind that look smart for eighteen months and terrible for the following fifteen years.

The Apps SDK is built on MCP, which is a joint standard for interactive UIs on the protocol. Meaning OpenAI and Anthropic are co-creating the shelf standard, and the server you build for ChatGPT is most of the work for a Claude listing too.

And Claude's side is growing too. The connectors directory has grown to around 900 connectors, Anthropic added consumer connectors (Uber, Spotify, Instacart, TripAdvisor, Resy, AllTrails) and shipped in-app submission tools and a developer dashboard in June.

You get two brand new distribution shelves with very little competition but a ton of growing demand. All on the same server. And mostly empty for consumer guidance ideas.

Instead of examples of who’s doing this (because we’ve already spoken about a few and you can just go to ChatGPT app store to poke around), to make this concrete I went back through the proven demand map and matched a stealable micro-app idea to what people are actually asking in each vertical.

Here are 3 quick ideas you could build to practice making an MCP app and getting a feel for it. All ideas are linked below.

What people are asking: offer evaluation and salary negotiation, the sharpest wedge on the whole map (career is 26% of guidance conversations, and most people negotiate badly or never).

The micro app: paste your offer letter, and an inline card breaks base/bonus/equity against market bands and drafts your counter script to negotiate.

Why this format: an offer is one atomic object which is great for inline apps. You can design one set of cards, one verdict, and one reply script to help people negotiate.

What people are asking: retirement math, how long until they reach their finance goals, real should-I and what-if decisions with real stakes (80% of chats are ranked high-stakes).

The micro app: an interactive projection widget with calculator toggles that does the actual compound math. This one has a clever edge too: LLMs guess at arithmetic, but widgets can calculate it. Your app can be the thing that makes ChatGPT's money advice actually add up.

I made this on web about a year ago when I was learning v0, it’s called The Millionaire Calculator. Check it out, it’s a stealable UI you can build into your app.

Why this format: sliders and projections want an interactive card, not a paragraph.

What people are asking: leases, eviction notices, deadlines. And 94% of legal guidance conversations are high-stakes, the highest in the study, and 2025's $3.5B of legal AI funding went almost entirely to law firms vs the consumer.

The micro app: paste the clause or notice, an inline card flags your deadlines and state-specific rights, and fullscreen mode drafts the response letter.

Why this format: triage inline, then take over the screen when it's time to produce the document.

I mapped one mini idea for all nine verticals—what people are asking (from the real usage data), a fun micro app that answers it, and the widget format it wants. It's free in Notion, next to the demand map:

TLDR: Intent used to be something you had to buy. This idea is about where you go stand. Pick your sub-topic, or think about your own business and how it fits into this channel, wrap it in an MCP server, and let the model hand your product to people mid-sentence. Here's the actual build path to “doing it”

And because I know some of you will want the exact clicks: I've packaged this whole build path including the setup prompts, the Chippy route, Dev Mode, the golden-prompt worksheet and the submission checklist into a free step-by-step guide. I’ve clearly been enjoying Notion recently.

Remember, the 2008 window lasted about two years. Who knows how long this one goes for. But I know which side of it I'd rather be standing on when it closes.

Go claim your shelf space.

p.s. If you ship something into either directory off the back of this, reply and tell me. I will 100% feature reader apps in a future edition.

› The primary—OpenAI's announcement of apps in ChatGPT and the Apps SDK, with the launch partners and demo videos. → Introducing apps in ChatGPT (8 min)

› The marking rubric—OpenAI's own guide to what gets approved and suggested. → What makes a great ChatGPT app (10 min)

› The census—the mid-year state of the directory: 1,624 apps, category breakdowns, verification stats. Your market-scan starting point. → State of ChatGPT Apps 2026 (12 min)

› The demand map—Our first breakdown of the primary research from Anthropic on what people are using Claude for. → Steal this data from Anthropic’s 1 Million user interviews (8 min)

› The bull case—Colin Matthews on why this is the next big distribution channel → ChatGPT apps are about to be the next big distribution channel (15 min)

🛠️ A tool I loved—Lazyweb.com by my friend Ali Abouelatta, that “puts product growth on autopilot”. It’s free and takes 300k+ learnings from the worlds best products (experiments, flows, etc) into ideas for you on activation, growth, etc.

📺 A video I loved—"The Golden Age of Apps", a nostalgic YT documentary on the 2008–2013 App Store era: fewer rules, weirder apps

📬 A read I enjoyed—"Striking It Rich in the App Store: more casino than gold mine", written right as the 2008 window was closing.