500 investment bankers review AI outputs and find none ready for client delivery

The more telling number isn't the zero client-ready outputs — it's that over half of 500 working bankers said they'd use these drafts as a starting point anyway. That's not a rejection of AI; it's a description of how AI actually enters professional workflows: through the back door, as a labor-saving first draft rather than a finished product.

This is largely disconnected from recent activity in our archive, as we have no prior coverage to anchor it to. But it belongs to a broader pattern visible across professional services: the adoption curve for AI in high-liability work tends to follow a 'draft layer' model, where the tool earns its place by reducing time-to-first-draft rather than by replacing final judgment. Investment banking is a particularly sharp test case because the cost of a client-facing error is reputational and regulatory, not just operational.

Watch whether either OpenAI or Anthropic responds to this benchmark by publishing their own task-specific fine-tuning results for financial document work within the next two quarters. If they do, it signals they see professional services reliability as a competitive front worth fighting on publicly; if they don't, the 'draft layer' framing likely becomes the default positioning by default.