“He’s just an AI rapper.”
This put-down sounds familiar to people who develop something new using artificial intelligence.
Push-back sounds equally familiar.
“Everything is a wrapper. OpenAI is a wrapper around Nvidia and Azure. Netflix is a wrapper around AWS. Salesforce is an Oracle database wrapper that’s worth $320 billion.” Arvind Srinivas, CEO of Perplexity says,
For those who are not familiar with the term “AI wrapper”, here is a good definition,
this is one dismissive word It refers to a lightweight application or service that uses an existing AI model or API to provide specific functionality, usually with minimal effort or complexity involved in its creation. A popular example of an AI wrapper are apps that enable users to “chat” with a PDF. This type of AI application allows users to upload a PDF document, such as a research paper, and interact with the AI model to get instant analysis and answers about specific content. In the early days of ChatGPT, it was not possible to upload documents as part of the prompt or create custom GPTs, so these apps became popular very quickly.
In my view, this AI wrapper debate misses a big point. Not all rappers are the same. Some people enjoy the race for a short time and only last until the big platforms collect them in their suite. But products that can live where the work is done, create proprietary data and learn from use, and/or may encounter existing distribution advantages. The wrapper label distracts from what I think really matters: (1) whether it’s a feature or a product, and (2) how big the market segment is.
Let’s first look at an old example of a wrapper that lets you chat with a PDF. Such a tool solves a narrow problem of answering questions about a document. It does not create new documents or edit existing documents. It typically does not capture any unique data, or learn from user behavior. So to me, it’s a capability rather than an end-to-end solution. A means to an end if possible. As a result, this type of feature resides inside a document viewer or editor, or in the core applications of model providers. So when the foundation models themselves (OpenAI/ChatGPT, Anthropic/Cloud, Google/Gemini) bundle this feature natively, the standalone tool becomes unnecessary. This is classic feature behavior – easy to copy, no end-to-end work, no moats or long-term defensiveness.
One caveat though; Even those that do have features can be interesting indie businesses that make money unless platforms build it into their apps.,
pdf.ai $500K MRR, PhotoAI $77K MRR, chatbase $70K MRR, interiorAI $53K MRR,
jenny ai Went from $2,000 to over $333,000 MRR in just 18 months,
Some wrappers are genuine products but occupy such a large segment of the market that model manufacturers and large technology platforms cannot ignore them. Two factors of competition come into play: (1) model penetration, and (2) distribution.
Coding assistants describe both. Tools like Cursor have turned a wrapper into an AI integrated development environment (ide) that reads repos, edits files, generates code, reverts changes, runs coding agents, and re-imagines the developer experience for the AI-era. The market justifies the attention. Software developers represent about 30% of the workforce at the world’s five largest market cap companies, all of which are technology firms, by October 2025.Development tools that increase productivity even by modest percentages unlock billions in value, This makes this segment a prime target for both model builders and incumbents who already have distribution channels in place,
But cursors and other such tools are almost entirely dependent on access to Anthropic, OpenAI and Gemini models open source Open-weigh and in-house models match or surpass Frontier models in quality. developer forum Paying customers are full of complaints about rate caps. In my own projects, I exhausted my cloud credits at Cursor mid-project and, despite preferring Cursor’s user interface and design, I moved to cloud code (and had to pay ten times more to avoid the rate cap). The interface could be better, but model access proved decisive.
Model access dependency has strategic implications beyond rate limits. OpenAI CEO Sam Altman argues that the right strategy assumes continuous model improvement,
“There are two strategies for building on AI right now. There’s a strategy that assumes models won’t get better. There’s another strategy that assumes models will keep getting better at the same pace. I feel like 95% of the world should bet on the latter category but a lot of startups are built in the former category. When we just do our fundamental work, because we have a mission, we’re going to steamroll you.”
Foundation Model Competition Spans Every Category of OpenAI Applications CEO Fidzi Simo Marked as strategic (knowledge/learning, health, creative expression and shopping) as well as other large market segments such as writing assistants, legal assistants, etc.
Distribution is another threat. Even where model makers hold out, startups face a different competing question – can they build a user base faster than by adding existing AI features to existing products and delivery? It’s the Classic Microsoft Teams vs. Slack DynamicThe challenge is to establish a loyal customer base before Microsoft embeds Copilot into Excel/PowerPoint, or Google adds Gemini into Workspace, or Adobe integrates AI into its Creative Suite, A standalone AI wrapper for a spreadsheet or presentation would have to address not only feature similarity but also bundling/distribution benefits and switching costs,
This distributional competition among incumbents also exists in other large markets such as healthcare and law. In these markets, regulatory friction and control systems of record favor established players like epic system In health care. For example, a clinical note generator that can’t write to an electronic health record (EHR) is likely to run into Epic’s distribution benefits sooner rather than later.
