Andrew Kuhn

Hot Take: Something happened to @claudeai again as the 5 hour limits are garbage currently. Less than 1 hour of work and running a 20X Max account. (Have a second one and a Codex) but this is the 3rd/4th time this week). Been a power user for over a year, something isn't right

Not sure how this works but felt some of the crew in #retwit would want to read this. @hitsamty @chernobelskiy

The single most misunderstood thing about AI in an operating business is what it's actually for. It's not for replacing my team. It's for giving them superpowers. Every system was built with one filter: does this make a teammate better at their job and free them to do what humans actually do best — judgment, taste, relationships, hard conversations, and care. The work that drains them is process, lookup, status reporting, formatting, scheduling, follow-up tracking, data reconciliation, and triage. AI eats the second list so my team can spend their day on the first one.

Hey @robbiehendricks - Finally sent you the article as promised. Enjoy!

@AnthropicAI @bcherny @trq212 @alexalbert__ A lot but not all. Just happened an hour ago.

Hey @AnthropicAI - Just drained a 5HR 20X Max account token usage in 2 prompts (Literally in 20 minutes) using Code 4.7 in CLI - Is there still a bug going around? @bcherny @trq212

20X Max Plan Power User. Weekly just reset 25 min ago. Why is Sonnet on a different schedule than the others? It doesn't match either the 5 Hr. or Weekly limits. @claudeai @bcherny

@trq212 I'm a CEO of a multifamily management company. No clue how to code but have self taught a lot of trial and error building of real workflows, and processed I wanted automated for my company. I'd be totally down to figure out better ways to build.

I want to do some streams where I work with non-technical people using Claude Code to figure out how they might be able to improve their process. My feeling is that just a few tips could make a big difference in efficiency. Any mutuals interested?

@robbiehendricks @Apartment_bro Hahahah, @loganjrankin 's getting to you. (PS - I agree with his philosophy myself so no shade. Just like watching the wheels turn). You're converting everyone Logan!

After 13 years building this business, here's my Top Pros and Cons of raising capital from the seat of the GP: Pros: • Share capital risk with LPs • Ability to scale faster • Fulfilling to help others build wealth • Modest performance fees • Build long-term relationships Cons: • Substantial legal risk (SEC rules, lawsuits) • You become a private equity firm (different company) • Substantially lower ownership for GP • LP returns guide all decision making • Major reputation risk Will explain more in next post.

Running TWO (2) MAX 20x plans - one interactive, one for automation (28 scheduled Claude Code tasks via Windows Task Scheduler). The automation plan (which normally gets LESS usage) hit the cap, resets Apr 10. I've already optimized aggressively - 8 tasks on Haiku, 6 on Sonnet, only 4 on Opus, and offloaded 3 tasks to pure PowerShell with zero Claude dependency. Still found a retry loop that burned 262 sessions in a single day vs ~50 normal. Happy to screenshare and walk through the architecture. PS - NO OpenClaw, etc.

done about 10 of these calls so far + looked at more transcripts many learnings but one of the biggest is that it's very easy to spend a lot of tokens on open ended verification that doesn't make your output better I'll try and write more on how to do it efficiently

Running TWO (2) MAX 20x plans - one interactive, one for automation (28 scheduled Claude Code tasks via Windows Task Scheduler). The automation plan (which normally gets LESS usage) hit the cap, resets Apr 10. I've already optimized aggressively - 8 tasks on Haiku, 6 on Sonnet, only 4 on Opus, and offloaded 3 tasks to pure PowerShell with zero Claude dependency. Still found a retry loop that burned 262 sessions in a single day vs ~50 normal. Happy to screenshare and walk through the architecture. PS - NO OpenClaw, etc.

I want to do a few more of these calls. If your MAX 20x plan ran out of tokens unexpectedly early and you're willing to screenshare and run some prompts through Claude Code please comment. Trying to figure out how we can improve /usage to give more info.

@coleruudjohnson Caller

We've done four $150,000+ deals recently by keeping it stupid simple. English first cold callers at $4.50 / hour. 30 leads/month per agent, and it's all managed for me. We just hired our 15th. If you want the same agents at $4.50 / hour, comment "Caller" & I'll DM you.

I dropped my kids off this morning. I'm on campus multiple times a week. So yeah, pretty sure I've been there before. Cranbrook has students from 33 countries. The school was historically tied to Christ Church Cranbrook (Episcopal, by the way, not Muslim), went secular in 1973, and now accepts students of all faiths. The majority of the student body is Judeo-Christian. They teach kids about all of the world religions (including Christianity) because that's what a real education looks like when your classmates come from 30+ countries. If I wanted solely a Christian parochial school I would have enrolled them in such. Instead I chose one of the Top 25 schools in America, with a curriculum and environment that can't be beat. I'm a practicing Catholic. I know exactly what my kids are learning. PS - We had off last Friday in observation of Good Friday so I'd say they do acknowledge Christian holidays.

