OpenAI and Thrive Holdings built Tax AI, a Codex-powered agent that helps prepare complex tax returns while preserving evidence for accountant review.

In this episode, Corey and Grant talk with OpenAI’s John de Wasseige and Arthur Fernandes Araujo about how expert corrections become structured signals, how Codex turns repeated failures into evals and scoped engineering tasks, and why the best AI deployments still need humans close to the work.

They also dig into what this pattern could mean for bookkeeping, audits, IT help desks, and other expert workflows where the system can measure what “right” looks like.

Relevant links:

OpenAI Tax AI case study: https://openai.com/index/building-self-improving-tax-agents-with-codex/

OpenAI Codex: https://openai.com/codex/

Harness engineering: https://openai.com/index/harness-engineering/

Thrive Holdings: https://www.thriveholdings.com/

Crete: https://www.cretepa.com/

Subscribe to The Neuron newsletter: https://theneuron.ai

Can AI Agents Learn From Expert Corrections?

The Neuron: AI Explained

Can AI Agents Learn From Expert Corrections?