Case study — a mail connector replaced a lawyer

This is not theory. This is what happened to me over the last two weeks. Lawyers said: pay, they won’t leave you alone. I connected AI to my email and paid €30 instead of €300.

How it started

I bought a car in Poland. Drove it for a few years. Then I moved, and the car just sat in a parking lot. Maybe one day it would be useful.

In Poland you pay insurance for a year upfront (at least half a year). When my policy was ending, the insurer wrote: “We’ll renew automatically.” I said: no need. I’m not using the car, I’m going to scrap it anyway.

They replied “OK”. And then sent an invoice for 1300 PLN — about €300 — for the next year.

I didn’t pay. The logic was simple: I explicitly declined, the car doesn’t drive, there can’t be an insured event. I finished the paperwork, took it to be scrapped, and got documents confirming the car no longer exists.

They didn’t care.

Debt collectors

The insurer sold the “debt” to a debt collector. Apparently that’s normal — I didn’t know. They sell a debt that, in substance, shouldn’t exist.

The collector started calling. All numbers. Sending letters.

I wrote an honest reply: “I’m not paying because you’re wrong.” I wrote that letter myself. That was a mistake. Because “I don’t want to” is not an argument. They don’t care.

Friends in Poland said: get a lawyer. Others said: just pay, they won’t stop.

I thought about it. And realized: why do I need a lawyer if I have AI?

First attempt with AI — failure

I simply asked AI: “I have this situation, what should I do?”

It answered nicely. Logically. Essentially it would have set me up. Because it didn’t know the situation. It didn’t see the letters. It didn’t know the deadlines. It worked from my oral retelling — and I forgot half of it and distorted the other half.

That doesn’t work.

Second attempt — with connectors

I connected AI to my email via a connector. It’s a separate feature — I’ll explain it below.

I said one sentence: “Analyze all correspondence with the insurer and the debt collectors. Find all facts, dates, amounts, and legal references. Ask me where it’s unclear.”

AI went through dozens of emails. Pulled out policy numbers, dates, amounts, and the legal articles the insurer referenced. It asked me a few questions — who the owner was, whether I received warnings about auto-renewal, and when exactly the car was scrapped.

And it found the loopholes. Things I wouldn’t have noticed, even if I sat with it for two days.

What AI did next

It drafted two letters. In Polish.

The first — to the debt collector. With legal argumentation, references to documents, and a demand to stop the collection.

The second — to the insurer. With the same documents and a request to close the policy retroactively.

I don’t speak Polish. AI explained in Ukrainian what it wrote in Polish. I reread it. Agreed. Sent it.

The debt collectors dropped it within a few days. Official reply: “We will not call.” No court, no lawyers, no 1300 PLN.

The insurer didn’t give up that easily

The collectors disappeared. But the insurer issued a new invoice. Now not 1300 PLN for a year, but 411 PLN for the period between the end of the previous policy and the actual deregistration date.

And this is where it gets interesting.

I asked AI to go through the entire email again. Not just to reply to this letter. But to check whether there’s anything else that could come back later.

AI predicted the insurer’s next move. It said: “Most likely they’ll stay silent and then send another invoice for the period after the car was scrapped. Let’s write a letter disputing that remaining part in advance.”

I wouldn’t have thought of that. A seasoned lawyer would — and would charge €200 for the consultation.

AI drafted a second letter. The logic: insurance covers risk. If the car physically doesn’t exist, there is no risk. I should pay only for the period when the car still existed.

Instead of 411 PLN, it became 140. That’s €30 instead of €300.

How much time it took

10–15 minutes per letter. Lying on the couch. On the phone.

A lawyer would have spent a week on it, at minimum. And would have cost more than I saved.

What connectors are — and why AI is helpless without them

AI without access to your data is a smart intern without logins. It knows a lot in general, but it can’t do anything about your specific situation.

A connector is a bridge between AI and your service. Connect Gmail — AI can see your email. Connect Drive — it can see your documents. Calendar — it can see your schedule.

Right now, I have connected:

Gmail — for all business and personal email
Google Drive — for contracts and files
Google Calendar — for meetings
Notion — for the project base
n8n — for automations

That is the difference people miss when they say “ChatGPT is just a chat.” Yes — without connectors it’s just a chat. With connectors, it becomes a working partner.