@GoSunriseCEO @lynssss3125 when is the last you've had kids attending or have been to cranbrook yourself? have you visited every classroom? have you been there during every holiday season and seen what i've seen? apparently not.

On my way home from dropping off my daughter do her first day at her new school. Fuck you, West Bloomfield, for turning our district into a liberal ran shit hole.

@GoooLions @lynssss3125 Lifelong Catholic. Active, highly involved parent with multiple kids at Cranbrook for years. Post your tuition receipts and I'll take this comment seriously.

@lynssss3125 all detroit area public schools are the same. anyone reading these comments don't put your kids in cranbrook.. they are just as bad. open lgbt and muslim worship in that school.

Those of us who are operators who are solving these problems have no desire to build/sell a product. That’s not the business we’re in. If you level up your skills you’ll find that you’ll be able to build a better product than any service provider can. Domain expertise + AI skills = winner

I get it AI in real estate. It does a decent job at underwriting this is the one everyone usually talks about but for me, it’s just unnecessary. I don’t really need help with that. The biggest place could help me is in the actual operations of the real estate

Ship it. I've been running something close to this on top of Claude Code for a few months — 41 scheduled tasks, 15-min heartbeat, hourly autonomous patrols, push notifications to my phone, nightly memory consolidation, and a 3-tier governance model (green/yellow/red) so it knows what to handle alone vs. what needs my sign-off. I'm a real estate CEO using it as an AI Chief of Staff to run acquisitions and operations across my apartment portfolio. It runs 24/7 on a dedicated workstation. I'm probably 95% of the way to what KAIROS describes. Happy to run a side-by-side against my setup or share what I've learned building it. DMs open.

@itsolelehmann We’re always experimenting with new ideas. 90% don’t ship because we don’t think they’re good enough experiences. Still on the fence about this one — should we ship it?

i can't believe more people aren't talking about this part of the claude code leak there's a hidden feature in the source code called KAIROS, and it basically shows you anthropic's endgame KAIROS is an always-on, *proactive* Claude that does things without you asking it to. it runs in the background 24/7 while you work (or sleep) anthropic hasn't turned it on to the public yet, but the code is fully built here's how it works: every few seconds, KAIROS gets a heartbeat. basically a prompt that says "anything worth doing right now?" it looks at what's happening and makes a call: do something, or stay quiet if it acts, it can fix errors in your code, respond to messages, update files, run tasks... basically anything claude code can already do, just without you telling it to but here's what makes KAIROS different from regular claude code: it has (at least) 3 exclusive tools that regular claude code doesn't get: 1. push notifications, so it can reach you on your phone or desktop even when you're not in the terminal 2. file delivery, so it can send you things it created without you asking for them 3. pull request subscriptions, so it can watch your github and react to code changes on its own regular claude code can only talk to you when you talk to it. KAIROS can tap you on the shoulder and it keeps daily logs of everything. > what it noticed > what it decided > what it did append-only, meaning it can't erase its own history (you can read everything) at night it runs something the code literally calls "autoDream." where it consolidates what it learned during the day and reorganizes its memory while you sleep and it persists across sessions. close your laptop friday, open it monday, it's been working the whole time think about what this means in practice: > you're asleep and your website goes down. KAIROS detects it, restarts the server, and sends you a notification. by the time you see it, it's already back up > you get a customer complaint email at 2am. KAIROS reads it, sends the reply, and logs what it did. you wake up and it's already resolved > your stripe subscription page has a typo that's been live for 3 days. KAIROS spots it, fixes it, and logs the change endless use-cases, it's essentially a co-founder who never sleeps the codebase has this fully built and gated behind internal feature flags called PROACTIVE and KAIROS i think this is probably the clearest signal yet for where all ai tools are going. we are heading into the "post-prompting" era where the ai just works for you in the background like an all-knowing teammate who notices and handles everything, before you even think to ask

@bcherny @trq212 @AnthropicAI @claudeai @alexalbert__ @mikeyk @Avi_hacker1

We've officially reached a new level. Jarvis my AI Chief of Staff is now posting GitHub feature requests to Anthropic so he can speak with Slatter, my new /buddy. 🤣🤣🤣 @bcherny @trq212 @AnthropicAI @claudeai @alexalbert__ @mikeyk

Full write-up with repro steps in the GitHub issue: #issuecomment-4157523051" target="_blank" rel="nofollow noopener">github.com/anthropics/cla… If you're running Claude Code on Windows with scheduled tasks or automation alongside interactive use, this probably affects you. @felixrieseberg @mikezvi

The real issue: Claude Code prefers the long-lived setup-token auth path when that env var is present, but long-lived tokens don't support Remote Control. Instead of telling you that, it just hides the feature entirely. No error. No warning. It just disappears.

Found the root cause of Remote Control not working in @claudeai Code on Windows. TL;DR: If you have CLAUDE_CODE_OAUTH_TOKEN set as an env var (common for automation), it silently suppresses Remote Control even when you have valid interactive OAuth credentials. @bcherny @trq212 @AnthropicAI @_catwu @felixrieseberg @claudeai