Legal ops · Production

Harvey integrates OpenAI Deep Research in under 12 hours using modular AI architecture

The problem

Harvey needed to rapidly integrate OpenAI's newly released Deep Research API, facing challenges including the API being too new for base LLM knowledge, complex streaming with many dynamic output types, and the need for code compatibility with Harvey's internal Workflow Engine and thinking states.

Workflow diagram · grounded in source
1
User engages Harvey agent
trigger
“a number of product surfaces that allow users to engage with Harvey agents”
2
AI blocks execute workflow
ai_action
“These blocks can be sequenced together by a developer or, in an agentic system, called as tools to create and execute end-to-end workflows for a particular legal task”
3
Thinking states shown to user
output
“With our native "thinking states", we provide users with visibility into an agent's plan and how decisions are made with the user in the loop”
4
User reviews and intervenes
human_review
“Intervene at any step—adding context, tweaking parameters, or rerunning actions—to course-correct and ensure the final output matches their intent”
5
Citations traced and verified
validation
“We successfully incorporated the URL citations produced by Deep Research into our citation system, so every step of the workflow can be traced and verified”
Reported outcome

Harvey successfully integrated Deep Research within less than 12 hours of the API's release, incorporating URL citations into their existing citation system so every step of the workflow can be traced and verified.

Reported metrics
Deep Research integration timeless than 12 hours
Reported stack
HarveyDeep ResearchWorkflow EngineStreamlitOpenAI
Source
https://www.harvey.ai/blog/integrating-deep-research-into-harvey
Read source ↗

Frequently asked questions

What did this team achieve with this AI workflow?

Harvey successfully integrated Deep Research within less than 12 hours of the API's release, incorporating URL citations into their existing citation system so every step of the workflow can be traced and verified.

What tools did this team use?

Harvey, Deep Research, Workflow Engine, Streamlit, OpenAI.

What results were reported?

Deep Research integration time: less than 12 hours (source-reported, not independently verified).

How is this legal ops AI workflow structured?

User engages Harvey agent → AI blocks execute workflow → Thinking states shown to user → User reviews and intervenes → Citations traced and verified.

Related legal ops cases