7-lesson course · about 18 hours

Coordinate proactive legal AI agent teams around one shared matter record

For Attorneys, paralegals and in-house legal teams

Everything for Legal teams

Build a Factory of three agent teams on a sample matter. Then test it with handoffs, late evidence, blocked work and a restart. Nothing is released until the attorney approves it.

Try lesson 1 freeSee what you’ll learn

What you’ll be able to do

By the end, you can…

Give every agent a defined job on a defined teamYou'll assign intake, chronology and evidence agents to a Matter Intelligence team, research, drafting and synthesis agents to a Work Product team, and source-checking and release-checking agents to a Review team. Each agent gets its inputs, outputs and stop rules. You'll find the places where two agents could change the same output, and you'll make sure no agent can expand its own access or release authority.
Build a shared matter record that handoffs pass throughYou'll set out the record's sections, each team's read and write rights, an approved source snapshot, a shared assignment ID, and the fields every entry must carry. You'll also write the acceptance criteria that work has to meet before it moves to the next team, and decide what may be shared with outside counsel and on what terms.
Have Dots or Claude Duet report back, and check what it saysYou'll pick OpenAI Dots or Claude Duet to follow the approved record across team handoffs. It reports source-linked changes, draft status and unresolved evidence questions. You'll check that account against the record and flag any statement the record doesn't support.
Make a late correction reach every affected deliverableYou'll trace a seeded contradiction or a late piece of evidence back to its sources and pause the work it affects. Then you'll ask the attorney to resolve it, and confirm the correction reaches the chronology, research notes and draft, not just one document. You'll keep allegations from becoming facts, and stop a newer document from quietly outranking a better-supported one.
Test the Factory under pressureYou'll run the Factory with changing inputs, withheld sources, changed permissions, blocked work and an interrupted run that restarts later. You'll check for duplicate work, check that it resumes from the last accepted step without losing the reason for each change, and check that every exception has an owner.
Set attorney gates and write down the limitsYou'll set approval gates for sources, conflicting evidence and release, and keep a gate record of what was reviewed, the decision, who made it and when. Rejected work goes back to its owner with reasons. To finish, you'll write a limitations register: what the Factory hasn't been shown to handle, and what needs more evaluation before anyone considers it for a real matter.

How it works

A tutor that meets you where you are

  1. Lessons that adapt to your levelYou'll work through seven lessons in three modules: Factory design, handoffs and contradictions, and attorney release. Each lesson adjusts to what you already know. If you've built strong agents in Dots or Claude Duet, it moves quickly. If a concept such as acceptance criteria or a shared assignment ID is new to you, it slows down and works through it on your sample matter.
  2. An AI tutor you can ask anythingYou can type your questions or ask them out loud. Ask the tutor why a handoff entry needs a certain field, how to word a stop rule, or whether your gate design holds work where it should. It works from the course material and your sample matter, and it treats you as a working legal professional.
  3. Quizzes and flashcards for each lessonEach lesson has its own short quizzes and flashcards. They cover the distinctions that matter in this work: fact versus allegation, accepted versus pending work, and which decisions stay with the attorney.
  4. Progress tracking and a certificateYour progress is saved across lessons, so you can stop and pick up where you left off. When you finish the course, you receive a certificate of completion.

Curriculum

The curriculum

The course follows the Legal track's three Capstone days. First you design the teams and the shared record. Then you connect the teams through review and test them against contradictions. Finally you tighten the instructions, set attorney gates and run the whole Factory through short update cycles.

Module 1Give the Factory clear responsibilities2 lessons

  1. 1Assign roles across the proactive AI agent teams
    • Assign agents to the Matter Intelligence, Work Product and Review teams
    • Define each agent's job, inputs, outputs and stop rules within its team
    • Identify overlaps where two agents could change the same output
    • Explain which decisions no team may take
  2. 2Define the shared matter record
    • Build the list of sections in the shared matter record
    • Define read and write permissions for each team
    • Choose the fields every record entry must carry
    • Explain what makes the record 'approved' and why handoffs go through it

Module 2Connect the proactive AI agent teams through review2 lessons

  1. 3Pass work between the proactive AI agent teams
    • Build a handoff entry that tells the receiving team what changed and what to do next
    • Choose OpenAI Dots or Claude Duet to follow the approved record across team handoffs
    • Check the tool's account of source-linked changes, draft status and unresolved evidence questions against the record
    • Identify statements in the account that the record does not support
  2. 4Reconcile conflicting outputs
    • Test whether a late piece of evidence reaches every affected deliverable
    • Identify contradictions between deliverables and the record entries each relies on
    • Build a correction that is made in the record and propagated to each deliverable
    • Prepare unresolved conflicts for the attorney queue and the decision record draft

