Richard Feldman

1/3 + 1/9 + 1/27 + ··· = 1/2. The proof is in the picture. No words needed.

What is your favorite git worktree integration for coding agents? Why?

I don't think most people would call C-3PO superintelligent.

Current LLMs don't feel like C-3PO to me. But I bet LLM agents could get there through incremental improvement from where they are today; I'd be surprised if it turned out a research breakthrough was necessary.

I think most people's idea of AGI is basically C-3PO. - Never forgets earlier parts of conversation, or past conversations - Annoyingly flawless with numbers - Main flaws: misinterpreting humans, skill limitations, making bad decisions - not confidently misunderstanding reality

@liran_tal Cool, thanks!

@rtfeldman Just from the past weeks for Claude Code for example: 1. /btw 2. /voice 3. learned they have a way to stash current chat

In the past 3 years, I haven't noticed any uptick in release speed for software I use. If productivity is increasing, I can't tell as an end user. I have noticed decreases in uptime, increases in bugs, and a HUGE increase in people bragging about how many PRs per day they land.

@liran_tal Their release notes can't tell me what you've personally found notable as an end user, which is the specific thing I'm curious about.

@rtfeldman You don't really need to ask me though? Just look at the Claude Code stream and everything coming out of Anthropic. Enough to follow the company accounts to see all the updates.

@keeslinp Hard for me to judge because I'm working on it, but I haven't heard people comment on it.

@rtfeldman I'm curious, do you think Zed has had an uptick? From what I understand the disconnect mostly stems from Jevon's paradox and most of the productivity gains are going towards increasingly lower priority items so the high level mission velocity is not affected

@liran_tal Fascinating! What are the new Claude Code features or improvements you've found personally notable as an end user?

@rtfeldman My experience is totally different, yes. As someone who builds with AI the productivity gain is probably beyond 10x too To your point: > I don't read release announcements Who has time and to what value you'd create release announcements when you release daily? :D

@modisulak In some cases the bragging is coming from some of the buggy, slow software I am literally using on a daily basis.

@rtfeldman the gains are real but they're in greenfield. the software you use is mostly legacy codebases where AI barely moves the needle. the bragging is coming from the other category.

@guidomb That's what it feels like to me too as a builder, but as an end user I'm just not seeing it. 😅

@rtfeldman But we are able to iterate much faster and stabilize much faster. We now have a testing infrastructure that would have cost a lot of time to develop prior to LLMs and probably would not have built it if it weren’t for the agents

@liran_tal Maybe your experience has been different though? Are there a bunch of new Claude Code features or improvements you've found personally notable as an end user? (I don't use OpenClaw.)

@liran_tal e.g. over the past 3 months, the main thing I've noticed off the top of my head is that the Plan Mode UI has changed in various ways. Sometimes I've seen a "talk about this plan" option, which is what I most often want, and it appears inconsistently. Can't say I'm impressed.

Thrilled to have Andrew Kelley, creator of @ziglang speaking at Software Should Work. Early bird ticket sales end at midnight!

I am sooooo excited for this conference 🤩 Early bird tickets close Friday, hope to see you there! softwareshould.work

I love incidents. Mostly because whenever there's a big public one I know I'll get to read @norootcause's write-up about it! The way large systems break under pressure is fascinating and Lorin does an excellent job sussing out the themes underlying failures across the industry. Reading Lorin's blog (Surfing Complexity) has completely changed the way I think about incidents and resilience. Naturally, I'm thrilled that Lorin will be speaking at Software Should Work. Early bird discount ends this Friday!

A short note that the predictions that LLMs would favor "boring technology" that's over-represented in the training data don't appear to be playing out as expected with the latest models - once you attach them to a good coding agent harness at least simonwillison.net/2026/Mar/9/not…