The planner worker architecture works well for me. About 3 layers is the sweet spot. From prompt -> plan -> task division -> workers.

Sometimes workers will task other workers and act as a planner if the task is more complex.

It’s a good setup but it’s nothing like Claude Code.

I've always liked the idea of intelligence in the autonomous ships of the Revelation Space universe. Little agents reporting to progressively more intelligent and higher level ones.

Supposing agents and their organization improve, it seems like we’re approaching a point where the cost of a piece of software will be driven down to the cost of running the hardware, and the cost of the tokens required to replicate it.

The tokens were “expensive” from the minds of humans …

I find it interesting that this line of adventure quickly lead to locking problems.

Is it too much to expect companies to share some of this in the open vs just the results?

Can a browser expert please go through the code the agent wrote (skim it), and let us know how it is. Is it comparable to ladybird, or Servo, can it ever reach that capability soon?

would really appreciate some elaboration as they gloss over the most important part in my kind. why can’t one agent just do it. that’s what ai seems to be - an amalgamation of all our knowledge. why split it back up into separate tentacles. i think focus should be on letting it envelop the problem like a fog and swallow it whole, instead of molesting it independently at touch points and reporting back to … the brain? it’s pretty ridiculous actually. just mimicking ourselves yet again.

Is the code not even compiling a feature or...

Very cool. Seems long running AI Agents are the new monuments.

At the same time they were doing this, I also iterated on an AI-built web browser with around 2,000 lines of code. I was heavily in the loop for it, it didn't run autonomously. You can see the current version of the source code here:

https://taonexus.com/publicfiles/jan2026/172toy-browser.py.t... (turn the sound down, it's a bit loud if you interact with the built-in Tetris clone.)

You can run it after installing the packages, "pip install requests pillow urllib3 numpy simpleaudio"

I livestreamed the latest version here 2 weeks ago, it's a ten minute video:

https://www.youtube.com/watch?v=4xdIMmrLMLo&t=45s

I'm posting from that web browser. As an easter egg, mine has a cool Tetris clone (called Pentrix) based on pieces with 5 segments, the button for this is at the upper-right.

If you have any feature suggestions for what you want in a browser, please make them here:

https://pollunit.com/polls/ahysed74t8gaktvqno100g