Hermes, the agent that kept me company without asking

A week without sitting down at the desk the way I used to

The first difference I noticed was geographic. Before, my work as a developer was tied to the chair, to the big monitor, to the operating system I knew well. Anything that broke out of that — moving a file, editing an image, running a script on another machine — required me to interrupt what I was doing.

With Hermes, that blurred. I write to it from my phone in a line at the bank. I write to it from bed at eleven at night when an idea shows up. I write to it from the Mac and also from a terminal on a server I did not even set up port forwarding for. The same conversation picks up where I left it, because it remembers what we have done together.

That is not innovation. It is a concrete convenience. But when you work alone, a convenience that shows up every day ends up changing how you organize yourself.

What nobody tells you about agents that "remember everything"

The pretty promise is that the agent remembers everything. The reality is more interesting and uglier at the same time. Hermes remembers well the things that matter: my configuration preferences, the directories where I keep projects, the tools I prefer for each job. It does not remember random details of a conversation from three weeks ago with the same precision, and that is fine.

What it does do, and that is the valuable part, is build a shared vocabulary. The third time I ask for something in Spanish, it already knows I prefer technical answers with context. The fifth time I mention a repository, it already has it loaded without me asking. That is not magic. It is repetition, persistent context, and a system designed to learn from its own operation.

There is an uncomfortable side, too. When an agent remembers a lot, you start wondering who owns the information. The short answer: I do. The memories live on my machine, in my configuration, in files I can read and delete. That changes the conversation. If you want to trust a tool, you need to be able to open the hood and look inside. With Hermes you can.

An ordinary afternoon

Something small happened to me the other day that sums all of it up. I was sitting in the middle of the afternoon with a silly question about a date format. Instead of opening a new tab, typing the search, reading five Stack Overflow answers, comparing two contradictory ones, and assembling my own, I asked Hermes. It handed back the answer, two examples, a clarification on the case I was dealing with, and, without me asking for it, a note on why my original approach had a small detail that could fail in a different timezone.

Three minutes. No context to assemble. No thread to lose.

Multiply that by a week and you understand why the phone is no longer just for reading messages. It is another work tool, and the Telegram integration stops being a technical curiosity and becomes a normal part of the day.

When it fails, and why I like that

Hermes gets things wrong. It is not perfect. Sometimes it calls a tool with the wrong parameters. Sometimes it reads my intent backwards. Sometimes it hands me an answer so long that I lose myself in it. But when it happens, it does not stay quiet and it does not pretend the operation was a success. I see it in the logs, in the error, in the conversation. That, oddly, gives me more confidence, not less.

An agent that never fails, or that hides its mistakes, is harder to use than one that fails transparently. With the second one you can build a workflow on top. With the first one, you never know when the thing will break without warning.

It does not replace anyone, but it changes the routine

There is a silly conversation that comes up every time someone talks about AI agents: are they going to replace programmers? No. The short answer is no. The slightly longer answer is that an agent does not replace you because it does not share the consequences of your decisions. If you ask it to migrate a database and it gets it wrong, the consequences are yours. If you ask it to write a contract with a client, the consequences are yours. If you ask it to publish something in your name, the consequences are yours.

What an agent like Hermes does, used with judgment, is move the starting point. You sit down to work and the codebase already has notes you left last week. The documentation already has a reasonable first draft. The repetitive scripts are already written. The human work is still human work, but it starts later in the line. That, for someone who works alone, is worth more than any promise of automation.

What it feels like

If you ask me what it feels like to use Hermes, I would say it feels like working with an eager intern who is curious and fast. One you still have to correct, but one you can hand a real task to and come back twenty minutes later to review.

It also feels like having a notebook that answers. I write a loose idea and it gives me back a first skeleton. I ask for a title and it gives me five options in the tone we have used before. I say I do not understand something and it explains it with the depth I need, no more and no less.

And it feels, above all, like a new layer over a way of working I already liked. It is not the tool that made me a better developer. It is the tool that let me use my time differently. You write fewer empty emails. You build fewer boards just to have something visible. You take fewer duplicate notes. The day yields a little more, and that little more, accumulated, is felt.

You do not need me to tell you this as if it were a revelation. Nor do you need to dismiss it as a fad. It is one more tool, with its limits and its virtues, and like every tool that truly sticks around, it becomes invisible when it works well.

If you are curious to try it, I installed it on my own machine following nothing more than what the official documentation says. You do not need a server. You do not need an expensive subscription. You need your laptop, an hour of free time, and the willingness to treat the agent like a colleague, not an oracle.