|
AI Builders Digest
Monday, September 28, 2026
|
|
Thin news day, which is itself a kind of signal. When the most-liked post in your feed is a birth announcement and the second is someone noticing Claude's rate limits quietly disappeared, it suggests the labs are in a heads-down phase. No splashy launches, no emergency blog posts. After last week's OpenAI agent disclosure, maybe everyone decided to keep their heads down for a Monday.
|
|
---
|
|
01
|
Anthropic quietly removed Claude's rate limits and almost nobody noticed
|
|
|
Peter Yang, a product leader, posted over the weekend that Claude's usage limits went from "barely usable" to "basically unlimited" with no announcement he could find. No blog post, no changelog, no press release.
|
Why it matters: Yesterday we covered Aaron Levie's argument that companies can't measure what their agents are doing. Anthropic's stealth limit increase points to the same pattern from the supply side: the product is changing faster than the documentation. If your team budgeted Claude usage assuming the old constraints, those assumptions are now wrong, probably in your favor, but worth checking before you over-provision alternatives.
|
|
Source →
|
|
---
|
|
02
|
Y Combinator president's new favorite debugging workflow uses GPT-6
|
|
|
Garry Tan, president of Y Combinator, posted that his preferred method for fixing production bugs now runs through Capy (an AI coding tool) with something called GStack's `/autoplan` command, backed by GPT-6 medium reasoning. Short post, no tutorial, but the fact that the YC president is casually mentioning GPT-6 as part of a daily debugging stack is worth noting.
|
Why it matters: Six months ago, GPT-4 was the default. Now "GPT-6 medium reasoning" is a throwaway detail in a bug-fixing workflow post. The capability ladder is moving fast enough that the baseline keeps shifting under teams that aren't paying attention.
|
|
Source →
|
|
---
|
|
03
|
One year of using Claude Code for video generation: the pace of change, illustrated
|
|
|
Thariq posted a reflection on a milestone: roughly a year ago, he was among the first to use Claude Code to generate videos, a process that required constant iteration, pointing out small errors, nudging the model toward the right output. He shared examples of what that looked like then, as a contrast to where things stand now.
|
Why it matters: This is a useful calibration post. If your team is still treating AI video or code generation as an experimental side project, you are probably a year behind where practitioners like Thariq already are. The "early adopter tax" of slow iteration that he's describing is largely gone.
|
|
Source →
|
|
---
|
|
04
|
A mysterious reset, 14,000 likes, and no context
|
|
|
Thibault Sottiaux posted "Resets all propagated. That will be all. Have a fantastic weekend." It got nearly 14,000 likes and almost 1,900 replies. There is no additional context in the post itself, and without knowing what system or incident this refers to, there is nothing substantive to report. The engagement suggests something significant happened in a community that already knew the backstory.
|
|
Source →
|
|
---
|
|
05
|
Vercel CEO Guillermo Rauch welcomed a daughter
|
|
|
Rauch posted a birth announcement for his daughter, Lisa Rauch. Not AI news. Congratulations to the Rauch family.
|
|
Source →
|
|
Follow builders, not influencers. A daily digest of what matters in AI.
Read online ·
Archive
|