Nicole Diaz had never written code. She then joined OpenAI’s legal team.
One year later, the associate general counsel is now using OpenAI’s apps to simplify and speed up his legal work. She used ChatGPT and the company’s coding agent, Codex, to build a tool to help transform law firms’ thick memos into plain English policies, prioritize employee emails, draft responses, and track results.
“My sense of what’s possible has expanded rapidly over the last six months, even three months,” Diaz said.
Diaz focuses on corporate compliance. This legal department is focused not on the products OpenAI puts out into the world, but on how the company conducts business and making sure its employees follow legal and ethical standards. This is also a feature that companies tend to build in as they prepare for the scrutiny that comes with an IPO.
Over the past year, Diaz has been using OpenAI’s technology to make that work faster and more manageable. She said the tools are not a substitute for her judgment. They absorb repetitive tasks around them.
The decision was shaped by a traditional legal resume. She studied law at Harvard University and learned the mechanics of high-stakes litigation at Skadden, where she worked as an associate. She then moved to the client side as a compliance attorney for Snap. None of those places taught her to build software. OpenAI changed that.
As such, Diaz has become a target of OpenAI’s pitch. The company is betting that its artificial intelligence agents can transform workers who have never thought of themselves as builders into people who create their own tools.
Her way of working also provides a window into the live debates in legal tech. As law firms and corporate legal teams consider whether to purchase specialized legal software like Harvey, Diaz uses the Frontier model directly to show what lawyers can do.
Magic wand for policy making
Diaz first used ChatGPT for one of the least glamorous parts of corporate compliance: policy creation.
Law firms often send out policies chock full of legal jargon and $10 words, but Diaz says such documents may work in more bureaucratic companies, but not with OpenAI. She would manually rewrite these policies and turn them into guidance that employees could actually use.
She created the ChatGPT skill for that purpose.
Skills allow users to turn repeatable tasks into a reusable set of instructions, eliminating the need to create long prompts each time. Diaz created a skill he calls “simplification.” Its job is to take the legal policy, shorten the text, cut out the legal jargon, and match it to the format of OpenAI’s policy template.
Diaz still reviews the output himself and sends the policy back to outside counsel for approval. But she said this skill allows her to create workable drafts faster.
Codex as a colleague
Employees often come to Diaz with questions about whether they need approval to take on an advisory role, invest in a friend’s startup, or attend a fancy dinner with government officials.
Diaz said she previously relied on Google Docs with draft answers to different types of questions. She still had to decide which guidance applied, write a response, and record how the question was handled.
Diaz now uses Codex agents, artificial intelligence that can perform tasks with little intervention, to help with triage. Every day at 5pm, the system scans her inbox for dispute emails, categorizes them by risk, and drafts responses based on pre-written guidance.
A screenshot of the instructions she provided to Codex regarding the “Conflict Triage” workflow. OpenAI
When an employee asks a question about angel investing, they may get a standard answer. Those who wish to join an industry association that also lobbies may receive a response requesting further information. Risky disclosures are flagged to get Diaz’s attention.
The system also keeps a log of the types of disclosures she receives, how she responds, and how long it takes to respond, giving her a better understanding of where employees are stuck and if OpenAI’s policies need more clarity.
She has (ChatGPT) skills
The tool still needs work. One of Diaz’s biggest complaints is that their responses sound too “legal.” She is working on an “about me” file that tells ChatGPT how she writes and speaks. This allows you to incorporate it into different skills and make the output sound more like her.
It’s not something they teach in law school. In OpenAI, learning to build seems to happen mostly through osmosis.
You can’t expect all lawyers to suddenly become Codex power users. Diaz credits his colleague Bright Kellogg with helping lead the legal team through a gentler kind of change management: sharing small lessons each week and highlighting examples of what other lawyers are building.
For Diaz, learning among peers is part of the fun. Looking at other people’s tools can feel similar to exchanging Pokemon cards, she said. One person has a skill that is “really amazing,” and others want to try it out.