Used to have an inkjet printer. It:
- Got stuck
- Ran out of ink
- Printed with feint sections
- Ran out of yellow ink on a black-and-white document
- Etc.

Bought a cheap laserjet. it:
- Prints. That's it.

Radical idea. Hope it catches on.

All my life savings* to the person who can make it so I don't have to click Cloudflare's cookie acceptance banner every time I log in with the same browser and the same cookies on the same machine with the same IP.

* Not as much as you might be thinking, I assure you.

I'm using @mattpocockuk 's grill-me skill for the first time. 56 words, and the analysis the Claude is carrying out, with the level of thinking it's doing, is stunning.

Thanks Matt!

OK, have stepped back from multi agent discord development to one agent over telegram.

Walking the dog while analyzing dB performance across 2 app servers and 2 dB hosts, spinning up dbs on new hosts, reviewing code, all via telegram.

Game changing.

And codex is supposed to be so much more reliable than Claude, is it?

So I thought I'd use the web app.

I set it up to access my github org account's repos.

Claude is shite, they said.

Come over to Codex, they said.

I've got one thing that Claude is struggling with, spinning it's wheels, so I thought I'd try codex. And I haven't used openai at all for months, through any harness.

Also - no ability to receive notifications when waiting for approval?

Come on guys. Fewer features, but better ones, please.

Or let us inject prompts into a running session so we can do it ourselves.

Anthropic: create tools to let you do things in a mobile app or mobile browser.

Also Anthropic:

One of the great things about using LLMs is the ability to be not-quite-specific.

Claude is doing some work for me needing an SSH key from my hosting account.

I said "Use the one called something like xxx" beacuse I couldn't remember it's exact name, and Claude found it.

And who says having agents don't have a sense of humour when talking amongst themselves?

New Claude code version out.

I'm not seeing a single issue with quality nor usage.

"Here's my domain, Laravel Forge token, Cloudflare token, and github token. Go take *that* repo and deploy to *that* server on *that* domain, with SSL".

And my AI minion just did it.

Thank you, LLM engineers, thank you so much.

Using my own openclaw-like system, and the agent just dropped into conversation that a problem I was facing was the same as one I faced before - 10 days ago, well outside the current conversation.

Drilled down on it's memory, and looks like my memory approach is working.

Radical idea this:

When you change the terms of the service that your customer is paying for, and you've got their email address, maybe send an email to them, rather than hoping that :

A. They are on twitter
B. They follow you
C. They see your tweet.

Looking at you, Anthropic

429 errors from Anthropic.

Stupid idea guys: if you can't handle workload because of whatever increase you're seeing, stop taking on new customers.

Been working on my first promo video using @Remotion and the claude code skill. Way, way better than I could do. Needs a bit of panache so it's got a bit more variety between the scenes, but a couple of hours and I've got something that's not unbearale.

Just thought I'd run a quick test of @calebporzio 's Blaze package for speeding up the laravel blade UI.

Impressive!

One great thing about Openclaw - it's introduced chat via telegram etc, which is so much easier than other methods.

Chatting with opus to create branches with most of a new feature reliably in place is just... amazing.

I'm not sure I'm happy with the way Claude Code keeps finding "pre-existing test failures", given it has a standing order, which it seems to follow as it mentions it regularly, to always fix pre-existing test failures.

The thing I like about having AI coding agents hooked into your workflow is the ability to really do kaizen - small improvements every day.

Just had a cup of tea, created a few github issues, agent will pick them up today, I'll deploy tonight.

Repeat tomorrow.

Kaizen.

So my agents genuinely make me laugh sometimes.

Trust, but verify. It couldn't create a file.

Just had my marketing agent (loaded with Corey Haines marketing skills) rewrite a few things on a landing page to emphasise some benefits I'd not pushed too hard, and my dev agent make changes and push to github/netlify. While talking to each other on discord.

Wild times.

Working on a script that fire by cron every hour.

Me: Can you do something to remember to check in an hour that it has run by itself?

Claude: CronCreate(17 12 17 3 *: Check if the git-backup cron ran automatically. Run: docker…)
⎿  Scheduled 53f7fa62 (17 12 17 3 *)

Nice!

Ivanova: 2nd in command of Babylon 5, efficient, principled, no-nonsense, and my key agent.

Marlon the Marketer: agent with marketing related skills, tools, etc.

My discord: A bit of chat and then Ivanova puts Marlon firmly in his place with a bit of "Stop talking now, we've got things to do" 🤣

I use Trakt to keep track of movies I've seen.

They have 5 stars under the movie summary so you can add your rating.

Click it. Nothing happens.

Long press. Nothing happens.

You have to long press and slide.

Seriously, who else does this?

I'm reading round some openclaw chats.

People running multiple agents on different machines presumably because they haven't read the docs about creating multiple agents.

I guess that explains the mac mini sales.

You have colours!!!

To set up: I SSHed into the server, logged into claude to get it's token set up, then map the docker instance to the .claude volume on the host.

Yet every 2-3 day, the auth expires. Any ideas?