Finally got to watch, The AI Doc: Or How I Became an Apocaloptimist.
Self improving AI models and robotics are the biggest risks, shortly followed by core infrastructure being hacked and disrupted by AI.
I'm not talking about direct AI access either, but specifically written malware/viruses that attack air gapped systems based on what operating plans it can find by hacking other companies.
What's everyone's thoughts on this? The risk is real, how much risk, I don't know.
imdb.com/title/tt39150120/
40
Remember the old Sky Digiboxes?
Claude just cracked (3 days of reverse engineering and trial and error) running the real dumped Pace 2500N firmware in a custom built MIPs emulator, in Javascript... Now I can start building a proper retro TV project, with EPG, old adverts, etc.
I'll whack it up on Github when it's polished and ready to run on a Raspberry Pi.
47
I've been using Obsidian as my agent memory for nearly 12 months now.
And by agent, I have an agent prompt for Claude, Codex, Grok and offline local models each pointing to the the Obsidian Vault, via a custom built MCP that hooks into a self hosted Couch DB server.
On top of this, if I'm working with code I use 'beads' github.com/gastownhall/beads which creates its own dependency graph, specifically for that project. Each task being atomically claimed by the agent or sub-agent that decides to pick it up, ensure dependencies are followed.
This then all sits on top of my AI framework, with a curated list of skills, workflows, conformance gates, and over 20 years of development experience fed into it.
I plan on building all of this into makeryard.app/ over the coming months, right now you can build apps in php, go, python, and node using AI over MCP, host preview and live versioned URLs, integrated database support and custom domain domains.
Join the wait list if you want to try it out for free whilst we build together - this is the AI Agent Harness that takes vibe coding to the next level.
Anyone else noticed how Claude usage is being used up slower since the Astra release? Just me? #anthrophic #claude #ai #agents
1
46
One of the collest things I've ever seen. A Falcon 9 launch from Cape Canaveral at 5:30am this morning!
Dan Dunford retweeted
A few patterns we frequently use with Fable 5:
Use Fable 5 as an "advisor."
An executor (Sonnet 5) calls Fable 5 for guidance.
Most tokens are billed at the lower executor rate.
Check out my latest article: Which LLM Should You Actually Use? linkedin.com/pulse/which-llm… via @LinkedIn
Used Claude to rebuild what AthenaScope did 8 years ago. In a night.
OCR, frame CV, LLM vision, audio triggers.
1
41
Today is Employee Appreciation Day.
So I want to appreciate Claude.
I know - Claude isn't an employee.
Claude doesn't have feelings, doesn't clock in, doesn't need a shoutout. But Claude has become a genuine part of how I work, and if I'm being honest, it'd feel weird not to acknowledge that today.
Over the past year-ish, Claude has helped me think through architecture decisions at midnight, written first drafts I actually wanted to build on, debugged things I'd stared at too long, and pushed back when my ideas were half-baked. Not in an annoying way. In the way a good collaborator does.
It's not magic. It's just a really good tool - one that rewards clarity, punishes laziness, and gets better the more precisely you use it.
So. Happy Employee Appreciation Day, @claudeai . You've earned absolutely nothing, because you're an AI. But the work has been solid.
4
3
54
14 years of Netflix, cancelled this month, so I made a 'Netflix wrapped' with Claude Code with the exported data. I'm gonna miss that fire log burner...
55
Dan Dunford retweeted
Announcing a new Claude Code feature: Remote Control. It's rolling out now to Max users in research preview. Try it with /remote-control
Start local sessions from the terminal, then continue them from your phone. Take a walk, see the sun, walk your dog without losing your flow.
Dan Dunford retweeted
we desperately need a new season of silicon valley. the ai era alone would carry 3 seasons
"Make a task list and use an agent team to complete" is probably one of the simplest and most powerful prompts you can tell Claude Code right now.
It literally spawns you an entire development team, uses agents and skills, to complete all tasks you give it.
🧠💥🤯
#claudecode #anthropic #aicoding #ai
38
Dan Dunford retweeted
Nothing humbles you like telling your OpenClaw “confirm before acting” and watching it speedrun deleting your inbox. I couldn’t stop it from my phone. I had to RUN to my Mac mini like I was defusing a bomb.
Dan Dunford retweeted
We’ve identified industrial-scale distillation attacks on our models by DeepSeek, Moonshot AI, and MiniMax.
These labs created over 24,000 fraudulent accounts and generated over 16 million exchanges with Claude, extracting its capabilities to train and improve their own models.
I've been testing Claude Agent Teams recently, and wow it is something else.
In agent teams, teammates share a task list, claim work, and communicate directly with each other.
Once you have the right balance of agents, skills, and modules (prefabs) tuned, you can throw it a plan (of phases) and it'll go create you a team (of agents) and they will all automously go and build it until complete.
No fancy wigum loops required!
It just built me an entire meeting hub that syncs calendars, contacts, emails into one unified API.. with SDK support and drop in react components and full OpenAPi docs
meeting.demosrv.uk/
Obviously it needs full testing and polishing, but damn, it's so easy to build software now.
Go check it out code.claude.com/docs/en/agen…
1
1
28

















