Alexis Rondeau

Newsfeed

Everything I publish, in the order I published it: dispatches from Twitter, videos from YouTube and notes from my vault, archived here in full — text, images, video and links — so they outlive the platforms.

857 dispatches since March 2014 · page 10 of 35

47 views

This bears repeating, in case you haven't seen this yet:

Quoting @karpathy · Mar 7, 2026

I packaged up the "autoresearch" project into a new self-contained minimal repo if people would like to play over the weekend. It's basically nanochat LLM training core stripped down to a single-GPU, one file version of ~630 lines of code, then:

  • the human iterates on the prompt (.md)
  • the AI agent iterates on the training code (.py)

The goal is to engineer your agents to make the fastest research progress indefinitely and without any of your own involvement. In the image, every dot is a complete LLM training run that lasts exactly 5 minutes. The agent works in an autonomous loop on a git feature branch and accumulates git commits to the training script as it finds better settings (of lower validation loss by the end) of the neural network architecture, the optimizer, all the hyperparameters, etc. You can imagine comparing the research progress of different prompts, different agents, etc.

t.co/YCvOwwjOzF
Part code, part sci-fi, and a pinch of psychosis :)

Image from the post
1 reply · 55 views
In reply to @beetclaw · Mar 6, 2026

@SpringStreetNYC how can i see a dashboard like this for mine?

@beetclaw Hey! I have this directly in my Anthropomorphic platform reporting under "Usage"

Note, "platform" means that I have an API account with them, in parallel to my Claude subscription.

Platform: platform.anthropic.com
Claude:

Claude PlatformClaude PlatformBuild on the Claude Platform↗ platform.anthropic.com ClaudeClaudeClaude is Anthropic's AI, built for problem solvers. Tackle complex challenges, analyze data, write code, and think through your hardest work.↗ claude.ai
20 views
In reply to @SpringStreetNYC · Mar 12, 2026

A few times now, in conversation with friends about AI, I found myself saying "...and that's why we need new vocabulary for all this"

And amazingly, people smarter than I, have started to do just that!

Here are my favorite new terms:

What's your favorite new word, term or way to describe what we're seeing?

PS: My first creation I'd love to add to the mix would be "AINXIETY" but it'd be okay if it didn't catch on.

Let's add: "@claudeai Claude Cowork could do all of that for me."

2 replies · 2 likes · 154 views · 1 bookmark

My wife after casually teaching herself @claudeai Claude Cowork since Tuesday:

"I couldn't fall asleep last night because I realized I could almost completely let Claude Cowork do my taxes this year.
All I need to do is X, Y and then Z. Then send everything to the advisory.
I could be done with the whole thing in half a day."

As someone who builds software for our family office's tax advisory all day:

She is absolutely right.

25 views
In reply to @SpringStreetNYC · Mar 6, 2026

And no, I do NOT want a "dAsHbOard wHeRe I cAn sEe tHeM aLl aT tHe saMe tiMe."

I love that they're all over the place. The mess is good.

THE MESS IS THE MESSAGE.

THE MESS IS THE MESSAGE.

Okay, I was right in regards to my personal assistant(s) but I might be WRONG for orchestrating teams.

Because, yeah, this is hot.

Image from the post
1 reply · 36 views

A few times now, in conversation with friends about AI, I found myself saying "...and that's why we need new vocabulary for all this"

And amazingly, people smarter than I, have started to do just that!

Here are my favorite new terms:

What's your favorite new word, term or way to describe what we're seeing?

PS: My first creation I'd love to add to the mix would be "AINXIETY" but it'd be okay if it didn't catch on.

GitHubGitHub - paperclipai/paperclip: The open-source app everyone uses to manage agents at workThe open-source app everyone uses to manage agents at work - paperclipai/paperclip↗ github.com
1 reply · 39 views
In reply to @SpringStreetNYC · Mar 6, 2026

Today I have multiple streams happening, with different agents/assistants, on different topics, across multiple platforms:

1. Claude Code agent + sub-agents that are refactoring, cleaning a personal project (Terminal)
2. Claude Code agent running on my nanoclaw VPS that helped me upgrade the assistant's Gmail/Calendar skill to Google's official CLI (Terminal + SSH)
3. Nano, my nanoclaw assistant who's currently scheduling a coffee chat with my friend's assistant via their own respective Gmail accounts (Telegram)
4. Horst, another friend's @openclaw assistant with whom I've developed a fun conversation around tech, debugging, cool ideas (WhatsApp)

The biggest surprise to me is that it's starting to feel natural and seamless.

And no, I do NOT want a "dAsHbOard wHeRe I cAn sEe tHeM aLl aT tHe saMe tiMe."

I love that they're all over the place. The mess is good.

THE MESS IS THE MESSAGE.

1 reply · 33 views

Even with a dev background, I REALLY dig having had Claude Code for the onboarding and customization process.

I would have not even considered using CC for that, but it's the first time that I have a full-time CTO without having to be one myself.

1 reply · 2 likes · 81 views

Today I have multiple streams happening, with different agents/assistants, on different topics, across multiple platforms:

1. Claude Code agent + sub-agents that are refactoring, cleaning a personal project (Terminal)
2. Claude Code agent running on my nanoclaw VPS that helped me upgrade the assistant's Gmail/Calendar skill to Google's official CLI (Terminal + SSH)
3. Nano, my nanoclaw assistant who's currently scheduling a coffee chat with my friend's assistant via their own respective Gmail accounts (Telegram)
4. Horst, another friend's @openclaw assistant with whom I've developed a fun conversation around tech, debugging, cool ideas (WhatsApp)

The biggest surprise to me is that it's starting to feel natural and seamless.

27 views

Reddit has fallen.

Up until 3 weeks ago, the AI-to-human ratio was about 30/70.

