problem hacker #23
The Distribution Game
The final Problem Hacker of 2025
Twenty-five days. Four frontier AI models.
November 17: xAI launched Grok 4.1 November 18: Google launched Gemini 3 November 24: Anthropic launched Claude Opus 4.5 December 11: OpenAI launched GPT-5.2
Each one immediately challenged by the next. Benchmarks leapfrogging weekly. A compression of innovation so intense that tracking it became nearly impossible.
Everyone focused on the model race. Who’s winning on coding benchmarks? Which one reasons better? Whose context window is bigger?
Then on December 17, OpenAI stopped racing.
They launched an app store.
The Year the Models Converged
2025 was supposed to be about model supremacy. Who could build the best AI?
For most of the year, that’s exactly what it was. Claude Sonnet 4.5 in September claimed “the best coding model in the world.” OpenAI countered with GPT-5.1 in November. Google dropped Gemini 3 six days later and briefly took the lead.
OpenAI called an internal “code red.” Rushed out GPT-5.2 less than a month after 5.1. Got back to parity on most benchmarks.
This is how everyone thought the game worked: Build better models. Win more users. Collect more data. Train better models. Repeat.
But here’s what actually happened in those 25 days:
The models converged.
Not perfectly. Claude Opus 4.5 is still better at some coding tasks. Gemini 3 has that 1 million token context window. GPT-5.2 leads on certain reasoning benchmarks.
But the gap between them is smaller than it’s ever been. And more importantly: the gap is smaller than most users can perceive.
When four models can all write production code, analyse complex documents, and reason through graduate-level problems, “better” stops mattering as much as it used to.
OpenAI saw this coming. And made a different bet.
The App Store Play
On December 17, ChatGPT launched an app directory. Not an API marketplace. Not a plugin system. An actual app store inside ChatGPT.
Apps from Spotify. Booking.com. DoorDash. Dropbox. Google Drive. Apple Music.
Developers can now submit apps for review. Build integrations directly into ChatGPT conversations. Get discovered by 800 million weekly users.
The strategic shift is profound:
OpenAI stopped competing on “best model” and started competing on “best distribution.”
This is Apple’s playbook. The iPhone wasn’t the best phone. It became the surface through which third parties reached users. The underlying technology became invisible infrastructure.
OpenAI is betting conversational AI can do the same thing.
If they succeed, asking ChatGPT to book a flight or order dinner becomes as natural as tapping an app icon. And the underlying service – Booking.com, DoorDash, whoever – becomes invisible.
Why This Changes Everything
Because distribution leverage is different from capability leverage.
Capability leverage: “Our model is 3% better at coding benchmarks.” Distribution leverage: “We have 800 million users and you need us to reach them.”
Capability leverage is temporary. Every model release gets matched within weeks.
Distribution leverage compounds. Every app that integrates makes ChatGPT more useful. Every integration creates switching costs. Users get invested in connected services. Workflows get built around specific integrations.
This is why platform businesses win. Not because their core technology is better, but because they control the surface.
Google didn’t win search by having the best algorithm. They won by becoming the default place people go to find things. Then they monetised that surface through ads.
Facebook didn’t win social by having the best features. They won by getting everyone’s friends on the platform. Then they became the surface through which you reached those friends.
Amazon didn’t win ecommerce by having the best logistics. They won by becoming the default place people go to buy things. Then they became the surface through which sellers reached buyers.
The pattern repeats: Control the surface. Make yourself indispensable. Let third parties do the work.
What This Means for the Other Players
Anthropic and Google face a different problem now.
Claude Opus 4.5 might be the best coding model. Gemini 3’s million-token context window is genuinely impressive.
But if ChatGPT’s app ecosystem creates switching costs – users invested in connected services, workflows built around integrations – competing on model quality alone might not be enough.
This is the uncomfortable question every challenger faces: Can you build a better product when the incumbent controls distribution?
Anthropic’s bet: Own enterprise coding. They positioned Claude as the default in Cursor, focused their entire first conference on developers, built computer use capabilities.
Google’s bet: Multimodal everything. Gemini 3 processes text, images, video, code. Veo 3 for video generation. Integration across Google’s entire ecosystem.
Both are solid strategies. Both might work.
But both require battling against distribution leverage.
