Yesterday we talked about the gap between AI demos and production reality. Today's posts show what happens when builders actually figure out how to cross that gap.
01
The token bloat problem nobody talks about
Peter Steinberger called out a widespread issue with AI agent skills: developers are writing "books" in their skill descriptions, and all that text gets loaded into every context window. He's even built a tool to find the worst offenders and recommends asking agents to be "token efficient" and relaxing grammar in skill files.
Why it matters: If you're wondering why your AI agent bills are higher than expected, check your skill descriptions. Every extra sentence in there is costing you money on every single request.
Peter Yang shared another insight from Ryan Carson (the founder running his entire company with AI agents): the old "ship fast, build systems later" advice is now backwards. Carson says you have to spend significant time upfront on documentation and systems before AI agents can do the work of 10 people.
Why it matters: Every startup founder who skipped documentation because "we'll fix it later" is now paying for it. Your AI agents are only as good as the systems you build for them to work with.
Anthropic's Amanda Askell warns about fake blog posts
Amanda Askell from Anthropic posted a heads-up that she hasn't written a personal blog post in over five years, so any posts claiming to be from her are fake. A simple warning, but telling that she felt the need to post it.
Why it matters: As AI-generated content floods the internet, even AI researchers have to actively deny authorship of fake posts. If you're seeing a sudden surge of "insights" from prominent AI figures, double-check the source.
Aaron Levie shared Goldman Sachs CEO's take on AI and jobs, arguing that technology historically makes work faster and easier but we just end up demanding more output. Instead of automating tasks to deliver the same value cheaper, companies expect more from everything.
Why it matters: This is the "productivity paradox" playing out in real time. Your company won't use AI to do the same work with fewer people — it'll use AI to do dramatically more work with the same people.
Y Combinator's Garry Tan predicts the rise of "prompters"
Garry Tan shared a post about someone landing a job specifically for their prompting skills, calling it a "brave new world" where "prompters of the world unite." Short on details but big on implications for where the job market is heading.
Why it matters: "Prompt engineer" is becoming a real job category, not just a meme. If your company isn't thinking about who owns prompt optimization, you're behind.