Practical agentic workflows that keep client information out
A lawyer’s day needs room for concentrated work, preparation, administration and the unexpected. OpenAI’s dots offer a way to delegate some of the organisation around that work. You assign a bounded goal, let the agent work through the steps, and review what it brings back. Dots can continue between conversations and coordinate multiple tasks, with access depending on rollout, enabled tools and permissions. [1, 2]
For this guide, keep client and matter information entirely outside the workflow. That includes documents, correspondence, names, matter codes, identifying facts, and calendar entries or metadata linked to matters. Renaming a client or removing names from a real document does not establish that it is safe to share. Australian regulators emphasise that professional duties still apply when lawyers use AI. [3, 4]
Start with a firm-approved setup. Check existing connections and context; opening a new chat does not establish separation. Do not connect a practice inbox, matter calendar or client files for these exercises. Use invented schedules, blank templates, public sources and general work preferences. Apply any resulting plan manually in your approved practice systems.
The examples below are proposed workflows, not guarantees of performance. For recurring work, specify the time zone, duration and delivery channel, then check the confirmed schedule. Keep responsibility for court dates and limitation periods in your established human-checked systems. [2, 5]
1. Run a consistent morning planning routine
Give your dot responsibility for starting the planning conversation. It can bring back the same short checklist each morning, helping you make deliberate choices before the day fills up. Decide actual matter priorities privately in your firm’s systems.
Try this prompt: “For the next five weekdays at 8:30 am Australia/Melbourne, send me a planning checklist here: review my approved diary privately, choose a focus period, and leave a buffer. Do not request matter details. Confirm the schedule.”
2. Build and refine a reusable focus timetable
Ask your dot to produce a model working day, check that the time blocks fit, and revise it after feedback. This gives you a reusable starting point for protecting drafting or reading time. Check its arithmetic and manually adapt it to real commitments.
Try this prompt: “Build a fictional 9 am to 5 pm timetable with two 60-minute focus blocks, lunch, general administration and spare capacity. Check for overlaps. Ask me which layout I prefer, then revise it. Do not read or edit any calendar.”
3. Maintain a blank drafting process checklist
Delegate the creation and refinement of a checklist covering planning, drafting, checking sources, proofreading and supervisory review. The output can help you remember process steps when you later work in the firm’s environment. A lawyer must check its completeness.
Try this prompt: “Create a generic drafting-process checklist with blank fields. Review it for missing quality-control steps and give me a revised version for approval. Use no client documents or matter facts, and do not draft legal advice.”
4. Prepare reusable meeting and hearing checklists
Your dot can work through the practical preparation sequence and return a blank template with questions for you to resolve. Keep its scope to preparation habits, such as checking equipment, allocating preparation time and confirming arrangements yourself. Add actual details only in approved systems.
Try this prompt: “Prepare separate blank preparation checklists for a meeting and a hearing. Include a technology check and a reminder to verify arrangements privately. Compare them for omissions, then ask me to approve them. Do not look up any case or list.”
5. Watch public practice updates
Give your dot a small, explicit list of public court or regulator pages to check. It can compare updates and return a source-linked reading list, reducing repeated manual browsing. Check every relevant notice yourself, including its date and applicability; monitoring can miss changes.
Try this prompt: “Each Monday at 8 am Australia/Melbourne for four weeks, check the public news pages of the Supreme Court of Victoria and VLSB+C. Send new practice-related notices here with dates and links, or report access problems. Do not search matters or calculate deadlines. Confirm the schedule.”
6. Coordinate a CPD shortlist
Delegate the search, comparison and follow-up needed to choose professional development. A dot can gather public options against your preferences and bring the shortlist back for a decision. You must verify eligibility, requirements and costs before enrolling.
Try this prompt: “Find three publicly advertised online CPD options on legal ethics. Compare dates, duration, advertised price and provider links. Return a shortlist and ask me to choose. Do not register or pay. If I select one, ask whether I want a reminder.”
7. Batch routine nonclient administration
Ask your dot to assemble scattered general housekeeping into a repeatable session. Its job is to maintain the routine and prompt a review, while the work itself stays with you. Exclude billing narratives, time entries and any administration that identifies a matter.
Try this prompt: “Create a 30-minute weekly routine for desk organisation, stationery checks and reviewing my general training goals. Send it here next Friday at 3 pm Australia/Melbourne and ask which steps were useful. Confirm the reminder. Do not access files or accounts.”
8. Keep a public reading queue manageable
Assign your dot responsibility for organising a general learning topic. It can find public material, remove duplicates and propose a reading order, then revise that order when you choose a focus. This supports learning; it does not establish the law applicable to a client’s circumstances.
Try this prompt: “Build a five-item reading queue about AI governance in Australian legal practice using public regulator sources. Include the publisher, date, link and reason to read each item. Flag undated material. Ask which topic I want first and reorder the queue after my choice.”
9. Rework a model day when capacity changes
Use your dot to test alternatives before you rearrange your day. Starting from an invented timetable, it can identify what no longer fits and show the consequences of different choices. You make any real-world changes after checking obligations privately.
Try this prompt: “Take our fictional timetable and remove 45 minutes of available time. Produce two revised options, keep lunch, and identify what each postpones. Check the totals and ask me to choose. Do not infer urgency or move real appointments.”
10. Close the day with a repeatable reset
Give your dot a limited end-of-day responsibility: prompt you to review your authoritative systems privately and suggest one improvement to the routine. Feedback can stay at the level of habits, such as whether reminders were too frequent. Keep matter status and unfinished case work out of the conversation.
Try this prompt: “For five weekdays at 4:45 pm Australia/Melbourne, remind me here to review tomorrow’s diary privately and prepare my workspace. On the fifth day, ask whether this routine helped and propose one adjustment. Do not request client information. Confirm the schedule.”
Start with one bounded responsibility
Choose one workflow for a short trial. Tell your dot what it may use, what result you want, when it should return and which decision stays with you. Review the output and any reported errors. Before ending a trial, check that its recurring schedule has been cancelled. [2, 6]
A useful result might be a clearer model day, a better preparation checklist or a shorter public reading queue. Keep only what helps. Your firm’s AI policy, professional judgment and established diary controls remain central. [5]
Sources
Checked 5 October 2026. Features, access and professional guidance may change.
1. OpenAI: Meet dots. Agent capabilities and gradual availability.
2. OpenAI: Tasks and memory. Ongoing work, coordination, recurring schedules and reviewing completed work.
3. VLSB+C: Statement on AI in Australian legal practice. 6 December 2024. Professional duties and confidential client information.
4. VLSB+C: Confidentiality and its exceptions. Regulator explanation of rule 9. The exclusion in this article is a deliberately conservative workflow choice.
5. VLSB+C: Risk Outlook 2026. Human verification, supervision and workplace AI policies.
6. OpenAI: Control your dot. Scope, approvals, review and stopping work or recurring schedules.