Gave an agent the tools I've been building and one idea (a little robot), then let it run the whole thing: generate it, drop it in a scene, make it walk, hand it a dance to copy.
Every frame generated. No camera, no footage, pretty fun watching the whole thing come together.
Spent the last couple weeks open-sourcing my AI media pipeline... eight small skills an agent runs end to end: generate an image, edit it, turn it to video, swap the character, watch the result, the works.
Some run on my GPU, some on a frontier API. Eight tools, one pipeline.
The boring truth about agents: better prompts only get you so far. Better tools change the ceiling. Otherwise they're like a printer intern without a printer.
Example from my setup: local MCP gateway for search, browser, docs, video, PDFs, and Whisper.
https://t.co/Hkrhy0s2bP
@ClaudeDevs This is the problem with treating frontier AI like normal SaaS.
You’re not just buying access to a model. You’re renting permission from a company that may be renting permission from Washington.
This is also why local/open models matter. Not because they’re always better... But because sometimes “worse but actually under your control” beats “best in class until a government memo says otherwise.”
Anthropic getting forced to kill access to Fable 5 is exactly the problem.
Everyone wants to pretend frontier AI is a normal SaaS business, until the US government can flip a switch and global customers just eat the outage.
Great IPO pitch: trust us, unless Washington says no.
As a result of a US government directive, we are suspending access to Claude Fable 5 for all users. You can continue to use all other Claude models.
Here’s what this means for you:
Across Claude products, new sessions will run on your selected default model or Opus 4.8, and existing Fable 5 sessions will end with an error.
On the Claude Platform, requests to Fable 5 will also return an error. Please update your integrations to other Claude models.
We know this is a disruption to your workflows; we appreciate your patience and support.
Sometimes I learn this stuff just to make my kids smile, like showing them dad flying spaceships or racing cars.
Same toolkit, broader work: software, automations, media. Starting to put more of it out here.
Made locally with LTX 2.3 + my own agentic workflow.
Been quiet here, but not inactive.
Most of my time has been going into software, local AI setups, homelab infra, media experiments, and small useful tools that mostly lived in terminals, repos, and private notes.