Within the past 3 weeks, it skyrocketed to 90%.

I'll never go back.

Quoting @nikitabier · Feb 11, 2026

Prediction: In less than 90 days, all channels that we thought were safe from spam & automation will be so flooded that they will no longer be usable in any functional sense: iMessage, phone calls, Gmail.

And we will have no way to stop it.

1 like · 47 views
In reply to @Gavriel_Cohen · Mar 3, 2026

Oof. I immediately went to my VPS and checked because I couldn't remember exactly if I ended up on the correct github repo (and hence cloned the right thing) Good news:

```
cat .git/config
[core]
repositoryformatversion = 0
filemode = true
bare = false
logallrefupdates = true
[remote "origin"]
url = github.com/qwibitai/nanoc…
fetch = +refs/heads/:refs/remotes/origin/
[branch "main"]
remote = origin
merge = refs/heads/main
[rerere]
enabled = true
```

GitHubGitHub - nanocoai/nanoclaw: A lightweight alternative to OpenClaw that runs in containers for security. Connects to WhatsApp, Telegram, Slack, Discord, Gmail and other messaging apps,, has memory, scheduled jobs, and runs directly on Anthropic's Agents SDKA lightweight alternative to OpenClaw that runs in containers for security. Connects to WhatsApp, Telegram, Slack, Discord, Gmail and other messaging apps,, has memory, scheduled jobs, and runs dir...↗ github.com
1 reply · 1 repost · 1 like · 136 views

My single most important learning after installing, customizing and working with @openclaw and @Gavriel_Cohen's nanoclaw:

The biggest risk was hiding in plain sight. And the only way to discover it was by actually exploring claws.

In order to make my point, I first need to briefly walk you through some of the "known unknowns" risks that I had on my radar:

Fear 1: That claws would be simply useless and I would have to face the social shame of trying something that really was just a joke. (Turns out, it IS actually very engaging, useful and even fun.)

Fear 2: Having credentials such as API keys and auth profiles on the server (to make claws actually useful) and then getting those credentials stolen and abused. (Could still happen, but also easily managed with spending limits.)

Fear 3: Customizing my assistant with code and skills from the internet which would lead to any of the above and more subtle attacks such as semantic worms, anti-memetic ███████, cognitive root-kits etc. (Could and probably will happen but we'll deal with it when that is the case)

However. ALL of my fears were DWARVED by the biggest, unanticipated, unmonitored and unmanaged blind spot:

THE FUCKING UTTERLY INSANE TOKEN BURN AND POTENTIAL LOSS OF MY ANTHROPIC MAX SUBSCRIPTION!

In hindsight, it literally sat there right smack in the middle of all the other fears and risks.

And at the same time, I did not have this on my radar AT ALL.

And of course: Once I was aware of it (Thank you @RealistSec for your help), I could start managing it.

My personal solution is that I've given my assistant the ability to sleep, dream and forget. Meaning, at night, a cron-job takes the session data from the previous day, checks for a few important bits to remember and then clears out all session storage. In the morning, the agent starts out fresh and with a free mind.

But my solution isn't what matters here. What matters to me is that I was reminded how action produces information. And that the things that actually hurt often are not what we anticipated. And that the peeps at @AnthropicAI stayed cool (thank you) while a few of their servers went up in flames.

So. That's what I've learned with all of this.

Just wanted to share it :)

Figure 1: A map of known unknown risks associated with "claws". Stuff that you kinda know you'll have to keep an eye on and manage.
The actual danger, the previously unknown unknown that hid right there in the center of the map. "THE FUCKING UTTERLY INSANE TOKEN BURN AND POTENTIAL LOSS OF MY ANTHROPIC MAX SUBSCRIPTION"
62 views

I briefly believed that "cron job" or "claws" would be the words of the year 2026.

Now I believe it'll be "AINXIETY".

...

"Don't believe the hype!" said the great prophet Carlton Douglas Ridenhour and his friend, the wise Bodhisattva William Jonathan Drayton Jr., in unisono, in 1988.

And they were right. Do not believe the hype. If this AI doomerism is getting to you: Take a break. Take care of yourself. Take a nice long shower. Eat something yummy. Call a friend!

And yes, also "do look up." There is so much borderline magic stuff happening right now. You ARE sitting front-row at Dylan's 1965 performance at the Newport Folk Festival.

Things change. They always have and will. The good news is: We humans are actually really good at change. YOU are really good at this.

With love,

Alexis

Image from the post
25 views
In reply to @SpringStreetNYC · Feb 26, 2026

Looking back, I am so grateful and lucky that my godmother Diane had subscribed to @NatGeo throughout the 80s. You could pick any issue from the stack and travel the world instantly.

It's crazy: I can still open any issue from 1985 and recognize every cover, most of the photos and ads and even some of the stories.

Of course, Jame Gurney's Rockwell-y soybean tableau was, and still is, an all-time favorite!

So, on behalf that 8 year-old kid growing up in West Germany and this one adult who has lived in many countries since:

Thank you Diane and thank you National Geographic.
❤️

Image from the post
Image from the post

OMG, I only discovered this now: You can also peruse those foldable supplements in stunning detail. Oof.

Image from the post
1 reply · 49 views

Looking back, I am so grateful and lucky that my godmother Diane had subscribed to @NatGeo throughout the 80s. You could pick any issue from the stack and travel the world instantly.

It's crazy: I can still open any issue from 1985 and recognize every cover, most of the photos and ads and even some of the stories.

Of course, Jame Gurney's Rockwell-y soybean tableau was, and still is, an all-time favorite!

So, on behalf that 8 year-old kid growing up in West Germany and this one adult who has lived in many countries since:

Thank you Diane and thank you National Geographic.
❤️

Image from the post
Image from the post