Lawyers, analysts, salespeople, writers, developers: each has tools built for the job. Three questions tell you whether you need one, and a map of the kinds that exist.
By now you've probably noticed the ads. AI for lawyers. AI for recruiters. AI for accountants, teachers, estate agents, dentists. Some are real products with real advantages. Many are a general assistant in a costume, with a logo and a higher price. This lesson is about telling them apart, and about the honest answer to "do I need one?", which for most people, most of the time, is no.
Three questions
Ask these before you open a comparison article. If all three are no, close the article.
Does it need my own data, live, every time? If your task is "answer from our CRM", "draft from this client's file", "check against our price list", a chat where you paste things gets old fast. You want a tool that's connected to the data, or a connector that brings the data to your assistant. Lesson 8 covers the second option, which is often cheaper than the first.
Is there a rule, format or compliance someone will check? Legal drafting, financial reporting, medical notes, regulated marketing. Specialised tools here encode the rules, keep audit trails, and are set up by people who know what the regulator wants. A general assistant can draft, but it doesn't know your jurisdiction's quirks unless you paste them, and nobody is accountable for what it missed.
Do I do this every single day? A daily task deserves a tool that lives where the work is: inside your email, your spreadsheet, your code editor, your meeting software. Copy-pasting into a chat is fine weekly, tiresome daily.
If all three are no, your general assistant is enough, and the time you'd spend evaluating tools is better spent on lesson 6.
The map
Here's what the categories look like, with the kind of product rather than a shopping list, because the names change and the kinds don't.
If your work is…
The specialised kind
What it adds over a chat
Legal
Contract review and drafting platforms, research tools tied to case-law databases
Citations you can open, clause libraries, audit trail, confidentiality terms written for law firms
Finance and analysis
Spreadsheet-native assistants, data-analysis modes in the big assistants, BI tools with a chat layer
Works on the actual numbers, shows the formula or code, charts from live data
Sales and customer work
CRM-embedded assistants, call recorders that summarise and log
Reads the account history, writes into the CRM, no copy-paste
Writing and marketing
Brand-voice tools, SEO-aware writers, design tools with generation built in
Enforces house style, checks against brand rules, templates per channel
Meetings
Note-takers inside Teams, Zoom and Meet, or standalone recorders
Transcript, summary, action items, searchable later
Software
Code editors with AI built in, agents that work in a codebase
Sees the whole project, runs tests, makes changes you can review
Translation
Dedicated translation services, plus the big assistants
Terminology memory, document formatting preserved
Research
Assistants with deep-research modes, academic search tools
Many sources, cited, in one report; still check the citations
Two things to notice. The big assistants keep absorbing these categories: data analysis, research, meeting notes and coding are all now modes inside ChatGPT, Claude or Gemini as well as separate products. And the specialised tool's real advantage is rarely a smarter model. It's the plumbing: connection to your data, the rules of your field, and living where you already work.
How to evaluate one without wasting a month
If a question above came up "yes", test a tool like this:
Pick one real task you did last week. Not a demo task. Yours.
Do it in the specialised tool and in your general assistant. Same input.
Compare on three things: was the result better, did it save real minutes, and did it get at the data without you pasting.
Check the boring parts: where your data goes, who can see it, what happens when you cancel. For anything regulated, this is the whole decision.
Decide in a week. If you're still "evaluating" after a month, the answer was no.
Ask for a trial. Most will give one. Don't buy annual on day one.
The costume problem
How to spot a general assistant in a costume: it can't connect to anything of yours, it has no feature a prompt couldn't replace, and its demo works on its sample data but it has no answer for "how do I get my documents in?" These aren't always bad; some add a good template and a nice interface. But you're paying for packaging, and the same result is one well-written request away in the tool you already have. Lesson 7 shows how to make that request once and reuse it.
The habit
Before any tool purchase, write the task it's for in one sentence and answer the three questions. If the honest answer is "I want it because the ad was good", that's a bookmark, not a tool. You met that item on the ignore list on the school's front page.
Try this today
Write down the three tasks that take most of your week. For each, answer the three questions. You'll probably find one that deserves a specialised tool or a connector, and two that your assistant already handles if you brief it properly. That's your shopping list, and it's short.
Check yourself
5 questions. Everything is in the lesson above. Get them all right and the lesson counts as done.
1. Your task is "write a weekly status email from my notes". Do you need a specialised tool?
All three questions come up "no". That's a general-assistant task with a reusable brief, which is lesson 7.
2. A contract-review platform costs much more than a chat subscription. What's the real thing you're paying for?
Specialised tools rarely win on the model. They win on connection to data, encoded rules of the field, and accountability.
3. Which sign most suggests an "assistant in a costume"?
Being built on a big model is normal. Having nothing but packaging is the tell.
4. What's the right way to test a specialised tool?
Demos work on demo data. Your task, side by side, decides in a week.
5. Your work needs the CRM's account history every time. Which direction is usually cheapest to try first?
Connectors (lesson 8) often solve the data problem without a new subscription. Pasting daily doesn't last; buying first skips the test.