Autoresearch for business forecasting

Your AI Could Be Running 100 Forecasting Experiments While You Sleep

Imagine you wanted to improve your revenue forecast. Traditionally, you’d hire an analyst, wait weeks for results, test one approach at a time, and hope they picked the right method. A new open-source tool from former Tesla and OpenAI researcher Andrej Karpathy has turned that process on its head. His concept, called autoresearch, lets an AI agent run roughly 100 experiments overnight on a single computer — testing different forecasting approaches, keeping what works, discarding what doesn’t, and repeating until morning. The only thing a human needs to provide is a plain-text document describing the business problem and what matters most. No coding. No data science degree. As Karpathy put it, “the human iterates on the prompt, the AI agent iterates on the training code.” You describe the goal; the machine figures out how to get there.

For business users, the breakthrough is not the technology itself but what it means for accessibility. The AI agent already understands modelling techniques, statistical methods, and optimisation strategies as that knowledge is built in. What it lacks is your expertise: which products are seasonal, why March is always slow, what happens when a competitor opens nearby, or which cost lines are most volatile. By writing your business knowledge into a simple strategy document, you give the agent the context it needs to run meaningful experiments against your actual sales, cost, and margin data. It might discover that your revenue forecast improves dramatically when weather data is included, or that a different modelling approach captures your seasonal patterns far more accurately than whatever your current spreadsheet assumes. Each of these discoveries would have taken a human analyst days. The agent finds them in minutes and moves on to the next idea.

The implications for financial planning are significant. Rather than relying on a single forecasting method chosen by one person, businesses can now let an AI systematically explore dozens of approaches tuned to their specific data. A convenience store chain and a professional services firm have completely different revenue patterns — and this tool adapts to each automatically. The original article on Garry’s List describes this shift clearly: the human role is moving from doing the analysis to designing the environment in which the AI analyses. For business owners, that means your competitive advantage is no longer about having the best analyst. It is about knowing your business deeply enough to point the machine in the right direction. The technology handles everything else.

Your competitive advantage isn’t having the best analyst.
It’s knowing your business well enough to point the AI
in the right direction.”