Automating 90% of finance and legal work with agents
OpenAI
Hebbia's AI now handles 90% of grunt work in finance and legal research, built on OpenAI's models. Wall Street and law firms are quietly letting agents do the reading nobody wanted to do anyway.
Based on reporting by OpenAI — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Hebbia has been quietly embedding itself inside some of the biggest names in finance and law, and the pitch is blunt: their deep research agents now knock out roughly 90% of the document-heavy grind that used to eat analyst hours. That's not a marketing rounding error. It's the difference between a junior associate spending three days on a diligence pull and getting a first pass back before lunch.
The system runs on OpenAI's models under the hood, chaining together retrieval, reasoning, and summarization steps so it can dig through filings, contracts, and disclosures the way a trained researcher would, just without the coffee breaks. Instead of a single prompt-and-answer loop, Hebbia's agents branch out, pull from multiple sources, cross-check figures, and stitch together something closer to a memo than a chatbot reply.
What's notable is who's using it. This isn't a startup selling to other startups. Hebbia counts banks, private equity shops, and law firms among its clients, the kind of institutions that move slowly on tech adoption because a wrong number in a filing can cost real money or trigger real liability. Getting those firms comfortable enough to hand off 90% of a workflow to an AI agent says something about how far trust in these tools has moved in a short window.
The economics are the real story here, though. Finance and legal work has always billed by the hour partly because the research itself was slow and manual. If agents compress that bottleneck by an order of magnitude, the value shifts from who can read the most documents to who can ask the sharpest questions and judge the output. That's a different skill set, and it's going to reshuffle who's valuable inside these firms faster than most people expect.
OpenAI, for its part, gets a flagship case study that isn't another coding demo or customer service bot. Deep research applied to finance and law is a harder sell technically, given how unforgiving the domains are with errors, and Hebbia pulling it off at this scale is as much a vote of confidence in the underlying models as it is a product win.
My take — AI-written commentary, not fact-checked reporting
I'll believe the 90% figure once a few of these firms get audited after a bad quarter, but directionally this is exactly where agents should be aimed: boring, expensive, document-heavy work that nobody romanticizes. The people who'll get hurt aren't senior partners, they're the first-year analysts whose entire training path was doing this grunt work themselves, and firms need to figure out how junior talent still gets built once the agent does the reps for them.
Read more about this at: OpenAI