Conversation, Not Delegation

This essay sets out the Conversation not Delegation framework that underpins the practice’s methodology: the operating principle behind every Diagnostic, workshop, and engagement.

Two ways to use AI

There are two ways your team can use AI:

1. Delegation: “Do this for me.”
2. Conversation: “Think with me.”

The difference matters more than most principals realise, and it is the single biggest predictor of whether an AI rollout makes a small firm sharper or quietly hollows it out.

Delegation develops dependency. Your team gets output, but not understanding. They collect answers without building the capacity to evaluate them. Over time they become less capable, not more, because they have outsourced the thinking, not just the typing.

Conversation makes them capable. Understanding is built alongside the output. People learn to ask better questions, spot weaknesses in reasoning, and develop judgement about what is good enough and what needs more work.

The goal is not to get AI to do your firm’s work. It is for your firm to become more capable, with AI as a thinking partner.

This distinction is not abstract. In a regulated practice, the difference between delegation and conversation is the difference between a junior solicitor who can defend the advice on the page and one who cannot, between a clinician who has actually thought through the differential and one who has accepted a model’s confident summary, between a tax position your firm can stand behind and one that will not survive scrutiny.

What is at stake, underneath all of this, is human agency: the firm’s ability to choose, steer, and stand behind its work rather than absorb whatever the model produces. Delegation transfers agency to the tool. Conversation keeps it where it belongs: with the practitioner whose name is on the file, whose registration is on the line, and whose judgement is what the client is actually paying for. Every other distinction in this piece (capability, judgement, ownership, professional obligation) flows from that one.

The mindset shift

Most teams start with the delegation mindset. It is natural. AI is fast, and getting a quick answer feels productive. But the firms that get durable value from AI are the ones whose people treat it as a conversation partner, not an assistant.

What conversation looks like in practice

Brainstorm with AI

Do not ask AI for the answer. Ask it to help you explore the space.

“I am structuring a Statement of Advice for a self-employed client with mixed business and personal income, two dependants, and a recent inheritance. What angles am I likely missing? Push me beyond the obvious.”

Then push back:

“Those are useful, but the inheritance includes a small holiday rental that complicates the deductibility question. How does that change the framing?”

Iterate with AI

The first output is a draft, not a deliverable. The value is in the refinement.

“Here is my first attempt at this client letter. What is weak about it? What am I missing? Where would a thoughtful critic push back?”

Then:

“Good points. Let me revise based on your feedback. Here is version two. What is better and what still needs work?”

Question AI’s output

AI sounds confident even when it is wrong. The job is to evaluate, not to accept.

• “What assumptions are you making here?”
• “What evidence supports this claim?”
• “What would someone who disagrees say?”
• “Are there contexts where this advice would be wrong?”

A note on citations: in law and tax especially, never accept a citation, section reference, case name, or paragraph number from a general-purpose model without verifying it directly against the source. Models fabricate authority that looks correct. Treating that fabrication as fact is one of the fastest ways for a regulated firm to end up in front of a tribunal.

Evaluate together

Use AI to help you develop your own evaluation criteria, then apply your judgement to the result.

“Help me build a checklist for whether this advice meets our internal quality standard. Then let’s run the draft through it together.”

Process over product

Here is the deeper principle:

The team members who obsess over the process (questioning, evaluating, iterating) will always outperform the ones who just collect AI outputs.

Why? Because judgement cannot be delegated to something that has read everything and experienced nothing.

AI has patterns. Your firm has context. AI has breadth. You have stakes: a client, a regulator, a court, a board. AI can generate options. Your people decide which options matter.

The teams that treat AI as a shortcut to the product miss this entirely. They get faster output but shallower understanding. Over time they lose the very skills that make them valuable: critical thinking, contextual judgement, the ability to evaluate quality. In a regulated practice that erosion is not just a productivity issue. It is a governance issue.

A note on agency

It is worth naming the concept that sits underneath this argument explicitly, because it is the one your professional obligations and the wider AI policy environment already lean on.

Agency is the practitioner’s ability to steer: to decide what to engage with, what to push back on, what to accept, what to discard. It is the active stance that turns AI output from a finished product into raw material for professional judgement.

Iteration is not the same thing as agency. You can iterate with AI for hours and still be drifting along behind whatever the model first produced: refining its framing, accepting its emphasis, defending its conclusions because they are now in front of you. That is delegation in slow motion.

Agency looks different. It is the moment you say “actually, that is the wrong question” and reset. It is the moment you ignore the elegant draft because it does not fit the client. It is the willingness to be the source of the work, not the editor of the model’s work.

This matters in regulated practice for three concrete reasons:

• Professional obligations presume it. Australian Privacy Principles, the Legal Profession Uniform Law, AHPRA expectations, APES 110, RG 175: all of them assume a practitioner is exercising professional judgement on the work. None of them recognise “the model produced it” as a defence.
• It is the ground for accountability. When something goes wrong, the question “who decided?” has to have a person’s name attached. Delegation makes that name harder to find. Conversation keeps it clear.
• It is what the client is paying for. A client can already get a passable first draft from a chatbot for free. What they cannot get without you is someone willing to choose, steer, and stand behind the work.

Every move in this framework (pushing back on the first answer, questioning assumptions, evaluating against your own understanding) is an exercise of agency. The framework is not a productivity hack. It is a practice for keeping agency where your registration says it has to be.

What process literacy looks like

Process literacy means knowing how to work with AI effectively, not just what to ask for. In a small firm it has four parts.

1. Knowing when to use AI (and when not to)

Not every task benefits from AI. Process-literate practitioners ask:

• Is this a task where AI adds genuine value, or am I just being lazy?
• Will using AI here help me learn, or prevent me from learning?
• Do I understand this well enough to evaluate what AI gives me?

