The problem is not legal AI, it’s lost context

A partner asks an associate to research an employment dispute. The associate starts in one legal AI tool, uploads a chronology, identifies the jurisdiction, frames the employment status issue, and runs several research prompts. Then they need to move into a second platform for drafting, document analysis, or authority checking.

At that point, much of the work starts again.

They re-explain the matter. They paste case facts into another prompt. They restate the key legal questions. They reconstruct the research trail, often from notes that don’t capture every qualification or assumption. The next tool may produce a useful answer, but it doesn’t know why the first answer mattered.

That is the friction DeepJudge’s Agent Handoff Protocol is trying to address. The protocol, announced with backing from Harvey and Thomson Reuters, is designed to help legal AI systems pass relevant context between tools and agents. The practical appeal is simple. A lawyer should not need to re-prompt an AI platform every time a research task moves to another system.

For a law firm owner or managing partner, this isn’t just a technology question. It is a margin question.

Legal teams can lose four to six hours per attorney each week to work that never reaches an invoice. Some of that is necessary professional work. A meaningful portion is rework, internal status chasing, matter administration, and research tasks repeated because systems do not retain the right context. In firms doing $1 million to $25 million in annual revenue, we commonly see total leakage land somewhere around $80,000 to $250,000 a year.

An AI handoff protocol will not solve all of that. It could remove a frustrating slice of duplicated research work if it works reliably across the tools your firm already uses.

The key word is if.

What an AI handoff should preserve

A useful handoff is not a giant transcript copied from one assistant to another. That creates noise, increases risk, and makes it harder for the next lawyer to validate the output.

For legal research, the receiving tool needs a structured matter packet. Think of it as a research brief that moves with the task.

That packet should include:

• Matter identifier and approved client or pseudonym reference
• Jurisdiction, venue, governing law, and relevant dates
• The question to be answered, stated precisely
• Confirmed facts, disputed facts, and facts still to verify
• Documents reviewed and their source location
• Search terms, authorities found, and citations used
• Research conclusions, including confidence and limits
• The next requested action, such as draft a memo or test a counterargument
• Access and confidentiality rules for the matter

Take a commercial litigation example. Your associate uses a research platform to assess whether a limitation of liability clause could be challenged for ambiguity under New York law. The first system identifies a line of cases, highlights wording patterns, and finds two facts in the contract that may change the analysis.

The next system should receive more than the prompt, “Can you draft a memo about this?”

It should receive the exact clause, the governing law, the client’s commercial objective, the cases already reviewed, the factual uncertainties, and a request for a short partner-ready memo that distinguishes adverse authority. That is a handoff with operational value.

Without it, the associate spends another 20 to 40 minutes recreating context. The drafting tool may also miss a limitation that the research tool identified earlier.

For firms using multiple platforms, that lost continuity compounds quickly across matters.

Why DeepJudge’s announcement matters to law firms

DeepJudge’s announcement matters because it points to a more practical model for legal AI. Firms do not need one giant tool to do every task. They need controlled movement of context between the tools that do specific jobs well.

A research platform may be strong at locating authority. A drafting tool may be better for organising a memo. Your document management system holds the governing source files. A workflow agent may be better at creating tasks, collecting approvals, and recording what was done.

The gap has been the space between those systems.

A handoff protocol could provide an agreed way for software agents to identify the task, package the approved context, pass it to another tool, and preserve a record of the exchange. That is more valuable than another generic chat interface.

It could also reduce a common risk in legal AI adoption. Lawyers start using several tools, each with its own workspace and prompt history. The firm’s knowledge becomes fragmented. Good research sits inside an individual chat. The next person cannot easily trace it, reuse it, or determine what source materials informed the answer.

The protocol is still something to test, not something to assume is production-ready for every practice. Announcements involving respected legal AI providers deserve attention, but they don’t remove your firm’s duties around confidentiality, supervision, privilege, conflicts, citation checking, and data governance.

The right response is not to wait for a perfect standard. It is to run a narrow, controlled pilot around work that currently gets repeated.

For a broader view of where those workflows fit, see Omni for law firms. It focuses on the operational work around legal delivery, not just the latest tool release.

Find the repeated research loop in your firm

Before testing any handoff technology, map the work as it happens now. Don’t start with a vendor demonstration. Start with three recent matters.

Pick matters where research moved between at least two tools or people. Litigation, employment, property, corporate, and family practices all have examples. Look for work where an associate had to explain the same facts several times.

Ask these questions:

1. Where was the matter first described?
2. Who created the initial issue list?
3. Which documents supplied the facts?
4. Which AI tool or legal database was used first?
5. How often was the task re-prompted or rewritten?
6. Did the next lawyer receive the sources and limitations, or only a summary?
7. How was the final answer checked and approved?
8. Was any of that time captured against the matter?

You will probably find that context disappears at ordinary handoff points. A partner gives a voice note. An associate turns it into an email. A paralegal saves documents in the DMS. Research happens in a legal AI tool. Drafting moves to another tool. The partner gets a Word document with no visible trail of how the result was reached.

None of those people are failing. The workflow is.

An AI continuity protocol becomes useful when it removes work at those joins without removing professional judgement. It should give the next agent or lawyer enough verified context to proceed, while making the provenance of the work easy to inspect.

That is different from asking an AI tool to “remember everything.” Firms should not rely on vague memory claims. They need a defined handoff object, defined access rights, and a clear review point.

A practical pilot for legal research handoffs

Keep the pilot small. Choose one practice area, one legal question type, and a small group of lawyers who will actually use it.

A good first use case might be employment advice on restrictive covenants, recurring commercial contract disputes, or discovery issue research. Avoid a high-stakes matter with an urgent hearing date. You want enough repetition to measure the workflow, without creating avoidable delivery risk.

