OpenAI’s Codex Powers New “File My Taxes” Tool, Promising Error‑Free Returns

OpenAI’s Codex 5.3 proved its mettle in a head‑to‑head test that many tax professionals would consider a “John Henry” showdown. Kyle Corbitt of Corbt documented the experiment, noting that he ran his 2025 return through both a seasoned accountant and an autonomous Codex agent, then compared the outcomes. The accountant, despite years of experience and a fee that ran into the thousands, missed a filing error that forced a corrective IRS notice. Codex, by contrast, returned a clean submission with no discrepancies, even after parsing a 10‑minute, unstructured voice dump that covered seven income streams, multiple K‑1s, crypto trades, a company sale with three compensation types, charitable stock donations, and state‑specific capital‑gain calculations. Corbitt concluded that “only one succeeded,” underscoring the AI’s capacity to handle tax complexity that would normally require multiple rounds of client‑accountant clarification【Corbt】.

The experiment also highlighted how Codex leverages context‑rich prompting to replace the back‑and‑forth that defines traditional tax preparation. Corbitt fed the model a decade‑long archive of his financial documents, then asked it to ingest his full 2024 return before launching into the 2025 “braindump.” Codex parsed the raw audio, extracted relevant data points, and generated a step‑by‑step filing checklist that included items most human preparers would overlook, such as partial‑year ACA coverage nuances and Washington state capital‑gain formulas that differ from federal rules. The model’s ability to maintain fidelity across disparate data types—PDF forms, CSV transaction logs, and spoken narratives—mirrors the multimodal capabilities OpenAI showcased in its GPT‑5.3‑Codex rollout, which Ars Technica reports is being positioned for “more than just code generation”【Ars Technica】.

From a practical standpoint, the Codex‑driven workflow cut preparation time dramatically. Corbitt estimates that the AI completed the bulk of the tax logic in under two hours, whereas his accountant required multiple days of document exchange, clarification calls, and manual recalculations. The speed advantage is not merely a convenience; it translates into cost savings that could reshape the market for high‑net‑worth individuals who currently spend upwards of $5,000 on annual tax services. TechCrunch notes that OpenAI is actively marketing Codex as a “developer‑first” tool that can be embedded into bespoke financial applications, hinting at a future where boutique tax‑tech firms might license the model rather than rely on human expertise【TechCrunch】.

Industry observers caution that Codex’s success in a single, albeit complex, filing does not guarantee universal reliability. Wired’s coverage of OpenAI’s broader ambition to make ChatGPT the “future operating system” emphasizes that regulatory compliance and auditability remain open questions for AI‑generated tax returns【Wired】. Nonetheless, the Corbt test provides a concrete data point: an AI model can, under the right prompting conditions, produce an error‑free return that satisfies the IRS’s technical requirements. As the tax code continues to evolve—particularly with emerging areas like digital asset taxation—AI agents that can ingest and synthesize heterogeneous data streams may become indispensable allies for both individuals and firms.

The broader implication is a potential shift in the accountant’s role from primary preparer to overseer of AI‑driven processes. If Codex can reliably handle the mechanical aspects of filing, human professionals may focus on strategic tax planning, risk assessment, and client education—areas where nuanced judgment still trumps algorithmic calculation. Corbitt’s experiment, while anecdotal, signals that the technology is reaching a maturity level where it can be trusted with high‑stakes financial outcomes. As OpenAI refines Codex and integrates it deeper into its ecosystem, the “file‑my‑taxes” promise may soon move from a novel side project to a mainstream service, reshaping the tax preparation landscape for the first time in decades.