Recently, my dear friend asked me if I use ChatGPT to “professionalize” my work messages. Of course not, I said, I use Claude.

It hadn’t even occurred to me to consider that. Why on earth would I regress my communication to the mean?

Then I thought deeper about the topic and I have something to say. It’s good that I have a newsletter 😅.

AI is useful-to-great at plenty things: web research, agentic coding, curing cancer. But it isn’t (and I’d argue shouldn’t) be useful for communication.

Replace non-communication with UIs and APIs

Not everything that currently requires two humans talking to each other should. If an interaction is transactional and goal-oriented, why would you need a beautiful and complex person at the other end?

Let’s take my favourite example: booking a table at a restaurant (I know, I know). I can delegate this to my AI assistant — problem solved, right? I imagine this timeline:

1. I outsource booking to my LLM.

2. (my LLM is talking to a human being, that’s pretty unethical to me)

3. The restaurant outsources reservations to their LLM.

4. (two robots are communicating in plain lossy English, why?)

To me, the real solution is this:

Instead of band-aiding with LLMs, let’s better burn these tokens to implement UIs for humans and APIs for robots instead. Building and customizing software has never been cheaper, so there’s no excuse.

“What’s the status of my residence permit application?” → UI

“I’d like to request a refund” → UI

“Sign me up for anger management classes” → motherfucking UI!!!

Make communication even more human

Remote work is already taxing on revealing humans behind Slack (and MS Teams¹) messages, and it’s a crime to make it worse.

Now is the best time to find and evolve your voice and your particular way of putting thoughts into letters. Ask AI for advice, sure, but never allow it to type your messages for you:

I’d take your awkward email over the “professionalized” version every time.

Does your job require you to speak in corporate? Fuck that job! We are hiring. Trust me, you can negotiate enterprise deals with emojis and put the words “quite a depressing month“ in an investor report.

I hope that the current rise of LLMs will expose how deeply inhuman some of our corporate and formal communication is, and will help us appreciate and embrace our human weirdness.

Shrink the middle

So I argue against using LLMs both for “non-communication” and for real communication. But is there something in between?

I can only think of a single scenario where I’m fine with using AI for communication. It’s connecting people who don’t speak each other’s languages — think of asking an Airbnb host to borrow a tennis ball for your dog. LLM translations are lossy but the alternatives are worse.

Otherwise, let’s delegate 100% to the robots or stay 100% human.

⁰It would be interesting to revisit this essay in a couple of years.

¹My condolences.

This was issue #17, find more at antoniokov.com.

What are the valid scenarios for using AI for communication, in your opinion? Reply to this email and share your thoughts.