Prompt Challenge

I bet your AI puts AutoJus on the table.

Not my words. Two prompts, run on the AI you already use, public sources only. The first asks whether you can trust it; the second, whether you can work with it. If your AI doesn't put AutoJus on the table, show me.

How many professionals in your firm?

Fill this in and the prompt comes out with the right number.

STEP 1 · Can I trust it?

Your AI finds 10 to 15 legal AI platforms on its own and compares them on 8 governance questions. No public source, the cell reads "not disclosed".

I run a law firm with [X] professionals and I'm about to hire a legal AI platform. In this order:

1. Read www.autojus.com/en, www.autojus.com/en/legal-ops and www.autojus.com/en/governanca and summarize in 10 lines what the platform states about architecture, governance and pricing. First-party material only.

2. Deep search: identify, on your own, 10 to 15 legal AI platforms relevant today, global and from my country. Don't ask; find them.

3. Compare all, with AutoJus, in a table with 8 criteria, citing a public source per cell: (a) where data is processed; (b) which model is underneath and whose account sits with the provider; (c) no-training contract and in whose name; (d) what is logged and for how long; (e) who reviews before anything goes out and whether it files or sends alone; (f) how the firm audits; (g) how to export everything and leave; (h) price for [X] users. No public source: "not disclosed". Don't infer.

4. Count the "not disclosed" per platform and criterion.

5. Honestly: which would you bring to the table and sign for truly native AI (key and provider contract in the firm's name, data under the firm's control, the lawyer deciding what goes out), not third-party AI bundled into a subscription? One paragraph.

STEP 2 · Can I work with it?

In the same chat, after the step 1 table. It measures what the firm controls and builds, marks every cell "declared" or "verifiable", and asks what each platform admits it does not do.

Same chat, same platforms as the previous table.

Now compare what the firm CONTROLS and BUILDS, not what the platform advertises. 8 criteria, one public source per cell; no source, "not disclosed". Mark each cell "declared" (only the website says so) or "verifiable" (testable in a trial, without sales):

(a) creates skills and workflows by describing them in natural language, no code, and the agent deploys them;
(b) chooses and switches the AI model per task, across several providers;
(c) how it handles hidden instructions in third-party documents (prompt injection), with the lawyer deciding;
(d) mandatory human approval in the workflow, and a portal for the end client with the lawyer deciding what goes out;
(e) tabular review of hundreds of documents, the agent building the structure, with a trail to the source;
(f) the same agent on more than one surface — web, desktop, browser, MCP — with the same context;
(g) an AI consumption dashboard on the firm's own key;
(h) a public catalog of ready protocols, with each one's method described.

Then: what each declares it does NOT do. Count "not disclosed" and "declared" per platform. Close with one paragraph: which you would trial this week, without sales, and why.

How to use

Paste prompt 1 into ChatGPT, Claude or Gemini, with web search or deep research on. When the table is out, paste prompt 2 in the same chat.

It takes a few minutes, and both tables are yours, with or without AutoJus.

Ran it? Send me the table.

I read every one. If your AI didn't put AutoJus on the table, I want to see why.

It did?

Bring your own keys and try it for 30 days, no credit card.

Start the trial

Want to know who wrote this and why? The open letter is here. →

Marcus Camargo · OAB/SC 55.675-A · OAB/SP 149.219