For a graduate solicitor, a junior accountant, or a new clinician, the third question matters most. If you cannot evaluate the output, you cannot use AI safely on that task yet.

2. Structuring the conversation

Effective AI conversations have shape. They are not random queries. They build on each other:

1. Set the context: give AI what it needs to be useful (without giving it what it should not see; see Why Enterprise AI Data Governance Gets It Wrong)
2. Explore the space: generate options, perspectives, and approaches
3. Evaluate and refine: challenge, improve, and iterate
4. Synthesise: pull together the best elements with your own judgement

3. Maintaining ownership

The advice has your firm’s name on it. Process-literate practitioners:

• Can explain every part of what they sign
• Can defend their decisions without referring to “the AI said…”
• Have improved the AI’s output with their own expertise and context
• Know where the AI contributed and where they did

This last point is also where your file note discipline matters. Practices that record how AI was used, not just whether it was, are in a stronger position when a complaint, audit, or insurer query arrives.

4. Reflecting on the process

After using AI, ask:

• What did I learn from this conversation?
• Where did AI help me think better?
• Where did I have to correct or improve on AI’s suggestions?
• What would I do differently next time?

The intern analogy

Think of AI as a smart, enthusiastic intern who:

• Works incredibly fast
• Has read an enormous amount
• Sounds confident even when wrong
• Lacks real-world experience
• Needs your expertise to do good work
• Gets better with clear direction and feedback

You would not hand an intern a critical matter and walk away. You would brief them clearly, review their work, ask them to explain their reasoning, point out what they missed, and have them revise.

That is a conversation, not delegation. Apply the same approach to AI, and apply it consistently, because intern-like outputs that go out the door without supervision are how firms end up in front of professional standards bodies.

Common traps

Each of the traps below is a way agency erodes. They are worth knowing by name because they show up unannounced, usually in the form of someone who feels productive while quietly handing the steering wheel over.

The speed trap

“AI gave me an answer in ten seconds. Why would I spend thirty minutes iterating?”

Because the ten-second answer is generic. The thirty-minute conversation produces something that fits the matter, has been stress-tested, and that the practitioner actually understands. Speed without understanding is just faster mediocrity, and in a regulated context, faster mediocrity is faster exposure.

The confidence trap

AI writes with authority regardless of accuracy. When the output looks polished and professional, it is tempting to trust it. But polish is not the same as correctness. Always ask: “Would I stake my registration on this?”

The replacement trap

“AI can do this better than me, so why bother learning?”

Because “better” depends on context, and context is exactly what AI lacks. AI generates plausible text. Your people generate meaning. If practitioners stop developing their own expertise, the firm loses the ability to tell the difference, and that is the difference your clients and regulators are paying you for.

The efficiency trap

“We are being more productive by delegating to AI.”

Are you? Or are you producing more drafts of lower quality, which someone senior then has to repair? Productivity is not volume. It is impact. One well-crafted advice beats ten AI-generated drafts that need to be unpicked.

Applying this in your practice

For legal practices

• Use AI to brainstorm angles on a matter, then refine with your own analysis. Never rely on AI-generated case citations or section references without checking the primary source.
• Use AI to draft a first pass of a client letter, then rewrite it for tone, accuracy, and your relationship with the client.
• Use AI to stress-test a position (“what would opposing counsel argue?”), then apply your judgement about what actually matters in this matter.
• Treat AI as a sounding board for arguments, not a source of arguments.

For medical and allied health practices

• Use AI to help structure patient education materials, then verify clinical accuracy against current guidelines.
• Use AI to draft administrative correspondence, then check it for tone and any clinical content that should not be included.
• Use AI to think through a differential or a referral letter, treating its suggestions as one more input, not as the conclusion. AHPRA expectations around clinical reasoning do not change because a model was involved.
• Never paste identifiable patient information into a tool you have not specifically approved for that purpose.

For accounting and financial advisory firms

• Use AI to explore options on a tax position or advice scenario, then apply your professional judgement and verify against current legislation.
• Use AI to translate technical content into plain language for a client letter, then verify the technical accuracy yourself.
• Use AI to check whether your reasoning has gaps, then close the gaps with your own work.
• Treat AI output as a draft prepared by a graduate, never as a position you can sign without review. APES 110 and your TPB obligations apply regardless of how the draft was produced.

Getting started

1. Next time you use AI, pause before accepting the first output. Ask: “What is weak about this? What am I missing?”
2. Try the brainstorm approach. Instead of asking AI for an answer, ask it to help you explore the question from multiple angles.
3. Practice the evaluation habit. After every AI interaction, ask yourself: “Do I understand this well enough to explain it without AI?”
4. Model the process. Partners and senior practitioners set the standard. If your team sees you iterate and challenge AI, they will too. If they see you accept first drafts, they will too.

The bottom line

AI is the most powerful thinking tool most firms have ever had access to. But like any tool, its value depends on how you use it.

Delegate, and you get output. Converse, and you keep agency, and capability follows.

The choice compounds. Every conversation builds your firm’s judgement and keeps the steering where it belongs. Every delegation quietly hands it over. Choose accordingly.

About this work

The Conversation, Not Delegation framework is now on the reading list at Kaplan Business School and is the foundation of the AI Readiness Diagnostic offered through borck.consulting: a fixed-fee, two-hour conversation with leadership and a written recommendation that answers the question many partners are quietly asking themselves: do we actually need to do anything yet?

For a deeper treatment of the methodology, including the structured six-step process and worked examples, see the book: Conversation, Not Delegation: Mastering Human-AI Development.