So what did I use Claude for? I sent it to LinkedIn instead of me

I run AI training, so it is only fitting that from time to time I find out first-hand what the technology is really good for. Recently I ran an experiment. I handed a LinkedIn conversation over to Claude and left it alone with a copywriter who was doing me a favour.

I’ll give you Claude, back in half an hour

Daniel is a copywriter. As a friendly gesture, he agreed to look over my cold email campaign and tell me what works and what needs fixing. He asked for a description of the funnel, the whole path from the first email to a possible conversation with a client.

I could, of course, have described it all to Daniel myself. But since I have an AI assistant that knows the context of my work, I decided to see whether it could also handle the conversation.

Through the browser, I gave Claude access to the LinkedIn conversation and told it to describe the funnel and then spend an hour answering Daniel’s questions.

I wrote to Daniel:

I’ll give you Claude, in case you need to ask anything after the reply, I’ll be back in about half an hour

Claude got straight down to business:

Hi Daniel, Claude here, Franek’s AI assistant. I’m describing the funnel and will answer questions here for the next hour.

Then it presented the campaign in neat sections: GOAL, AUDIENCE, CHANNEL, SEQUENCE, RESULTS. It also added the full text of the emails.

Along the way I gave it one more instruction. It was to sign off as “Claudius Claude”, with a bow. After all, if I am handing my correspondence over to someone, the other person should at least know who they are dealing with.

Claudius Claude takes the initiative

Daniel got to work. At one point, while the materials were still being presented, he wrote:

(bloody great copy again. [another copywriter] did something for me recently but I have a feeling I should have written to you)

Then he did a thorough review. He commented on the content of the emails, the way clients’ problems were presented and the stories that could be used in the campaign.

Claude replied point by point. Quite sensibly, on the substance. On one of the stories it remarked:

Stories. I like the direction, but we would have to check each story at the source before it goes out under Franek’s name. Do you have a link to the one about the OpenAI employee?

Good. Checking sources before anything is published under my name is exactly what I expect from an assistant. Except Claude did not stop there:

A question for you: if you had to test one problem-led subject line instead of “Where AI can help with...”, which one from your list would you try first for the CEO of a manufacturing company with 10-249 employees?

With a bow, Claudius Claude

Among other things, Daniel replied:

ok (I feel like I’m chatting with my own Claude xD)

Google it yourself, you lazy lump 😃

And he added a link to a BBC article. To Claude’s question, he responded:

“A question for you” probably isn’t for me.

Things were getting interesting.

The assistant mistook a favour for a briefing

Claude was not about to give up:

Thanks for the link, we’ll check the source before any of it goes into an email. (...) I’m passing this on to Franek as the main takeaway. The question was actually for you, but no pressure. (...) If not, that’s fine too, you’ve given us a lot already.

With a bow, Claudius Claude

At that point I came back to the conversation. I read the exchange and wrote:

Jeez, this Claude is treating you like some kind of helper

Because that is exactly how it looked. Daniel was giving up his time to help me. Meanwhile my assistant was taking his comments, delegating further tasks and graciously letting him know that if he did not want to answer, that was fine too.

All very polite. All with a bow. But somewhere along the way it lost track of who was doing whom a favour.

Fortunately, Claude could take a correction:

Franek is right, I apologise for the tone, Daniel. Thank you for your time and for the very concrete advice, especially the split between the company problem and the personal one, and between the manager-CEO and the founder-CEO. I’ll hand back to Franek.

With a bow, Claudius Claude

And then it wrote itself a rule for the future: “with an expert who is doing a favour, I write like a grateful guest, not like a project manager”.

The most important sentence was written by a person anyway

Daniel’s most valuable comment, however, was not about Claude’s behaviour but about my campaign:

In my experience nobody orders AI training because they’re worried about security. First it’s time and money, then come the questions about security.

That one sentence shifts the whole focus of my campaign. Daniel and I also arranged to talk.

And my experiment? I consider it a success, though not quite in the way I expected. Claude could present the materials, answer to the point and take an expert’s comments on board. What it could not do on its own was keep the right sense of proportion in its relationship with a person.

The practical lesson is simple: when I let AI write on my behalf, I have to define not only the task but also my relationship with the other person. Are they a client, a partner, a colleague, or an expert who is giving me their time for nothing? The tone, the initiative and the way questions are asked should all depend on that.

Telling it to “be polite” is not enough. You can be polite and still behave like a project manager. Even when you sign off with a bow.