|
AI Builders Digest
Tuesday, June 9, 2026
|
|
Yesterday we talked about token costs eating enterprise budgets. Today we're seeing why those bills might be worth it: developers are running Claude autonomously for days at a time and discovering what AI can actually do when you stop micromanaging it.
|
|
01
|
Anthropic's Claude Opus emerges as the go-to model for long-running AI work
|
|
|
Boris Cherny shared five tips for running Claude Opus autonomously for hours or days, based on new benchmarks showing it outperforms other models on extended tasks. His advice: use auto mode to skip permission requests, deploy dynamic workflows to orchestrate hundreds of agents, use goal commands to keep Claude working until completion, run it in the cloud so you can close your laptop, and make sure Claude has enough usage allowance to finish the job.
|
Why it matters: This is the first practical guide for letting AI work unsupervised on complex projects. If the early reports hold up, we're moving from "AI as a smart autocomplete" to "AI as a research assistant you can assign multi-day projects to."
|
|
Source →
|
|
02
|
Former Google engineer explains why we have coding agents but not knowledge work agents
|
|
|
Madhu Guru argues that training data quality, not quantity, is the real bottleneck for advancing AI capabilities. The reason we have software engineering agents but not financial planning or legal research agents comes down to documentation: coding has Stack Overflow, GitHub, and clear examples everywhere. Knowledge work relies on domain-specific expertise built over years using legacy tools that don't integrate. Most of that knowledge has never been written down.
|
Why it matters: Your industry probably won't get AI agents until someone figures out how to capture the tribal knowledge that exists only in senior employees' heads. The sectors with the best documentation will get AI first.
|
|
Source →
|
|
03
|
Box CEO Aaron Levie: AI won't kill enterprise software companies after all
|
|
|
Aaron Levie pushed back against the "AI will eat enterprise software" narrative, pointing out that building good software was never the main cost for enterprise companies anyway. Most of the budget goes to sales, marketing, and customer success because enterprise software categories are winner-take-all markets. AI making development cheaper doesn't eliminate the need for distribution, brand trust, and customer relationships.
|
Why it matters: All those AI startups betting they can build better software and steal enterprise customers are missing the point. The hard part was never the code. It was convincing a Fortune 500 CTO to bet their career on your product.
|
|
Source →
|
|
04
|
Anthropic's Codex team launches 100-day power user experiment
|
|
|
Thibault Sottiaux from Anthropic announced they'll select one Codex user per day for the next 100 days and give them 10x usage limits for a month to see what's possible with unrestricted access. The goal is to find people doing "impressive or incredibly useful work" and remove the usage constraints that might be limiting their experiments.
|
Why it matters: Anthropic is essentially crowdsourcing R&D by giving power users unlimited budget to push their tools to the breaking point. The projects that emerge from this will preview what's possible when cost stops being a constraint.
|
|
Source →
|
|
05
|
Product manager Peter Yang's AI dad joke experiment
|
|
|
Peter Yang asked his followers to finish a dad joke: "My wife asked me to take out the trash. I said, 'You shouldn't be prompting me anymore. You should be designing loops.'"
|
|
Source →
|
|
Follow builders, not influencers. A daily digest of what matters in AI.
Read online ·
Archive
|