Because here’s what OpenAI’s app store creates:
- Users discover apps contextually (ChatGPT suggests them during conversation)
- Apps get triggered automatically based on intent
- Transactions happen without leaving the interface
- Every successful interaction trains the recommendation system
- Developers invest in the platform with the most users
- More apps make the platform more useful
- More useful attracts more users
That’s a flywheel. And flywheels are hard to compete with.
The Problem Nobody’s Talking About
Everyone’s focused on whether the app store will work.
The real question is what it means for strategy in every other industry.
Because what just happened in AI is happening everywhere:
Capability advantages are temporary. Distribution advantages compound.
Your product is 10% better? Great. That advantage lasts until your competitor’s next release. Usually weeks or months.
You control distribution? That advantage compounds. Every user you add makes you more valuable to suppliers. Every supplier you add makes you more valuable to users.
This is why:
- Amazon beat better-funded retailers (they controlled discovery)
- Spotify beat labels (they controlled the listening interface)
- Uber beat taxis (they controlled rider demand)
- Airbnb beat hotels (they controlled guest supply)
None of them had the best “product” in the traditional sense. They had the best distribution.
The mistake most businesses make: Optimising for capability when the game is about distribution.
They invest in features. Improve quality. Reduce costs. Ship faster.
All good things. All temporary advantages.
Meanwhile someone else is building distribution leverage, becoming the surface. Controlling the customer relationship; making themselves indispensable.
What This Looks Like in Practice
A SaaS company spends millions improving their product. Better UI. More features. Faster performance.
Their competitor builds a Slack integration. Then a Teams integration. Then APIs that let customers embed their service everywhere.
Guess who wins? The one that became infrastructure.
A retailer invests in better inventory, faster shipping, nicer stores.
Their competitor builds a marketplace. Lets third-party sellers use their platform. Takes a cut of every transaction.
Guess who wins? The one that became the surface.
A consultancy hires better people, develops better methodologies, builds better case studies.
Their competitor builds a platform. Creates a community. Becomes where clients go to find solutions.
Guess who wins? The one that controls discovery.
The pattern is everywhere. Capability advantages are temporary. Distribution advantages compound.
The Uncomfortable Truth
Most businesses are playing the wrong game.
They’re optimising their product when they should be thinking about their position.
Position questions sound different:
- Not “how do we build a better product?” but “how do we become indispensable?”
- Not “how do we improve features?” but “how do we control distribution?”
- Not “how do we compete on quality?” but “how do we make switching costly?”
- Not “what capabilities do we need?” but “what surface do we control?”
These aren’t better questions because they’re more strategic. They’re better questions because they acknowledge the actual game being played.
2025 in AI was supposed to be about model supremacy. It ended up being about distribution.
The models converged. The distribution advantage compounded. And OpenAI made the most important move of the year by launching something that had nothing to do with model capability.
The Hack
The Distribution Audit
Answer these three questions about your business:
1. Where are you optimising for capability? Better features. Faster service. Lower prices. Higher quality. These are all capability advantages. They’re good. They’re also temporary.
2. Where could you optimise for distribution instead? Could you become the surface customers go through? Could you control discovery? Could you make third parties need you to reach customers? Could you create switching costs through integrations?
3. What would it take to own the interface? Not the product. The interface. The place customers go first. The layer between customers and solutions. The surface that makes other services invisible.
Most businesses will realise they’re playing the capability game when the real game is distribution.
They’re competing on having the best product when they should be competing on controlling access to customers.
OpenAI just proved this in the most competitive market in the world.
They were losing the model race. So they changed which race they were running.
They stopped trying to build the best AI. They started trying to become the surface through which people access AI.
That’s not a model strategy. That’s a platform strategy.
And platform strategies win because distribution compounds in ways that capability never can.
A Note on 2025
This is the last Problem Hacker of 2025. Thank you for reading this year.
What started as identifying problems businesses don’t realise they have has become something bigger: watching the patterns that keep repeating across industries.
The problems are different. The pattern is the same.
Most businesses optimise for the wrong thing. They perfect their answers while the question changes. They improve execution while strategy becomes irrelevant.
2026 will bring new problems. Different industries. Fresh challenges.
But I suspect we’ll keep seeing the same pattern: Businesses solving problems they don’t have while ignoring problems they don’t see.
That’s what makes this work interesting.
See you in January.
The Problem Hacker identifies problems businesses don’t realise they have. Published by Mark Jefford of Jefford Consultancy. If you’re winning the wrong race, we should talk.