In partnership with
AI Spotlight · The Market Map
The AI Market is Splitting Into Three Camps — Which Side Are You On?
Infrastructure builders, workflow platforms, and prompt packs dressed as products. The tools you use every day fall into one of these three camps. Knowing which is which may be the most useful thing you learn this month.
At some point in the last two years, every person with a laptop and a ChatGPT account became an AI company. That era is ending. A Google VP stated in early 2026 that two once-hot AI business models are now looking like cautionary tales: the LLM wrapper and the AI aggregator. The venture capital money that chased shiny prompt interfaces is drying up. The market is consolidating fast, and it is consolidating around a very clear structure.
The AI market in 2026 is splitting into three distinct camps. Infrastructure builders at the bottom, genuine workflow platforms in the middle, and what developers bluntly call prompt packs pretending to be platforms at the top. Each camp has a fundamentally different relationship to durability, pricing power, and long-term survival. And every tool you use today sits somewhere in this map whether you realise it or not.
This issue gives you a clear framework to identify which camp each of your tools belongs to, what that means for how much you should rely on it, and where the smart bets are for solo operators and small teams building on AI right now.
* * *
Camp One
The Infrastructure Builders: They own the ground everything runs on
At the bottom of the AI stack sit the companies that own the physical and computational layer: the chips, the data centers, the cloud infrastructure, the foundation models, and the raw APIs that everything else is built on. This camp includes Nvidia, Amazon, Google, Microsoft, Meta, OpenAI, and Anthropic. These are not just AI companies. They are the terrain itself.
What makes this camp structurally different is that it wins regardless of which specific application wins. Amazon does not care whether ChatGPT or Claude or Gemini dominates consumer mindshare, because all three run on AWS infrastructure. Nvidia does not care which model architecture prevails, because all of them need GPUs. The infrastructure layer is the toll road, and every AI product that ever gets built pays to use it.
|
How to recognise a Camp One company They sell to other AI companies, not just to end users. Their primary customers are developers and other businesses building on top of them, not individuals using a tool. If they disappear, everything built on top of them disappears too. Dependency flows upward. They are not dependent on anything above them in the stack. Their moat is physical and capital-intensive. You cannot replicate AWS by writing better code. You replicate it by spending hundreds of billions on data centers, which no startup can do. Examples: Nvidia, AWS, Google Cloud, Azure, OpenAI API, Anthropic API, Mistral API, open-source model providers like Meta (Llama). |
As a user, you almost never interact with Camp One companies directly. But every tool you use does. Understanding this camp matters because it tells you who the landlords are in the AI economy, and what happens to your tools when those landlords change their terms.
* * *
Camp Two
The Workflow Platforms: They own the outcome, not just the interface
The second camp is where the genuinely interesting and durable AI businesses live. These are companies that use AI as one component of a deeper product: a product that is integrated tightly into how users actually work, that owns proprietary data or workflows, that would not collapse if the underlying AI model changed, and that solves a problem specific enough that a general-purpose model like GPT or Gemini cannot easily replace it.
The test for a genuine workflow platform is simple: if OpenAI shut down their API access tomorrow, would this product still work in a meaningful form? If yes, it is a platform. If no, it is a wrapper. The best Camp Two companies use AI to amplify a product, not to be the product. They own the workflows, the data, the integrations, the edge cases, and the user habits that make switching away genuinely painful. In 2026, the companies that have figured this out are the ones showing the most durable growth and the strongest enterprise margins.
|
How to recognise a Camp Two company Switching is genuinely painful: Your data, your history, your templates, your integrations, and your team's habits are all baked in. Leaving costs more than staying. They own proprietary data or workflows: The product gets smarter or more useful the longer you use it, because it learns your specific context, terminology, or processes rather than being a generic interface. They solve a problem too specific for a general model to own: A legal contract review platform that knows your firm's clause preferences. A customer support tool trained on your product's unique edge cases. A coding assistant embedded in your specific codebase. Examples: Cursor, Replit, Notion AI, Harvey (legal), Abridge (medical), GitHub Copilot, Databricks, vertical SaaS tools with deep domain AI. |
Camp Two is where solo operators and small teams should be spending the majority of their tool budget. These products are worth paying for, worth integrating deeply, and worth trusting as durable parts of your stack. The key signal: the product was useful before AI was added, and AI made it significantly better, rather than AI being the entire reason the product exists.
AI Agents Are Reading Your Docs. Are You Ready?
Last month, 48% of visitors to documentation sites across Mintlify were AI agents, not humans.
Claude Code, Cursor, and other coding agents are becoming the actual customers reading your docs. And they read everything.
This changes what good documentation means. Humans skim and forgive gaps. Agents methodically check every endpoint, read every guide, and compare you against alternatives with zero fatigue.
Your docs aren't just helping users anymore. They're your product's first interview with the machines deciding whether to recommend you.
That means: clear schema markup so agents can parse your content, real benchmarks instead of marketing fluff, open endpoints agents can actually test, and honest comparisons that emphasize strengths without hype.
