You gave your AI agent (OpenClaw, Hermes, Claude Code) a whole server. Why does email still depend on 3rd party? Introducing AIMX: a self-hosted, open-source email server designed for AI agents. How Postfix would look if it were built for AI agents. uzyn.com/2026/introducing-ai…

Waiting for my spell cooldown to hit 100% so I can cast it with "Reset for free". @claudeai @ClaudeDevs

This is **ONE** prompt on @claudeai Code on the Cloud. $100. $100. Just for the VPS "rental" for about 40 minutes, exclusive of LLM tokens! Just a few days ago, this was free. You just had to pay for Claude tokens.

Thanks @sketch for still championing the pay-once, keep-forever licensing practice today. My copy, which I still use semi-regularly, is 7 years old & 40 releases behind. I've upgraded twice, probably time for me to upgrade again for the 3rd time. Still my fav design software.

TISC 2026 by CSIT SG starts tomorrow night, runs to 20 Sep. A fun, extremely challenging CTF you can do from home. Like last year, I crafted one of the challenges, so no prizes for me. Share more after the contest. Sign up: csit.gov.sg/events/tisc/tisc…

Apple's amazing at sales. I hated the fact that Apple even explored fold phones. I thought it's dumb & gimmicky. After browsing the product page, for the first time I'm quite sold on the point of the fold. Let's see if Apple will convince me on the rumored touchscreen MacBook.

AI is perfect for ADHD individuals. But at the same time it’s also making it worse. Start one prompt, while waiting, 3 other prompts on separate projects just come up. It’s Hydra. Arrrgh!!!

You don't pay for gas to drive around to the gas station. You don't pay a carpenter to spend all day sharpening the saw. You don't rent a warehouse to store the shelving. Same for LLM tokens. Stop burning them to tune harnesses that burn more tokens. Build something!

Love that @claudeai is not shy at spawning agents.

A good day is when you have 6 long-running agents set up and running well before you head out for lunch appointment.

Claude gets smarter and token-hungrier , but our monthly sub allowance remains flat. @AnthropicAI can we have 100x Max plan at $500/month? Will happily pay for that. 5+ days till reset but I'm already maxxing out my weekly credits. I should touch grass for rest of the week?

You have nothing to worry about @Apple if this is what @OpenAI hardware team came up with: youtube.com/watch?v=m8uUUUsM… No better way to demonstrate how bad of an engineer are you with Codex Micro.

Claude has been pretty much the cool kid the past 6-12 months. With just 1 ad, @ChatGPTapp now turns it around. This is such a kickass ad. I'm all-in on @claudeai I do miss ChatGPT voice, even the prior version. youtube.com/watch?v=EAN5Cj34…

Conducting playtest on a CTF challenge: Told @claudeai specifically not to cheat, and set up a separate dir with just the challenge. Surprised that it solved it in 1-shot and wrote a report on how it solved it, outlining all the right methods. Thankfully tool calls are logged. Checked its tool calls, it cheated and wandered looking for files, i.e. solutions and hints outside of the workdir; and not mentioning it at all in the report! Tips: check Claude Code logs at `~/.claude/projects` if you need to conduct forensic work on misbehaving Claude.

Me: "Hi, how are you?" Fable 5 with Ultracode: - spawn 10 agents - 1 hour and 1M token later "I am good, you?"

My regular go-to GPU rental service for ad-hoc ML fine-tuning is vast.ai. It's a marketplace of spare GPU resources, great for non-regular rental. Interestingly, for the first time yesterday I encountered a host that appeared to be from the US, but it was actually not. It felt a little slow when I was on it. Turns out it was actually a Hong Kong host, proxying vast.ai traffic from a US. Chinese hosts are usually cheaper and initially avoided due to very limited international bandwidth. Now it's worse as AI services like @claudeai actively block access from Chinese IPs, which makes the workaround not just slow but quietly unusable for real workloads.