There are three caveats here: (1) First, a market rally can create exit options even without long-term defensiveness; Tools like cursors may lack control over its core dependencies (model access), but rapid growth makes them attractive targets for model builders wanting an instant market presence. (2) second, superior execution sometimes outweighs structural advantages; MidJourney’s product quality reassures Meta use it Despite Meta’s much larger budget and distribution power. (3) Third, foundation models may avoid some markets despite their size; Regulatory burdens in the health care and legal sector, or reputational damage from AI companions or pornographic adult content may provide opportunities for operators willing to face excessive regulatory scrutiny or controversy.
the opportunity remains bigBut competition (and/or acquisitions) could be knocking.
cursor Recurring revenue went from zero to $100 million in 18 months, and became the subject of repeated OpenAI acquisition rumors.
windsurfAnother coding assistant received $2.4B
acquisitionLicensing deal with Google.gamma Revenue reached $50 million in about a year.
lovable Generated $50 million in revenue in just six months.
Galileo AI Acquired by Google for an undisclosed amount.
Not every market gap attracts model builders or big tech. There exists a long line of jobs that are too small for enterprise scale but large enough to support multimillion-dollar businesses. These spaces are suitable for frugal founders with a disciplined scope and lean operation.
Consider those astrology or manifestation or dream interpreter AI apps. A dream interpreter that lets users record dreams every morning, generate AI videos based on them, maintain some kind of dream journal, and solve an entire task by uncovering patterns over time. Yes, users can describe dreams in ChatGPT and it also stores history/memory, but a dedicated app can structure dream capture with specific fields (recurring people, places, things, themes, etc.) and integrate with sleep tracking data in ways that a normal chatbot probably can’t. Such a spot is small enough to escape the model’s attention but seems quite big To maintain a profitable indie business.
While the previous categories create opportunities for new ventures, existing ones face their own strategic choices in the wrapper debate when it comes to model builders. In my view, people participating in model builder competitions will share two characteristics.
-
First, they will own the results, even if they do not own the model. Applications already embedded in user workflows (Gmail/Calendar, Sheets, EHR/EMR, Figma) do not require any new habit formation, and building these platforms from scratch is much more difficult than adding AI capability to an existing platform. When these applications send the action directly to the owner system of record (controlling calendar events, filing claims, creating purchase orders, and so on), “getting done” happens in the incumbent’s environment. AI becomes another input rather than a replacement for existing workflows.
-
Second, successful incumbents will generate proprietary data from customer usage. Improvements, edge cases, approvals, and any human feedback become training data that refines the product over time and a frontier model would not have access to. Cursor, although not an incumbent and despite its reliance on external models, plans to compete by capturing developer behavior patterns, CEO Michael Truell notes in his note. stratcherry interview,
Ben: Is that a real sustainable advantage for you going forward, where you can really dominate the space because you have usage data, it’s not just the call to LLM that got you started, but now you’re training your own models based on people using cursors. You started with the full context of the code, which is the first thing to get it done, but now you have your own data to train on.
Michael: Yeah, I think that’s a big advantage, and I think this dynamic of high ceiling, you can choose between products and then this kind of third dynamic of distribution gets your data, which helps you improve the product. I think all three of those things were shared by search in the late ’90s and early 2000s, and so in many ways I think that really, the competitive dynamics of our market reflect search more than typical enterprise software markets.
Both critics and defenders of AI rappers have a point, and both are missing something. Critics are right that some rappers lack defensiveness and will disappear when platforms absorb their characteristics. The defenders are right that every successful software company wraps something up.
But I think the insight lies between these positions. Even if a new application starts out as a wrapper, it can endure if it lives where the work is done, writes to proprietary systems of record, creates proprietary data and learns from use, and/or captures distribution before incumbents bundle the feature. More importantly, even as competition arrives, wrappers continue to rapidly ship features that meet users’ needs, making it difficult to compete with them. These are the same characteristics that distinguish permanent products from transitory features.
If you enjoyed this post, please consider sharing it on your social media or with someone who might find it interesting. to follow me on x.com Or Linkedin To discuss technological and business trends as they occur.