THE COMPANY-BUILDING FIELD NOTEBOOKRESEARCH EDITION / SEPTEMBER 2026
Startup
Research.
Search
POST-MORTEM LIBRARY / AI-era failures (2023–2026)

The GPT-wrapper cohort (a category entry, not a single company)

  • Company — Collectively: the document-chat, PDF-summarisation, AI-writing and single-prompt-tool companies launched 2022–23
  • Sector — AI applications
  • Founded / Died — Mostly 2022–2023 / mostly 2023–2025
  • Lifespan — Typically 6–24 months
  • Capital raised — Mostly bootstrapped, pre-seed or seed. SimpleClosure's 2025 shutdown data puts the median capital raised by shutting-down AI companies at ~$2.4M, below the ~$2.8M overall median.
  • Peak scale — Typically thousands to low tens of thousands of users; MRR from hundreds to low tens of thousands of dollars.
  • What it built — Interfaces over a third-party model for one task: chat with a PDF, summarise a document, generate marketing copy, write SQL.
  • Stated cause of death — Where founders wrote anything, the dominant stated cause is "OpenAI shipped it." The reference event is OpenAI DevDay, 6 November 2023, when ChatGPT gained native file upload and document analysis; a widely circulated comment at the time was "many startups just died today, because OpenAI added PDF chat."
  • Evidenced causePartial divergence; the honest version is less flattering. The feature release was real and destroyed some businesses overnight. But these products had no proprietary data, no workflow lock-in, no distribution advantage and no switching cost — three or four weeks of engineering any competitor, including the model provider, could replicate. Most were not killed by DevDay; they were revealed by it. Many died more slowly of inference-cost economics: flat-rate pricing against variable per-token costs produces negative gross margin on the heaviest users, who are also the least likely to churn.
  • Warning signs visible earlier — Retention curves that flattened near zero after month one. A feature list identical to a dozen competitors. A cost of goods sold that rose with usage.
  • Money outcome — Small. Mostly founders' own time and small angel cheques. Some founders did extremely well on the way (several reached $10k–$100k MRR before the collapse) and exited with cash.
  • Post-mortem qualityVariable and thin. A handful of good Indie Hackers and X threads; mostly silence. AI represented 15.9% of all documented startup shutdowns in 2025, per SimpleClosure, with wrappers and apps dominating the composition — so the volume is real even where the documentation is not.
  • Transferable lesson — The genuinely transferable version is not "don't build wrappers" — several wrappers became large businesses. It is: ask what you will own after the model improves by an order of magnitude. If the answer is "the UI," you have a product with a known expiry date, and the correct strategy is to extract cash quickly rather than raise on it.
  • SourcesSimpleClosure, State of Startup Shutdowns 2025 · TechCrunch on DevDay's PDF feature · TechCrunch on 2025 shutdown rates

Read the wider evidence

This entry is reproduced from the supplied research, with its inline source links retained. It has not been independently re-reported for this website conversion.

Read the complete chapter, source list, and methodological notes →
← Back to post-mortems