shipping things outside the big lab RL circuit. building self flying airplanes @ https://xitter.cf/t.co/P0GOCDqMFH
San Francisco
Joined March 2010
- Tweets3.3K
- Following2.1K
- Followers2.1K
- Likes113K
In other words, they're pushing beyond the 1M token context window and need to train on longer RL traces.
Anthropic is now recommending that companies REMOVE unnecessary subagents from their AI systems.
This is a pretty significant shift in agent architecture, with their team finding that a single agent equipped with skills can frequently outperform complex multi-agent systems, and one customer seeing roughly 2x faster execution.
The emerging rules of thumb:
→ Skills: Teach the agent HOW to do something, including procedures, expertise, and instructions.
→ Tools / MCP: Give the agent the ability to DO things through APIs, actions, and live data.
→ Subagents: Use for parallel work, isolated context, or tasks a single agent struggles with.
The interesting part is progressive disclosure. Instead of spawning specialized agents or stuffing everything into a giant prompt, one agent discovers and loads skills and tools as needed.
Their advice is to fold existing subagents into the main loop, turn their instructions into skills, expose their actions as tools, and run evals to see if performance holds up.
As models have gotten smarter, improving agent architecture is becoming less about what you can add and more about what you can delete.
claude.com/resources/article…
Survival of the fittest let it rip
Some stats on #php2shell. In the 10 hours before it was publicly announced, across the @Cloudflare network we saw thousands of attempts to exploit it. That jumped to tens of thousands in the 45 minutes post announcement, hundreds of thousands within 2 hours, and millions now.
IDEs are as useful as a steering wheel on a Cybercab.
The enshittification of once relevant tech companies continues.
We connected IntelliJ IDEA’s code intelligence to Claude Code and Codex, then tested the setup with Netflix.
The first public results are out.
Typical efficiency gains: 20–30%.
Best code-navigation case: ~65%.
These results come from two real codebases. The efficiency metric averages token savings and time savings.
Correctness stayed stable or improved.
Here’s what we actually built.
The code engine behind IntelliJ IDEA ran headlessly, giving agents access to its capabilities through a CLI backed by a daemon.
LSP alone wasn’t sufficient.
Agents needed a convenient way to invoke the tools. The backend also needed to stay running and keep the project index warm between calls.
The idea is straightforward.
An IDE has already mapped the structure of your code and its dependencies — including what’s inside compiled libraries.
An agent can use that knowledge instead of spending time and tokens reconstructing it.
But the most interesting finding was which models benefited most.
The strongest models saw the biggest gains.
Especially at higher reasoning effort.
In these experiments, more capable models were also better at taking advantage of code intelligence.
My takeaway: better models and better tools aren’t competing bets.
You don’t have to choose between investing in model intelligence and giving agents better access to your codebase.
The two can reinforce each other.
Huge thanks to @pbakker and the Netflix team for working with us on this!
Fire your designer. How does this landing page not have flying cars all over it? @Tesla
🤖 Made with AI
hey chatgpt dress me up in the traditional costume of the country i look like im from
I guess I look Norwegian
ALT A smiling man stands outdoors in front of lush green foliage, wearing a navy EpicWeb T-shirt, blue jeans, and a black smartwatch, with both hands tucked casually into his front pockets.
ALT A smiling man stands outdoors in front of lush greenery wearing an ornate traditional Norwegian-style bunad: a black embroidered jacket and trousers with red trim and floral detailing, a red patterned vest, white high-collared shirt, and decorative silver buttons and brooch.
This argument is so dumb.
Asset ownership only works within a system where property rights are stable and people still have something of value to exchange.
Imagine a world where human labor is economically irrelevant. A small group owns the assets while the majority has little or no economic leverage because their labor is no longer needed.
Telling people to buy land and gold to prepare for that world is a category error.
Ownership itself depends on a social order capable of enforcing ownership. If the majority has no economic stake in that order and no meaningful way to exchange labor for the things they need, the question eventually becomes political rather than financial. Property rights ultimately rest on collective legitimacy and, at the outer limit, the state’s monopoly on legitimate force. You cannot indefinitely preserve an extreme distribution of assets in a society where most people have no productive claim on the economy simply by telling them to respect the title deed.
For most of human history, people have essentially exchanged labor for energy (labor itself being a form of energy) and the goods that energy produces. If energy becomes effectively free, as it should, and robots become the labor, humans collectively become the owners of the robots.
At that point, there are only a few plausible equilibrium states. Either the benefits of effectively free energy become so abundant and broadly distributed that money and conventional asset ownership lose much of their meaning or society develops some mechanism for distributing access to that productive capacity.
If neither happens, there is no obvious stable equilibrium. The alternative is not some clever portfolio allocation. It is eventually a contest over who has the legitimacy and power to control the robots, which ultimately means violence.
So seriously, what the fuck are you going to do with your assets in that world?
Hear me out. The value of human labor is trending toward zero.
Some of the most talented people I've ever worked with can't find jobs right now.
And the people who don't see what's coming are responding the exact wrong way. More degrees. More certificates. More skills. The belief that if they make themselves more valuable to an employer, they'll be safe.
This is not the way.
The world is dividing into two classes: the labor class and the asset class.
The only way out is owning assets.
Scrimp. Save. Convert every dollar you can into things that can't be inflated away. Gold. Land. Bitcoin. To a lesser extent, stocks.
Investing in the value of your own labor is a great way to go broke slowly.
If you do invest in your skills, do it for exactly one reason: to convert that labor into assets as fast as humanly possible.
Labor is the on-ramp. Assets are the only way out.
Texas Roadhouse is a regular stop for me on the way to Tahoe. It’s underrated.
1908: The Horseless Carriage Means Trouble
2026: AI Agents Mean Trouble
Some people never change.
recommended reading. for perspective.
blog.bustikiller.com/2026/09…
Jake Murzy retweeted
To know what Chinese labs are doing you can just read their papers.
To know what American labs are doing you have to wait for @karpathy to do a 1 year internship and then post a GitHub repo with what he learned.
Two things can be true at once, but this coexistence is often temporary.
The truths in tension today are between those betting on capability overhang, positioning for a future where accumulated theory becomes obsolete, and craftsmen who understand systems deeply but remain skeptical that the singularity is coming to save us. If the overhang pays off, the craftsmen lose. If it doesn’t, the extractors burned the accumulated theory they’ll now need.
Place your bets.
recommended reading. also read part II. also read naur's original article.
cekrem.github.io/posts/progr…