I wake up at 7am on a Friday. The date is April 20, 2029. My dog (Trim) stretches by my bed, moving his tail frenetically.

“Morning, Mark.”, I say to my LLM.
“Good morning, Steve,” Mark replies. “wanna start the day clearing off some pending tasks from the night?”
“Sure.”

Three seconds of silence, and then Mark the LLM says:

“It’s two payments. The plumber’s LLM sent an invoice for the Tuesday sink fix. Price is what we had agreed. I guess that’s a Yes?” Mark sounds a little bored.
“Yeah.”, I reply, patting Trim’s head, who yawns.
“Next: your sister wants you to chime in for her friend’s fundraiser. Something about property rights for LLMs.” his voice sounds amused this time.
“Gah, no!”

I get up from bed and walk by my server microrack, picking up some clothes to put on. Lights twinkle and the GPUs hum softly. Everything’s good.

Inside that 10x10x10 cm³ cube, a dozen or so LLMs are running a multitude of tasks for me. Some are responding to real-world events. Some are browsing the web. Some are processing my email, private notes, and databases. Some are talking to other people’s LLMs. Some are requesting new tasks from other agents.

I only own 8 GPUs, so most of my batch tasks (news analysis, etc.) do time-sharing and get scheduled according to priority. Of course, the scheduler is an LLM too. It gets a list of task descriptions and picks the next one to run every 200ms.

Right now, it assigns high priority to analyzing a social media post. It was just published by a close friend. Another LLM, fine-tuned with my social network and interactions spins up. The full tweet is sent to it, along with a prompt engineered to analyze social posts. The tweet contains a link, so the contents of the link are appended too.

The LLM reads through everything. It decides some ideas might be new and interesting to me, so it drafts a two sentence summary. After a few revisions, it sends it to my message queue, and terminates. All of this happens in roughly 92ms, while I brush my teeth.

I look at myself in the mirror and say: “Can I get breakfast with Freddie?”.

“Negotiating…” replies Mark in a quiet, professional voice. Some seconds later, “got it, his LLM agreed you can meet at Beppu Café in 15.”
“Cool, thanks.”

On my way there, I listen to a podcast about the latest open-source LLM agents. It’s amazing what people are coming up with these days. Meanwhile, someone mentions my name online. A public relations LLM spins up, reads the blog post, and decides it’s about another person with the same name. It goes back to sleep, and I don’t get notified.

I arrive at the café early, so I decide to check out one of the new agents I just heard about. It’s called MoneyLLM, and it’s supposed to help you with financial decisions. I download the model’s weights to my home server, and a couple of LLMs spin up. First, a code auditor agent looks at the open-source code for about 10 seconds and gives a preliminary green light. No obvious backdoors, data stealing, or malicious tracking. Good start.

I move forward, knowing that a more in-depth analysis will be running for the next hours. Now, additional agents check the code’s performance and the creator’s online presence. Everything looks good, so I run MoneyLLM for the first time. After getting access to my local financial data, it sends a short message:

“Hi Steve! I recommend you move your DAI from Aave to Compound. It seems the yield is better there now, and you’ve used Compound before. Do you want to move forward?”

“Yeah, sounds good. Mark, can you take care of that?”

Mark chimes in: “Yep, I’ve sent the instruction to your crypto agent to craft and sign the transaction… Done.”

I look around and my friend hasn’t arrived yet. I order a coffee to drink while I wait for him.

I receive a high priority notification: “Fine-tuning task complete: MarthaBot”. I get a rush of dopamine: I’ve been waiting for this for days! Martha was my grandmother and life mentor. I lost her two years ago. But there’s this open-source code that lets you train an LLM with conversations you had with someone. Similar to that Black Mirror episode, but without the creepy robot.

I see my friend approaching Beppu Café, so there’s no time to chat with her just yet, but I don’t want to make her wait. I spawn MarthaBot and assign her plenty of resources. I also add 5 helper LLMs for her to command. I send her the message: “Hi grandma-bot! Let’s talk in a couple of hours. Feel free to look around my stuff meanwhile.”

I smile at my friend as he approaches my table.

“Crazy days, huh?” says Freddy.

“Crazy days…” I reply.

Final words

If you liked what you read, consider subscribing:

Comments and Discussion

Email a private comment