Mintlify powers documentation for over 20,000 companies, reaching 100M+ people every year. We just raised a $45M Series B led by @a16z and @SalesforceVC to build the knowledge layer for the agent era.
Camp Three
The Prompt Packs: A nice interface sitting on top of someone else's product
The third camp is the most populated and the most precarious. These are products whose entire value proposition is a well-designed interface wrapped around an API call to GPT, Claude, or Gemini. The underlying AI is someone else's. The data stays nowhere. The workflow integration is shallow. If OpenAI released the same feature natively, or if the model provider raised its API prices, the product would either vanish or become unaffordable overnight. Developers call these LLM wrappers. A more honest name is prompt packs pretending to be platforms.
This does not mean every Camp Three product is useless right now. Many of them are genuinely pleasant to use and solve real problems in the short term. The issue is durability. The AI wrapper boom of 2023 and 2024 produced hundreds of these products. As of 2026, the graveyard of shut-down AI writing tools, AI summarisers, AI image prompting apps, and AI email assistants is enormous. The survivors are almost exclusively the ones that either moved into Camp Two by building genuine workflow depth, or found a very narrow niche that the big models have not bothered to address yet.
|
How to recognise a Camp Three product The "if they disappeared" test: If OpenAI, Google, or Anthropic shut off the API key this product uses, would the company survive? If the honest answer is no, it is Camp Three. The "built in a weekend" test: Could a developer replicate the core functionality in a weekend with a direct API call and a basic frontend? If yes, the moat is essentially zero. The "native features" test: Has the underlying model provider already built or announced a version of this feature natively? If ChatGPT, Claude, or Gemini now do what this product does, the product is on borrowed time. Warning signs in the wild: AI writing assistants with no document storage, AI summarisers with no integration into where you actually read things, AI image tools with no workflow beyond generation, AI email tools that just draft replies without learning your voice or managing your actual inbox. |
The practical advice is not to avoid Camp Three products entirely. Some of them are useful enough right now to be worth using on a month-to-month basis. But you should never build a critical workflow around one, never pay annually in advance, and never assume it will exist in its current form in two years.
* * *
The practical part
How to audit your own AI stack right now
Pull up the list of AI tools you currently pay for or use regularly. For each one, run it through three questions in order. The answers will tell you exactly which camp it belongs to and how much structural trust to place in it.
|
The three-question camp test Question 1: If the underlying AI model disappeared, would this product still exist? If yes, it is almost certainly Camp Two or Camp One. The product has something real beyond the model: data, workflow, integrations, or physical infrastructure. If no, it is Camp Three. The model is the product. Question 2: Does this product get more useful the longer I use it? Camp Two products have compounding value. They learn your context, accumulate your data, integrate into more of your tools over time. Camp Three products deliver the same output on day one as they do on day three hundred. There is no learning, no compounding, no depth. Question 3: Would switching to a competitor be annoying or genuinely painful? Annoying means Camp Three: you lose a nice interface but nothing important. Genuinely painful means Camp Two: you lose your data, your history, your integrations, your team's workflows. That pain is the product. It is called a moat. |
* * *
Where to place your bets
Which camp is actually worth betting on as a user and builder
| Camp One | Camp Two | Camp Three | |
|---|---|---|---|
| Durability | Extremely high | High if moat is real | Low to very low |
| User control | You rarely interact directly | Deep, daily interaction | Surface-level interaction |
| Build on it? | Yes, but accept dependency | Yes, invest deeply | Use lightly, never depend |
| Annual commitment? | Depends on use case | Yes, worth locking in | No, stay month-to-month |
| Risk when market shifts | They gain from the shift | Protected by user lock-in | First to be disrupted |
The clearest advice is to invert the typical approach. Most people spend the most time evaluating the shiniest, newest Camp Three tools and the least time thinking about whether the Camp Two tools they rely on are as deeply embedded in their workflow as they could be. That priority order is backwards. The tools worth obsessing over are the ones that compound. The tools worth staying curious about but cautious with are the ones that dazzle.
* * *
The question the market is answering right now
The consolidation happening in 2026 is not a crash. It is a clarification. The market is running a brutal, efficient experiment to determine which AI products create genuine value and which ones were riding the wave of novelty. The three-camp structure is the output of that experiment so far.
Infrastructure builders will survive because they own the ground. Workflow platforms will survive because they own the habits. Prompt packs will largely disappear, or transform, or get acquired for their user bases, and be replaced by native features in the tools that own the actual workflow layers beneath them.
The most useful thing you can do with this framework is not to use it as a reason to distrust AI tools. It is to use it as a lens for knowing where to invest your attention, your data, and your deepest workflows. The tools in Camp Two that are right for your specific work are the ones worth building around. Everything else is a feature, not a foundation.
|
The AI tools worth betting on are not the ones with the best demos. They are the ones that make it genuinely painful to leave. That pain is not a flaw. It is the product. |
* * *
|
Before you go Which AI tool in your stack do you think is secretly Camp Three disguised as a real platform? Hit reply with the name and why. The most honest answers will shape a follow-up issue where we put specific popular tools through the three-question test publicly. |
Until next time,
AI Spotlight
Practical AI, translated into real work, once a week.