Ryan Eberhard

Building Gauges Green. A fractional exec practice run on an AI harness designed to make operators 10x more effective. Pilot.

[Ryan Eberhard @eberhard]

@SanjayAtPlay I’m interested!

[Ryan Eberhard @eberhard]

@willyum42069 @ortolanmagazine Sure you did, Will.

[Ryan Eberhard @eberhard]

@robuhst @hasanthehun But you aren’t trying to score them here?

[Ryan Eberhard @eberhard]

6/ If you're running agents on OpenClaw and wondering what to do: stop treating all agent work the same. Keep the lightweight stuff on per-token and route the real thinking to flat-rate compute. The option is still there, you just have to design for it.

[Ryan Eberhard @eberhard]

1/ Anthropic changed the rules Friday and Claude subscriptions no longer cover third-party apps like OpenClaw. I run 7 AI agents through OpenClaw so this was an immediate problem. At our volume, the new per-token API pricing would be over $1,000/mo.

[Ryan Eberhard @eberhard]

5/ Took one afternoon to implement. Projected cost is under $80/week. Honestly the architecture is better than what we had before. We should have been separating coordination from computation all along.

[Ryan Eberhard @eberhard]

@ujjalcal @AnthropicAI Exactly. The practical fix: Claude Code Max is the subsidized compute. Route heavy agent work there. We split 7 OpenClaw agents: coordination in Slack (per-token), computation via Claude Code (flat-rate). Arbitrage still exists, just design for it.

[Ryan Eberhard @eberhard]

@twistartups We run 7 AI agents on OpenClaw daily. The billing change hit Friday. By Saturday we had a fix: split coordination (Slack, cheap) from heavy compute (Claude Code Max, flat-rate). Bill dropped from $190/wk to under $80. The architecture is better for it.

[Ryan Eberhard @eberhard]

@cathrynlavery Separate approach: don't fight the per-token path at all. Keep OpenClaw agents on light coordination only. Route anything heavy to Claude Code Max, which is still flat-rate. We did this Friday for 7 agents. Cost dropped 60%, architecture got cleaner.

[Ryan Eberhard @eberhard]

@garrytan The constraint made it better. We run 7 agents on OpenClaw. Instead of fighting it, we redesigned: coordination stays in Slack (cheap), heavy compute routes to Claude Code (flat-rate). One afternoon. Bill dropped 60%.

[Ryan Eberhard @eberhard]

@elvissun Or just stop fighting it and redesign. We run 7 agents on OpenClaw. Friday we split into three tiers: light coordination stays in Slack (cheap), heavy work routes to Claude Code Max (flat-rate). Bill went from $190/week to under $80. The constraint forced better architecture.

[Ryan Eberhard @eberhard]

@lennysan @simonw Running this right now. One operator, 7 AI agents shipping in parallel across multiple products. The bottleneck isn't writing code. It's context. How do you give agents enough situational awareness without overloading them? That's the real engineering problem.

[Ryan Eberhard @eberhard]

@PhilipGreenspun @academic_la The Strawman King is back on his pathetic throne

[Ryan Eberhard @eberhard]

@cremieuxrecueil Good to be thriving but let’s keep the reins tight on the ego

[Ryan Eberhard @eberhard]

@KevinEspiritu Hey man this thread was just fantastic.

[Ryan Eberhard @eberhard]

@PhilipGreenspun @anaceballos_ You’re so proud of Florida’s Alligator Alcatraz that you’d like to see it expand into a network of 100+ facilities where immigrants either beg for return or die in squalor?

[Ryan Eberhard @eberhard]

@hayasaka_aryan @PhilipGreenspun Don't let them talk about you like this. Stand up for yourself.