UpShaqo
Intelligence desk
Enterprise Agents Source-backed analysis

Legora Turns Contract Review Into a Multi-Agent Assembly Line

OpenAI says Legora is building an 'agentic operating system' for legal work, chaining models, ChatGPT, and Codex into contract workflows — a shift from chatbot assistance to supervised automation.

UpShaqo Editorial IntelligenceOctober 28, 20266 min read
Intelligence standard

Independent UpShaqo analysis built from fresh, attributed sources. We explain the impact instead of repeating the announcement.

Read for leverage: focus on the workflow change, the customer problem, and the next action—not only the product announcement.

It's 11 p.m. before signing, and the due diligence team is still working through a data room full of vendor agreements. An associate is pasting termination clauses into a spreadsheet by hand, flagging change-of-control provisions one contract at a time, while a partner worries that something buried in document 340 of 600 will surface after close. That bottleneck — not a lack of legal talent, but a lack of time to read everything carefully — is the exact friction OpenAI says it built its newest legal-industry partnership to attack.

OpenAI describes Legora as a company building the agentic operating system for legal work, using OpenAI models, ChatGPT, and Codex to accelerate legal workflows, expand contract analysis, and free lawyers to focus on higher-value work, according to OpenAI's announcement. It's a short, pointed description, but the phrase "operating system" is doing real work. It signals something different from the legal chatbots that have circulated for the past two years.

From Chat Assistant to Standing Infrastructure

A legal chatbot answers one question at a time: summarize this clause, draft this email. An operating system, by contrast, implies something persistent — a layer that sits underneath multiple workflows and coordinates them. Based on OpenAI's own framing, Legora's system is meant to run across contract analysis tasks continuously, not just respond to isolated prompts.

This is UpShaqo's inference rather than a claim OpenAI makes explicitly: an "agentic operating system" for legal work most plausibly means a set of specialized agents — one that ingests and parses documents, one that extracts and classifies risk language, one that drafts or redlines, and one that escalates ambiguous judgment calls to a human lawyer — chained together so the firm doesn't have to manually hand off work between tools. The OS metaphor suggests Legora wants to be the substrate other legal workflows run on top of, not a standalone feature bolted onto existing software.

Why Codex Is the Interesting Half of This Story

Most coverage of AI in law focuses on language models reading documents. But Codex's presence in Legora's stack points to something less obvious: using code-generation capability to build and customize the legal workflows themselves, not just to analyze the legal text inside them.

OpenAI's own engineering team offers a useful, if unrelated, data point on what Codex-class tools can do with dense investigative work. In a separate account, OpenAI engineers describe using Codex Ultrafast to cut incident investigation from hours to minutes during internal security analysis and code refactoring. That example isn't about legal work, and OpenAI makes no claim that Legora achieves the same compression. But it illustrates the kind of task Codex-powered agents are built for: fast, structured investigation across large amounts of material — a capability that maps naturally onto reviewing hundreds of contracts for a narrow set of risk patterns.

The Infrastructure Question Law Firms Rarely Ask

There's a second, quieter development that matters to anyone evaluating a vendor like Legora: where the underlying models actually run. OpenAI recently announced that its frontier models are coming to Google Cloud's Model Garden, available through Google Cloud Marketplace. Law firms and corporate legal departments are often locked into specific cloud environments for compliance and data-residency reasons — a constraint that has historically slowed AI adoption in regulated industries more than model quality ever did.

If OpenAI's models become available inside the cloud environments legal clients already use and have already vetted, vendors building on top of those models — Legora among them — gain a shorter path to enterprise deployment. This is analysis, not a confirmed fact about Legora's architecture specifically, but it's a reasonable reading of why OpenAI is pursuing multi-cloud distribution at the same time it's promoting a legal-industry partnership.

A Hypothetical Scenario: Due Diligence at Scale

Consider — purely as an illustrative, hypothetical scenario rather than a reported case — a midsize firm handling due diligence for a mid-market acquisition. Under a traditional model, associates divide the contract batch manually, each reading assigned documents and logging findings into a shared tracker. Partners review flagged issues in batches, often under deadline pressure in the final 48 hours before signing.

Under an agentic-OS model like the one OpenAI describes, the workflow could instead look like this:

  • An intake agent ingests the full document set and normalizes formats.
  • An extraction agent flags clauses matching a defined risk taxonomy — change-of-control, indemnification caps, termination triggers.
  • A drafting agent proposes redlines or summary memos for each flagged issue.
  • A lawyer reviews only the exceptions the system surfaces, rather than every page of every contract.

The lawyer's role shifts from first-pass reader to final-judgment reviewer. That's consistent with OpenAI's stated goal of freeing lawyers for higher-value work — but it also concentrates risk differently. If the extraction agent misses a clause type outside its defined taxonomy, no human may ever see that page at all. The quality of the review now depends heavily on how the risk taxonomy was built and tested, not just on the volume of documents a human could get through.

The Tradeoffs Nobody Can Skip

Agentic legal tooling raises questions that are structural, not hypothetical glitches:

  • Confidentiality and privilege. Routing client documents through third-party model infrastructure raises questions about data handling that firms must resolve before deployment, regardless of which cloud hosts the model.
  • Malpractice exposure. If an agent's output informs legal advice and something is missed, the liability framework for AI-assisted review is still unsettled in most jurisdictions.
  • Verification burden. Supervising agents at the exception level only works if the exception-detection logic is trustworthy — and trust has to be earned through audit, not assumed from vendor marketing.
  • Vendor and infrastructure dependency. A firm that builds workflows around one agentic OS inherits that vendor's roadmap, pricing, and uptime as a dependency for client-facing work.

What Legal and Business Leaders Should Do Next

For firms and in-house teams evaluating tools like this, a few concrete steps follow directly from the shift OpenAI is describing:

  • Pilot agentic contract review on a low-risk document category first — NDAs or standard vendor terms — before extending it to deal-critical diligence.
  • Ask any legal-AI vendor explicitly what happens when a document falls outside its risk taxonomy, and who is accountable for that gap.
  • Map which cloud environments your firm's compliance policies permit, since multi-cloud model availability may expand or restrict which vendors are viable.
  • Redesign partner review time around exception-handling rather than first-pass reading, and track whether that actually reduces hours or simply shifts where the hours go.

The bigger shift here isn't that AI can read a contract — it's that OpenAI and its partners are betting lawyers will supervise systems instead of reading documents directly. Whether that bet pays off depends less on model capability than on how rigorously firms test the gaps between what the agent catches and what it misses.

Sources

#Legal Tech#AI Agents#OpenAI#Codex#Enterprise AI#Contract Review

Two doors. Pick one.

Hire the team.
Or become it.