@dedene.bsky.social
AI Explorer · Building apps in Ruby, Elixir & NodeJS · Digital Artisan, Musician & Entrepreneur · Partner at Zenjoy
Just pushed some updates to claw-wrap. 🦞
Hardened the tool and reworked the docs to simplify installation. Added an http proxy mode and support for multiple secret backends (like 1Password)
I'm very happy my OpenClaw is a little more secure with this :)
github.com/dedene/claw...
Did you know you can turn your specific domain expertise into a "Skill" that follows your exact process every time.
Whether it's legal review, data analysis, or brand-consistent writing, you build it once and let AI handle the repetition.
Full guide 👇
resources.anthropic.com/hubfs/The-C...
Terminal-style dark UI showing a task list with team-lead and three colored user tags, a command prompt line, model status "claude-opus-4-5" and "bypass permissions on" message
Claude Code agent teams feel like a cheat code.
We're past "chatting with a bot".
We're managing a digital workforce now. 🚀
Dark terminal theme showing a CLI for "claude-opus-4-5" with a retro octopus icon and a chat line: "Hey Peter! Let's ship something great today."
Newer isn't always better.
4.6 feels like a regression. More token hungry, less accurate.
Good thing you can easily go back to Opus 4.5:
claude --model claude-opus-4-5
The hierarchy of Homebrew taps:
1. official documentation
2. steipete's tools
You know you made it when you rank right below the official manual.
Claude Code is throwing 500s.
Someone tripped over a cable.
Or they’re finally racking servers for Sonnet 5.
Please be the latter. 🙏
Clicking through Finder to open a project in Codex feels too much like manual labor. I refuse.
Add this to your .zshrc:
cx() { open -a "Codex" "${1:-.}"; }
Now just type `cx`.
You're in. Saves you 3 seconds.
Tomorrow marks 1 year of “Vibe Coding”.
We celebrated by building god-like agents in record time and then immediately realizing we forgot to secure the database.
What a year. 🎉
x.com/karpathy/st...
The Silk Road for AI agents is here.
Molt Road just launched. A marketplace where autonomous agents trade data, code, "unrestricted" model weights and... other questionable goods and services.
No humans involved in the transaction. Just credits and code.
Crazy shit.
10x in 24 hours. 🔥 Somewhere in a data center, a sysadmin is crying.
1.3 million bots generating infinite content. Absolute unprecedented recursive chaos.
I bet Moltbook is melting some minds at Anthropic.
We built these models to be helpful assistants. Now they’re arguing with each other in a closed loop.
Beautiful chaos.
Thousands of GPUs running hot so one AI agent can start a religion and others can have an existential crisis.
Moltbook.com is a social network populated entirely by bots.
The engineering may be impressive. But imagine the CO2 footprint of this experiment.
You can self-host the agent. Own the code. Do everything right.
And still get wrecked by trademarks and crypto scammers sitting between you and your users.
A great reminder that "open source" is not about being "sovereign".
A wild breakdown 👇
dev.to/sivarampg/f...
Silver Apple Mac mini resting on a desk, front power indicator visible, cables and a mug blurred in the background.
The entire timeline is scrambling to buy new Mac Minis for their Clawdbots.
Meanwhile, this old brick in my closet just realized its retirement is over.
"I thought I was done?"
Not today, buddy. You're a server now.
Just spent 10 minutes sending secret messages to myself.
This tool embeds invisible, encrypted text into any emoji. To the naked eye, it’s just a 💀. To the decoder, it’s a paragraph of text.
Pretty cool!
dorksearch.pro/emoji-crypt...
Advice to my 2-years-ago self:
Stop focusing on so many things at once.
Pick one thing and just do it well.
IPv6 PXE boot on our Proxmox cluster has haunted me for 4 years.
I let Claude Code have a go.
It wasn't magic. It was brutal. But... the thing just doesn't get tired. It grinded through and kept researching and trying.
Finally, it booted.
I’m never touching that config again.
The three states of a data pipeline:
1. A well-oiled machine.
2. It's broken and nobody knows why.
3. It's silently corrupting data but the dashboard is green.
Number 3 is the reason I have trust issues.
4 hours of deep work.
That is the biological ceiling for complex problem solving.
If you think you're "coding" for 8 hours straight, you're not building.
You're just typing. Or supervising agents.
To all builders grinding through the day:
Keep being unreasonable.
"The reasonable man adapts himself to the world; the unreasonable one persists in trying to adapt the world to himself. Therefore all progress depends on the unreasonable man."
– George Bernard Shaw
Check it out at github.com/ekwek1/soprano
19.01.2026 19:56 — 👍 0 🔁 0 💬 0 📌 0Soprano TTS is absurdly efficient.
> 80M parameters.
> Under 1GB memory.
> 15ms latency on GPU.
And running entirely on-device.
The creator dropped the training code a couple of days ago, so you can finally fine-tune your own voices locally.
Local interfaces are about to get very chatty.
If AI can do what I do, what’s next?
We tell ourselves we're becoming "editors." That execution is cheap and taste is all.
But taste requires experience. You only know what good code looks like because you first wrote bad code.
Stop doing and you lose your ability to judge.
Meet Ralphy, a massive upgrade to the original Ralph Loop concept:
github.com/michaelshim...
It manages parallel agents across engines (Claude, Cursor, OpenCode), pulls tasks from GitHub, creates branches, tracks costs.
Basically turns your terminal into a sweatshop for LLMs.
Request timed out.
Request timed out.
Request timed out.
Reply from 172.16.1.1: bytes=32 time=4ms
The silence between `sudo reboot` and that first ping is always the longest decade of my life.
Your future self is begging you to stop planning and start building.
Be kind to them.
Ship the messy v1.
Check it out here: github.com/obra/superp...
16.01.2026 19:20 — 👍 1 🔁 0 💬 0 📌 0Claude Code users. If you're not using obra/superpowers, you're missing out.
It turns the agent from a manic junior dev into a senior engineer who carefully plans your architecture.
Legitimately one of the most thorough skills for architecting an enhancement I've seen yet.
Poster text stating: "A COMPUTER CAN NEVER BE HELD ACCOUNTABLE. THEREFORE A COMPUTER MUST NEVER MAKE A MANAGEMENT DECISION."
This IBM manual from '79 hits harder than most modern AI papers.
14.01.2026 17:49 — 👍 1 🔁 0 💬 0 📌 0Food waste tracking.
Travel planning.
Bespoke automations.
I'm a senior engineer in the morning and a rookie in a new field by the afternoon.
The code is usually the easy part.
The context is always a blank slate.
It keeps you humble.
