The "AI Last" Principle

This essay sits primarily under the Should we do AI? question of the practice’s methodology — the discipline of solving with conventional means first, and reaching for AI only where it adds value the firm cannot generate itself.

The core idea

Before reaching for an AI tool, solve as much of the problem as you can with conventional means. Use your own expertise, the tools your firm already pays for, and your established workflows. Only hand off to AI the parts you genuinely cannot solve yourself.

This is not anti-AI. It is pro-efficiency. AI is powerful, but it is also expensive in subscription dollars, exposure dollars, and reviewer minutes — and often overkill for tasks that a competent staff member can finish in seconds with what is already on their desk.

This principle is the practical counterpart to the AI Readiness Diagnostic conversation many firms quietly need: do we actually need to do anything yet? Often the answer is “fix the workflow first, then ask the AI question.”

Why this works

Your team already knows more than they think. Most tasks in a regulated practice have well-understood solutions. Document templates, precedent banks, checklists, your practice management workflows, your accounting software’s automation, your clinical software’s letter generators — these do not need a large language model. They need a practitioner using the tools the firm already pays for.

AI costs compound. Every prompt run, every page reviewed by a human afterwards, every supervision hour is real cost. Subscription seats stack up. So does insurer attention. The fewer matters that need AI in the loop, the more concentrated your AI effort can be on the work where it actually pays.

Smaller, focused prompts produce better results. When you have already narrowed the problem to the specific hard part, the model has less ambiguity to deal with. You get more precise, useful output because you have done the cognitive work of framing the question well — which is the work that protects you anyway.

Your team stays sharp. Outsourcing every decision to AI erodes problem-solving ability over time. Doing the groundwork yourself — or having juniors do it — keeps the firm’s underlying capability current. In a regulated practice, the underlying capability is the product.

There is also a quieter benefit. Reaching for AI last keeps the practitioner in the position of deciding — which parts of the problem are theirs, which are the model’s, and where the line sits. That decision is agency. Reaching for AI first cedes it.

The practical workflow

1. Start with what you know. Draft the letter from precedent. Build the schedule from your template. Structure the file from the engagement letter. Get as far as you can on your own.
2. Use conventional tools for conventional problems. Practice management workflows, document automation, precedent banks, calculation engines, your accounting platform’s reports, your clinical letter templates. These are fast, predictable, and already paid for.
3. Identify the genuine sticking point. What are you actually stuck on? An unfamiliar area of law? A complex transaction you have not seen before? An angle on a clinical case you want stress-tested? Three competing tax positions you want to think through?
4. Now bring in AI, scoped tightly. Give it the specific problem, the relevant context, and a clear question. Do not dump the whole file in and say “draft the advice.” That is the prompt that produces the answer you cannot defend.
5. Validate and integrate the output yourself. AI gives you a draft, not a finished product. You still own the result. Your name goes on the file.

Where AI earns its keep

The principle is not “never use AI.” It is “use AI where it adds value you cannot easily generate yourself.” That typically means:

• Exploring unfamiliar territory where the firm needs to get up to speed quickly
• Generating boilerplate for matters or formats you rarely deal with
• Stress-testing arguments and positions
• Brainstorming approaches before committing to one
• Translating technical content into plain language for clients
• Processing or summarising volumes of material that would take hours by hand
• Drafting first-pass communications you will substantially rewrite anyway

The distinction is between using AI as a crutch for laziness and using it as a lever for capability. The first wastes money and dulls the firm’s skills. The second multiplies effectiveness.

The cost argument

Even on flat-rate subscriptions, the principle holds. The real costs are not API tokens. They are:

• Partner review time on a draft no one needed
• Insurance exposure when an AI-generated draft slips through unchecked
• Junior staff who never built the underlying capability because the AI did the work that would have taught them
• Leadership attention spent on AI vendor evaluations rather than on the practice

The most cost-effective AI pattern is simple: minimise what you send to AI, and maximise the value of what comes back.

About this work

borck.consulting helps regulated Australian small firms decide whether — and where — AI fits in the practice before deciding what AI to buy. Engagements begin with the AI Readiness Diagnostic: a paid two-hour conversation with leadership and a written recommendation, often answering the partner’s quiet question with “not yet, and here’s why.”