Who owns the work in an agent-assisted legal practice?

Which operating responsibilities remain with lawyers in an agent-assisted practice?

Automating a task does not assign professional judgment, client communication, supervision, or accountability to software. A practice needs explicit owners for the engagement, data permissions, review, client decisions, external actions, billing, and error correction. The exact duties come from the governing jurisdiction and engagement, not from a product's automation features.

Start with the governing professional framework

ABA Formal Opinion 512 discusses competence, confidentiality, communication, supervision, candor, and fees in generative-AI use. ABA Model Rule 5.3 addresses supervisory responsibilities for nonlawyer assistance. These are model materials; applicable jurisdictional rules must be checked. [S1] [S2]

• Identify the licensed professional responsible for the matter.
• Identify the actual engagement scope and client decision rights.
• Check local rules, court orders, and professional obligations for the intended use.
• Distinguish a software permission from authority to exercise professional judgment.

A proposed responsibility map

Assign one accountable owner for each result, even if several people or tools contribute. Name the artifact that demonstrates completion so responsibility is observable. This is an original operating template, not a claim that every practice must use these titles.

• Engagement owner: scope, conflicts process, and authorized instructions.
• Data owner: approved sources, destinations, access, and retention decisions.
• Review owner: factual and legal accuracy for the stated use.
• Decision owner: client or authorized person for choices requiring their decision.
• Action owner: final recipient, version, signature, send, filing, or other commitment.
• Correction owner: investigate errors, mitigate effects, and preserve the record.

Synthetic workflow: a client update email

A fictional agent prepares an update from authorized matter notes. The draft accurately summarizes a filed motion but calls a hearing 'confirmed' when only a proposed date appears. The reviewer corrects the status, checks the intended recipients, and approves a specific version. The sender then records the actual transmission result. Draft approval and sending remain different events.

• Source review resolves the hearing-status claim.
• Client communication owner checks clarity and relevance.
• Authorized sender verifies recipients and attachment versions.
• A failed or unknown send outcome is reconciled before another attempt.
• The task closes only after the intended result is evidenced.

Inspect the whole workflow, including exceptions

A practice review should examine where work can escape supervision: reused drafts, changed attachments, stale instructions, direct tool access, and unreviewed exception paths. Include time spent checking and correcting output when assessing efficiency or billing. The conclusion should reflect observed work, not a vendor claim.

• Sample completed tasks and reconstruct source-to-result history.
• Check that required client decisions were actually recorded.
• Review who can override a hold and how the reason is preserved.
• Track errors reaching external recipients separately from errors caught internally.
• Update the smallest responsible procedure after a verified failure.

Scope and limitations

Independent practice-design research grounded in ABA model materials. This domain is not a law firm, does not offer representation, and does not establish an attorney-client relationship.

No firm's compliance, lawyer's conduct, client matter, billing record, or operational system was audited.

Sources

1. American Bar Association — Formal Opinion 512 (2024) — ABA interpretation of Model Rules for generative AI, including competence, confidentiality, supervision, communication, and fees; local law must be checked. Checked 2026-11-05
2. ABA Model Rule 5.3 — Responsibilities Regarding Nonlawyer Assistance — Model-rule supervisory responsibilities; not proof of any jurisdiction's enacted rule. Checked 2026-11-05

Explore further

Agentic Attorney at https://agenticattorney.com/ is a separate platform for exploring legal-team AI implementation and workflow assessment. This research page is not a law firm or a route to automatic representation. Explore Agentic Attorney.