Run the pilot across 10 to 20 research requests over four weeks.

Start by setting a common intake brief. Each request should capture the legal question, jurisdiction, matter facts, approved documents, desired output, deadline, and reviewer. Then define when a task can move from research to drafting.

The research agent or researcher creates a handoff record containing:

• The question asked
• The source documents and document references
• The research path and key search terms
• Authorities located, with citations to verify
• Relevant excerpts and short reasoning notes
• Open factual or legal questions
• A clear instruction for the receiving tool or lawyer

The drafting stage then receives the approved packet. It should produce a bounded output, such as a one-page research memo, a chronology, a list of counterarguments, or a draft advice note. It should not invent missing facts. It should identify any authority that needs human validation.

Finally, a supervising lawyer checks the citations, reasoning, and factual assumptions before anything reaches a client.

Measure four things:

• Minutes from research assignment to first usable draft
• Number of times facts or instructions were re-entered
• Number of material corrections at review
• Time recorded, written off, or never entered against the matter

If the protocol saves 15 minutes on a research-to-draft handoff, that may sound minor. Across 15 matters a week, 48 weeks a year, and several fee earners, it becomes meaningful. At associate billing rates often sitting in the $200 to $400 range, the economic effect depends on how much saved capacity becomes billable work, faster turnaround, or reduced overtime.

Don’t count a theoretical saving as profit. Count the work you can see.

If you want help identifying the best pilot and measuring the operational result, Book a 60-min Omni Audit. It is a working session, not a slide deck.

Where Omni agents fit around the research handoff

Research continuity is important, but it sits inside a wider chain of firm operations. The strongest result comes when matter context is captured well before research begins.

The Intake Voice Agent answers calls after hours, at lunch, and on weekends. It captures the caller’s details, runs the first layer of conflict-check information, records the nature of the issue, and books an appropriate consultation directly into the firm calendar. For practice areas that get a high volume of urgent enquiries, this prevents a promising lead from sitting unanswered until the next business day.

Industry operators often see 30 to 40 percent of after-hours legal intake fail to convert when there is no timely response. Not every missed call is a good fit. Still, a firm should not lose a suitable matter just because the phone went unanswered at 6:15 pm.

The Matter Triage Agent then reviews web forms, inbound emails, and captured call notes. It classifies the practice area, scores fit against your criteria, identifies missing information, and routes the lead to the appropriate partner or team. The routed brief should contain one clear paragraph that tells the receiving lawyer what happened, what the client needs, what deadlines may exist, and what needs to be checked next.

That brief can become the beginning of the matter context used later in legal research. The legal issue is not retyped from a receptionist note, then from an email, then again into an AI prompt.

Once the matter is active, the Document Review Agent can perform a first pass across contracts, discovery batches, and matter files. It flags relevant clauses, extracts dates and parties, groups documents by issue, summarises positions, and produces an associate-grade memo for review. It does not replace the supervising lawyer. It reduces the amount of time a junior associate spends locating the basic shape of the file before legal analysis begins.

You can see the operating model behind these workflows in Omni Ops and learn how the phone layer works through Omni Voice. The point is not to add three disconnected AI products. It is to create one controlled chain from first contact to matter triage, document review, research, and lawyer approval.

Questions to ask before you connect legal AI tools

Legal technology teams should be demanding here. A handoff protocol needs better answers than “the AI can share context.”

Ask prospective providers and internal stakeholders:

• What exact data leaves each system during a handoff?
• Can the firm limit handoffs by client, matter, practice group, or user role?
• Is the handoff logged with a timestamp and an identifiable source?
• Can a lawyer see the documents, instructions, and research outputs that informed the next step?
• How are client confidentiality and privilege handled in the connection?
• Are matter-specific retention rules applied?
• Can the firm stop, revoke, or correct a handoff?
• What happens if the receiving system cannot access a cited document?
• Does the system distinguish source content from AI-generated summaries?
• How are conflicts handled before a new matter context is passed into a workflow?

Your answers will vary based on firm size, jurisdiction, client requirements, and existing agreements with technology providers. A five-lawyer employment practice will not need the same governance model as a 200-lawyer national firm. Both still need a named owner for the workflow and a clear review standard.

Don’t let the technical discussion obscure the commercial question. Are your lawyers spending less time repeating administrative work and more time applying judgement where clients value it?

Use the intake checklist before you automate

AI research handoffs are only as good as the facts and instructions entering the system. If intake notes are incomplete, the AI can move incomplete context faster.

Our AI Client Intake Checklist for Law Firms gives you a practical worksheet for reviewing what your firm captures at first contact. It covers the client and opposing-party details needed for conflicts, matter facts, urgency, documents, consent, routing, and follow-up ownership.

You can also download the checklist directly and use it with your intake team this week. Take five recent enquiries and check how much of the required context was available when the responsible lawyer first reviewed them.

That exercise often exposes the same issue visible in AI research. People are working hard, but critical context is arriving late, scattered across systems, or not captured at all.

Make continuity measurable, then expand carefully

DeepJudge’s Agent Handoff Protocol is worth watching because it addresses an everyday frustration in firms that use more than one legal AI tool. Context should not disappear every time a task changes platforms.

Start with a contained research workflow. Build a structured handoff record. Keep human review at the point where legal advice is formed. Measure re-prompting, turnaround time, corrections, and recovered capacity.

Then look upstream. Better intake and triage create better matter context. Better document review gives research a clearer factual base. Better handoffs make each stage easier to supervise.

If your firm wants to identify where that chain is breaking, start with the AI audit for law firms. In 60 minutes, we map the manual workflow, identify the highest-value automation opportunities, and outline a practical next-step plan. No deck, no generic technology pitch.

When you are ready to put numbers against the repeated work in your firm, Book my Omni Audit.