Module 3Approve a reusable matter system3 lessons

  1. 5Refine proactive AI agent team instructions
    • Identify the instruction responsible for each defect found in Modules 7 to 9
    • Revise that instruction and record why it changed
    • Check that key terms mean the same thing across all teams
    • Test each revised instruction on the sample matter
  2. 6Set attorney approval gates
    • Define the approval gates the Factory cannot pass without the attorney
    • Build a gate record showing what was reviewed, the decision, who made it and when
    • Explain why privilege determinations and discovery requests stay with the attorney
    • Test that the Factory holds work at each gate
  3. 7Run and review the Factory
    • Test the Factory with changing inputs, blocked work and a later restart
    • Check that it resumes from the shared record without losing the reason for each change
    • Review the account of source-linked changes, draft status and unresolved evidence questions against the record
    • Identify what needs further evaluation before use on a real matter

Who it’s for

Made for you if…

Made for: Attorneys, paralegals and in-house legal teams

Attorneys who direct matter preparation

You already check single tasks at the source and verify authority yourself. This course shows you how to make sure that's still true when several agent teams are working on the same matter. You set the gates, and nothing goes out without your approval.

Paralegals and case managers who run agent workflows

You've built intake, chronology, drafting or evidence agents on a sample matter. Here you learn to coordinate them: who writes to the record, how work moves between teams, and what happens when evidence arrives late or a run stops partway through.

In-house legal team members

You'll design a shared record with clear permissions, including what's shared with outside counsel. You'll also produce a written account of what the system can and can't yet be trusted to do.

Questions

Good to know

What do I need before I start?

You should have completed days 1–7 of the Legal track, or have equivalent experience. That means you've built source-linked fact lists, verified authority and written client explanations, and you already have working intake, chronology, drafting and evidence agents on a sample matter in OpenAI Dots or Claude Duet. You'll also need access to one of those tools, with app connections approved for non-confidential sample material only, and a working knowledge of a lawyer's duty to review AI output (ABA Formal Opinion 512 or your local equivalent).

Can I use this on a live client matter?

No. All the work happens on a bounded sample matter with no privileged, confidential or identifying client information. Building the Factory in a course doesn't show that it's suitable for live work. That's why the course ends with a limitations register listing what still needs evaluation.

Should I use OpenAI Dots or Claude Duet?

Either one works, and the course teaches them as equals. Both ask for approval before taking consequential actions, and in Dots, background work is read-only. Coordinated 'teams of dots' is OpenAI's stated vision, not a current product. In this course you design the team structure yourself, and the tool you choose follows the approved record across the handoffs.

Will the agents make legal decisions?

No. Privilege determinations, discovery requests and legal advice stay with the attorney and are never given to agents. The attorney verifies authority in an appropriate legal research source and approves any advice, communication or filing before it's used.

What will I have at the end?

You'll have a working Factory on your sample matter: role definitions for three teams, a shared matter record with permissions and acceptance criteria, handoff entries, a gate record, and source-linked deliverables such as an issue matrix, review questions, an attorney queue and a decision record draft. You'll also have a limitations register.

How long does it take?

Plan on about 18 hours across seven self-paced lessons.

Does it cover court rules on disclosing AI use?

No. Standing orders and local rules on disclosing AI use aren't covered, so check the rules that apply to you. The course also doesn't cover building individual agents from scratch, single-task prompting, vendor pricing comparisons, billing models, or custom code outside Dots and Claude Duet.

Legal AI Capstone: build and test a Factory of proactive agent teams

$397

One payment · lifetime access

  • 7 self-paced lessons in 3 modules, about 18 hours in total
  • Build Matter Intelligence, Work Product and Review teams around one shared matter record
  • Have OpenAI Dots or Claude Duet follow the approved record, and check its report against it
  • Recovery tests: late evidence, changing inputs, blocked work, an interrupted run and a restart
  • Attorney approval gates, a gate record and a limitations register
  • An AI tutor, quizzes and flashcards for each lesson, progress tracking and a certificate of completion

Secure checkout by Stripe. Have a promotion code? Enter it at checkout.

Refunds: Full refund if you ask before you open your first lesson. Once you’ve started, purchases aren’t refundable, except where the law gives you that right.

Already bought it? Sign in