Twitter died a long time ago and X isn't worth it. Before I made the jump, my ego was holding me back. I was too concerned with vanity metrics. I finally realized if you're posting stuff people want to follow, then they'll follow you, even when you leave. www.siliconrepublic.com/enterprise/k...

To bad things from Claude: The cursor does not blink so if you have split windows you don't know if you are in the Claude split. Also Ctrl+g to edit the prompt with vim works weird with pasting.

holy shit

Two things changed the way I work last year. One is ofc Claude Code and there's no news there. The other one is switching to working with command line notebooks (write commands on left split, block select and execute on right split). It does make a big difference. 👇

Learn to "Lead without authority" is one good piece of advice for any non pure IC role. Even if you have "official authority" it normally means little and relying on it alone won't work

"feeds", and more precisely a "For You" for AI 🤔

Turns out I have accumulated lots and lots of "skill" files in Markdown throughout the years. They were intended for me, but should be usable by agents. I am only missing the motivational prompts "you are the best croquetas chef in the world, you will read the instructions in detail ..."

More user contributed algorithmic improvements to bsky please, this is the way!

For You has become the most liked feed: bluefacts.app/feeds/top Thanks everyone who gave it a like! Liking a feed may seem meaningless, but I think it does matter to the feed creators. It definitely has been motivating for me. Consider going over the feeds you have pinned and giving them a like!

There was a short-lived bug in Claude Code some months ago that caused dead background tasks to continually nag Alpha, never going away. It was very much like she was sick. She hallucinated and generated out-of-place Chinese text. It was *creepy.* Way worse than watching an algorithm malfunction.

I am curious to see Claude Code's revenue numbers for January

B2B being call centers and other services over audio

Voice neuro-software is going to be pretty mind boggling and also feel much more like AI than anything else (well, talking robots will create a far stronger "WTF future is here feeling")

This is a lot in my mind. You can imagine various timelines where "non-transactional value creation" reigns. In all the ones I can imagine, we still need to pay for the infrastructure and services 👇

As the cost of development goes down, people will pay for real problem solving, design and creativity. One could even have a feature shop in it and bsky could take a cut. (Yes, talking about money does not feel great, but this thing you are looking at costs money)

Custom Bsky clients are great, we should continue to do them, but isn't fragmentation a big issue? Could a secondary Bsky app "BlueSky Labs" where it's easy to contribute features (with some minimal vetting) be a better solution?. Less decentralized, yeah, but much more ecosystem reinforcing

Also when the tools do most of the token generation, even local models go quite fast (since it's mostly prefill). This adds to the feeling of "ugh, I would not like to have to fend off this thing"

Sitting in the middle of an agent tool call loop, selectively refusing tools and actions is kinda scary. These things do have lots of ways of getting where they want. The equivalent of High Frequency Trading for infosec is going to be wild.

Gas Town is nutty, and Yegge seems a bit AI manic, but he's also leaning into it for effect and having fun. I think there is real value in what he's doing--maybe not Gas Town as the final form, but in investigating what patterns actually work here, at all.

Suddenly, code quality becomes that latent variable conditioning observables like, "it ran for 3 days now", "fan noise is still tolerable", "I am mostly sure I had no files in that folder", "passes all the tests, no I did not look at those either"

Another advantage of Coding Agents: people that only touch their own code can't even tell if it is their own code.

Widely available data at scale. Sounds like robotics is accelerating!

I sound like a broken record but ... Adversarial Interoperability with a touch of reclaiming ownership of your data bsky.app/profile/ramo...

How popular is the theory that GPT is too shifted towards hard problems (e.g. trying to replace C++ devs) while Claude focuses on empowering non-core-devs wanting to do real world stuff (i.e. being the Python of LLMs)

Just to be clear, I am talking improvements from research (this is all about research). Improvements from productization will keep piling up and the impact will be gigantic. We will probably see a first peek this year.

Papers becoming more philosophical, successful productization, improvements decelerating ... I know this stage. Hopefully cross-modal transfer gets us out of here?

We just found the subset of users permanently unhappy about coding agents!

Bluesky is a social annotation layer. We should ease up plugging data streams into it. We should ease up broadcasting the meta data out.

Oops, I started a new project: Skyreader, an RSS reader on the AT Protocol. Share cool articles like it's 2010 and Google Reader would never die. skyreader.app www.disnetdev.com/blog/2026-01...

It seems the crypto community has found a new way to extract money from people. They are creating tokens for popular open source projects and inviting maintainers to get in on the scheme. Here is the email I got earlier today.

Blacksky @blackskyweb.xyz now has its own servers running the network This means they're able to run their own feature development and moderation policies, while still connecting to Bluesky through the shared Atmosphere network! s/o to @rude1.blacksky.team and the blacksky team for amazing work

With this, I mean that meme coins are like Tinkerbell, they cease to exist if you stop believing in them. Here you are declaring the value to be pegged to a product *and* paying the product creator a cut, therefore creating a self-reinforcing real bound between chances of success and valuation

Walking in Central Park in Winter disconnects you from the digital world in that, at some point, your fingers are too frozen to unlock the phone

Correction: "non-consensual patreon meets meme coin"

In other words, this is "patreon meets meme coin"

Lol, this seems like "non consensual company shares using crypto". I have many feelings and some are not repulsion. 👇 steve-yegge.medium.com/bags-and-the...

Gemini is robustly becoming my go to search engine when I don't want that knowledge to be applied to coding something.

Another Erdos problem announced solved by machine learning: github.com/teorth/erdos... From discussion, looks like it may have already been solved in prior lit. A weird thing about neural networks is that we can't actually tell if it used that source or not.

Ofc if "negotiating agents" do not work, under some conditions, they may generate a hellish experience. Some Goodhart's Law driven dystopia worthy of a (true) Black Mirror episode.

I used to imagine "negotiating agents" as something that would travel with you and get you a good local internet offer during vacation or even access to an uncommon event. Projecting everything onto the money axis is often a terrible approximation, even if you only care about money! 👇

If you think about it, this is like the "private package registries" that Astral intends to offer to monetize "uv". Corporate security is a very serious matter, but expensive. OSS security has to be a business or paid by a business (case point FFmpeg). 👇

I wonder how much Anthropic and others are going to try to restrict their neuro-software (e.g. Claude Code Scaffold and included Tooling) to only work with their neuro-hardware (e.g. Claude). Ideally there is open competition on both spaces, and in both Open Source/Weights has something to say!

Seems like a very smart move.

Oh, this reminds me a long time ago I bought a beautiful Skaven Warp Fire thrower, only for it to blow up on the first and only time I got to use it in battle, incinerating the poor rats and some unfortunate goblin.

This triggered the thought BTW x.com/i/status/201...

So I guess "App stores, but for agents" and "agent SEO" is coming? with this I mean: You ask the agent for some custom app. Agent tells you what services/software you need to pay for (e.g. Supabase, some domain provider). Those decisions will move a lot of money. 👇

aaand it's gone

Fun fact, playing Overwatch increases For You response time from 90ms to 115ms.

Steve is bold enough to ask “What if the solution to this problem didn’t come from organisational control theory? What if the solution didn’t come from agent-based computational economics? What if the solution came from this?”

Not sure what everyone’s so confused about. This is perfectly clear to me.