Product-market fit has a prerequisite that most AI founders ignore. Before the market can pull your product, the model must be capable of doing the job. That's Model-Market Fit.
The core assumption is that application builders don’t have much influence on model performance - so it has to be good enough when the model comes out of the lab.
I think it’s a correct assumption - in-context learning is not really learning at all - but what a weird world this creates!
For example, you rightfully talk about how models are bad on financial workflows, and hence the whole sector of applications is stalled. But when OpenAI gets to prioritize training GPT for financial workflows, all it takes is to hire a hundred investment bankers to label data for fine tuning (https://fortune.com/2025/10/22/sam-altman-openai-wall-street-junior-bankers-ai-entry-level-jobs/)
Doesn’t that blow your mind? There are 400,000 investment bankers in US. Why is the whole sector of innovation in financial AI stalled waiting for OpenAI to hire a 100 of them?
Good point. I think financial services are about to be completely disrupted, the same way Claude Sonnet 3.5 disrupted software engineering. The next generation of models will be trained on millions of Excel files and will nail the most complex financial models in one shot. At Fintool, we've done the work of building the best agent for financial services. We already have customers running tens of thousands of queries. The next model generation will 10x our growth.
your insights are super valuable Nicolas. much thanks for sharing. keep them coming ✊
I’ve been thinking about this a lot too.
The core assumption is that application builders don’t have much influence on model performance - so it has to be good enough when the model comes out of the lab.
I think it’s a correct assumption - in-context learning is not really learning at all - but what a weird world this creates!
For example, you rightfully talk about how models are bad on financial workflows, and hence the whole sector of applications is stalled. But when OpenAI gets to prioritize training GPT for financial workflows, all it takes is to hire a hundred investment bankers to label data for fine tuning (https://fortune.com/2025/10/22/sam-altman-openai-wall-street-junior-bankers-ai-entry-level-jobs/)
Doesn’t that blow your mind? There are 400,000 investment bankers in US. Why is the whole sector of innovation in financial AI stalled waiting for OpenAI to hire a 100 of them?
Somebody needs to go disrupt this nonsense.
Good point. I think financial services are about to be completely disrupted, the same way Claude Sonnet 3.5 disrupted software engineering. The next generation of models will be trained on millions of Excel files and will nail the most complex financial models in one shot. At Fintool, we've done the work of building the best agent for financial services. We already have customers running tens of thousands of queries. The next model generation will 10x our growth.