DEV Community

AISoloDesk
AISoloDesk

Posted on

Two professions, one AI rule: automate the admin, never the judgement

Two professions, one AI rule: automate the admin, never the judgement

Two of the guides I published this week are about professions most "best AI tools" lists never touch properly: accountants and solo lawyers. Not because the tools are different — the tools overlap a lot — but because the question is different. For a freelancer, the question is "what saves me time?" For an accountant or a lawyer, the question is "what am I allowed to let a machine touch?"

That distinction matters more than any feature list, and it is the reason so many professionals either buy the wrong tool or buy nothing at all.

The work AI is good at in a practice

Start with what is boring, repetitive, and low-risk. For an accountant, that is chasing receipts, reconciling transactions, reminding clients that the documents are still missing, and drafting the first pass of a routine email. For a solo lawyer, it is the intake form, the appointment booking, the reminder before a consultation, and the first draft of a standard letter.

None of that is the professional judgement. It is the admin wrapped around the judgement. And it is where the hours go: the average small practice loses more billable time to chasing and reminding than to anything a client would ever pay to watch.

The guides I wrote this week price that admin in minutes, because "saves time" is too vague to buy against. An accountant who knows a tool returns forty minutes a week can decide whether it is worth $20 a month. A lawyer who knows the intake chatbot answers the "do you take new clients?" question at 11pm can decide whether that is worth the setup afternoon. Minutes are the only honest unit here.

The line AI must not cross

Here is where the two professions agree, and where most AI marketing goes quiet: some work never leaves human hands, no matter how good the tool gets.

For an accountant, that line sits at anything that changes what a client owes, claims, or files. AI can prepare, organise, and flag. It does not sign off. For a solo lawyer, the line sits even earlier: anything that touches client confidences, anything that gives legal advice, and anything that decides strategy. AI can draft the letter. It cannot decide what the letter should say.

This is not caution for its own sake. A wrong number on a return and a wrong sentence in a client file are both mistakes that cost more than any subscription ever saved. The professionals who get the most out of AI are not the ones who automate the most. They are the ones who are clearest about what stays human.

How to buy against that line

If you run a practice — accounting, legal, or anything where a client trusts you with something sensitive — buy tools in this order:

  1. The chaser. Whatever reminds clients, collects documents, and follows up without you thinking about it. This is the highest-return automation in any practice, and the lowest-risk one.
  2. The drafter. A tool that writes the first version of routine documents and emails, which you then edit. First drafts are cheap; final judgement is yours.
  3. The intake. Something that answers the basic questions and books the appointment while you are with another client. It does not advise. It routes.

What not to buy first: anything that promises to replace the judgement itself. If the sales page says the tool "decides", "advises", or "signs off", that is the moment to close the tab.

The two guides

I wrote up both professions in full this week — the tools, the pricing, and exactly where each one draws its line:

Read the one that matches your practice. Then read the other one anyway — the line is drawn in a different place, and seeing both makes your own line clearer.

The rule underneath both

Automate the admin, never the judgement. If a tool cannot name the specific hour of your week it gives back, don't pay for it yet. And if a tool asks you to hand over the decision itself, the price is not the subscription — it is the mistake you will not catch until a client does.

Top comments (1)

Collapse
 
ztrx_febb1a49041b23a5125a profile image
ztrx •

The line I would draw is not draft versus send, it is who presses the button. Drafting a chase email is admin. Deciding that this particular client needs chasing today, given everything that happened last week, is judgement, and it depends on context no model has. So the pattern that has held up for me: let the tool find who is overdue and write the first pass, keep the send manual, and make it obvious why each name is on the list. If the user cannot see the reason, they do not trust it and go back to checking the inbox by hand.