Quick answer
For two years VCs said "AI wrappers won't survive when GPT or Claude improves." That thesis is now visibly wrong. Cursor, Perplexity, Granola, Lindy, Captions, Beehiiv — every notable wrapper has gotten faster, not slower, since the underlying models improved. The moats aren't the model. They're distribution, workflow, brand, and data.
The 2024 VC consensus was clear: thin AI products are doomed once the foundation models add the same feature for free. By mid-2026 that consensus looks naive. Cursor passed $300M ARR. Perplexity is on track for $400M. Granola, Captions, Lindy all crossed $50M. Far from being crushed by GPT-5 and Opus 4.8, they grew alongside them.
Why the "wrappers die" thesis was wrong
- Foundation models added some wrapper features (file uploads, web search) but never integrated them tightly enough to replace the wrapper experience.
- Wrappers built distribution. Perplexity has 150M monthly users — those are users ChatGPT doesn't get back.
- Workflow lock-in is real. Cursor users have custom rules, snippets, and codebase indexes — switching cost is high.
- Brand matters more than expected. "Use Perplexity for research" is now a habit; people don't reopen the question every day.
The real moats AI wrappers have built
- Distribution: 100M+ users compound. Foundation labs can't out-distribute incumbent wrappers.
- Workflow integration: a Cursor user has Cursor in their muscle memory. A new feature in ChatGPT doesn't change that.
- Data network effects: the more meetings Granola records, the better its templates get. Foundation labs don't see that data.
- Specialisation: vertical AI tools (legal, medical, devops) have domain depth foundation models don't.
- Vendor diversity demand: enterprises want to not be locked to one foundation model. Wrappers provide that abstraction.
Which wrappers still die?
The thesis isn't totally wrong — it's just narrower than VCs claimed. Wrappers without distribution, without workflow lock-in, without specialisation, do still get crushed. "AI ChatGPT for X" with no other moat dies. But wrappers that built any of the four moats above survive and grow.
If you're building an AI product in 2026, ask yourself: which of distribution, workflow, data, or specialisation am I building? If the honest answer is "none," your product is the thin wrapper everyone warned about.
Related reading
Bottom line
AI wrappers won. The doomsayer thesis was wrong because it underrated distribution and workflow. If you're shipping AI, stop worrying about being a "wrapper" and start asking what moat you're building.
