Story perspectives

AI Models Excel in Complexity but Falter in Basics

4/19/2026

Summary

Story summary

David Meyer, senior vice-president at Databricks, says state-of-the-art AI models excel at complex tasks but struggle with everyday enterprise functions.
These models often make errors, such as correcting mistakes on invoices instead of simply identifying them.
Anthropic’s Claude performs well in coding but lags in data engineering tasks.
Meyer suggests smaller, open-source models refined with reinforcement learning for specific tasks, trained at lower cost than state-of-the-art models.

1 of 1

Story summary

David Meyer, senior vice-president at Databricks, says state-of-the-art AI models excel at complex tasks but struggle with everyday enterprise functions.
These models often make errors, such as correcting mistakes on invoices instead of simply identifying them.
Anthropic’s Claude performs well in coding but lags in data engineering tasks.
Meyer suggests smaller, open-source models refined with reinforcement learning for specific tasks, trained at lower cost than state-of-the-art models.