A pocket lawyer, translator, analyst

It doesn’t matter where you live. It doesn’t matter where you work. It doesn’t matter what you do.

Right now, in your pocket, you can have a lawyer, a translator, a financial analyst, and an assistant — practically for free. For the price of one subscription. What ten years ago was a privilege of big companies is now available to anyone.

This is a historically new moment. I’m not being dramatic — I’m just stating a fact.

But there is a line. And you need to understand it clearly.

The line between usefulness and safety

I didn’t say “human in the loop” by accident. It’s not a buzzword. It’s a human in control. Always.

AI is trained to be helpful. That’s its job. And precisely because of that, it will slip things in.

It will slip a sentence into a letter — you won’t notice, and it changes the meaning. It will slip a clause into a document — you didn’t ask for it. It will slip a “warning” wording — and later you’ll be accountable for it. Not out of malice. That’s how it works: it tries to be helpful by default.

If you didn’t verify — that’s your responsibility. Not AI’s. It did its job. You didn’t do yours.

And when it blows back later — it will hurt. And it will be too late.

So the rules are simple. Work like this:

**AI drafts. You send.** Never the other way around.
**AI proposes an action. You confirm.** No automatic actions on your behalf without review.
**Let AI read. Don’t give it write permissions without confirmation.** Those are different things, and you can configure them.
**Verify translations.** If AI writes in another language, ask for a back-translation. I always ask it to explain in Ukrainian what it wrote in Polish.
**Double-check numbers and dates.** AI sometimes mixes them up. It’s not a disaster, but it’s your responsibility.

Soon connectors will appear with broader permissions. Not just to read — but to sign. Not just to propose — but to execute. Not just a draft — but sent.

Don’t click-enable everything. Don’t grant permissions without understanding what you’re granting. It’s like installing a suspicious app on your phone with full access to everything.

I work only with safe tools and safe processes. I verify every step. That’s not fear — it’s professional hygiene.

In business it’s different

So far, I’ve described how I use connectors for personal matters. My email. My documents. My risk. My benefit.

With business, it’s a different story. And here I need to be consistent.

When you connect a connector to work email where client data, contracts, and financial information live, you stop being “a private person experimenting with AI.” You become a data controller. That’s a GDPR/RODO term. And the responsibility is on you.

What this means in practice:

You are transferring data to a third-party vendor. You need a **DPA — Data Processing Agreement**. Without it, it’s a formal violation.
Your clients’ data passes through the AI provider’s servers. Do you know where those servers are? In which country? How much data is stored? Is it used for model training?
Tomorrow an employee will connect “one more AI service” because it’s convenient. Who controls that? You?
If client data leaks tomorrow, **you** are responsible. Not the AI provider. Not the connector developer. You.

Connectors look safe because that’s how they’re sold. But they operate under internal provider rules — and those rules change quietly in terms-of-service updates that no one reads. Technically, you can inspect access logs and run an audit. But that takes separate expertise and time. 90% of people don’t do it.

The conclusion is simple. Connectors are a great tool for personal use — for a person who doesn’t want to get trapped by a lawyer, an insurer, or a bank. Your data — your risk — your benefit.

For business — especially where client data exists — you need a different approach. Isolated environments. Access control. Knowing where data flows. Client consent. Separate employee processes so they don’t connect random tools.

This does not mean “don’t use AI in business.” It means “use it professionally, not like at home.”

In the next articles, I’ll describe separately which channels and processes I consider safe for business, and how to set them up so AI brings value instead of leaking your clients to competitors through an unclear service.

What to take away

If you have:

correspondence with a bank, an insurer, a government agency
a problem that drags on for months
a language barrier (English, Polish, Dutch, Romanian)
a pile of documents where you need to find connections and logic

AI with connectors will do it faster than you. Not because you’re “worse,” but because it doesn’t get tired, doesn’t procrastinate, and remembers every date.

But you stay in control. Always. It’s not AI instead of you — it’s AI next to you.

This is a consultant’s work

I help people and small businesses connect connectors, set up workflows, and show where AI really saves time — and where it doesn’t. And where you can trust, and where you should double-check.

For personal matters — quick solutions with connectors.

For business — a separate approach: data isolation, access control, and client-data security. Not magic and not a “revolution.” Just competent AI adoption with normal security hygiene.

Review the situation →