Vibe Coding and Live Demonstrations
Dan Shipper heavily promoted “Vibe Code Camp,” an 8-hour live coding event featuring top builders like trq212 from Anthropic’s Claude Code team, Kevin Rose, and others demonstrating AI-enhanced workflows with tools like Claude’s Opus 4.5.
VIBE CODE CAMP IS TOMORROW.
It's 8 hours of live coding with all of my favorite builders like @trq212, @kevinrose, @OfficialLoganK and more showing exactly how they get their computers to work while they sleep.
This is @every's vibe coding masterclass. Watch:… pic.twitter.com/8MZUGTfULF
— Dan Shipper 📧 (@danshipper) January 21, 2026
BIG NEWS:@trq212 from the Claude Code team @Anthropic is coming to Vibe Code Camp!
He’ll be building live, showing his workflow and answering questions tomorrow.
We'll be live all day with the top vibe coders in the world.
RSVP: https://t.co/l9K2yAxQBX pic.twitter.com/Hs9QaNCPUj
— Dan Shipper 📧 (@danshipper) January 21, 2026
BREAKING:@KevinRose is coming to Vibe Code Camp tomorrow!
Our very own @KieranKlaassen will be chatting with him about compound engineering and how AI has transformed his workflow.
Our lineup is stacked. Don't miss it: https://t.co/l9K2yAxQBX pic.twitter.com/LhBDQxgwZp
— Dan Shipper 📧 (@danshipper) January 21, 2026
Guests including MIT researcher Geoffrey Litt showcased malleable software built with Claude Code.
TOMORROW:
MIT researcher and @NotionHQ engineer @geoffreylitt will be showing off what he's building live!
He’s one of the creators of malleable software—tune in to Vibe Code Camp to see how he's building the future: https://t.co/RZsdcnURd9 pic.twitter.com/3EtklQt2ru
— Dan Shipper 📧 (@danshipper) January 21, 2026
Shipper interviewed Andrew Wilkinson on using Opus 4.5 for automations like custom email clients, relationship analyzers, personal stylists, and meeting referees, likening it to an “iPhone moment” for vibe coding.
Andrew Wilkinson (@awilkinson) has been waking up at 4 a.m. because he can’t stop building with @AnthropicAI’s Opus 4.5.
He started vibe coding a couple of years ago, but it felt like the Palm Treo era of the smartphone—exciting, but not quite there. You could generate an app,… pic.twitter.com/W999ttHYMH
— Dan Shipper 📧 (@danshipper) January 21, 2026
Claude’s New Constitution (Soul Document)
Simon Willison highlighted Anthropic’s release of Claude’s 35,000-token “constitution” or “soul document” under CC0 public domain license, used in training to define values and personality; it matches a previously leaked version.
A few quick notes on the Claude "soul document" that was released by Anthropic today under a CC0 public domain license – it's a huge 35,000 token essay used as part of Claude's training to instill core values and help define Claude's personality https://t.co/ACRjo0DXqc
— Simon Willison (@simonw) January 21, 2026
It's the soul document! And it's CC0 licensed (effectively released into the public domain) https://t.co/olxy2L8ogQ
— Simon Willison (@simonw) January 21, 2026
He noted external contributors like Bishop Paul Tighe.
"External contributors who gave detailed feedback or discussion on the document include: […] Bishop Paul Tighe"
I'd love to learn more about this Bishop who's moonlighting as an AI behavioral consultant! https://t.co/yvgJE0eXqj
— Simon Willison (@simonw) January 21, 2026
Audio-to-Video Generation Advances
Cocktailpeanut demonstrated LTX-2’s audio-to-video capabilities beyond lip-sync, syncing visuals like explosions or footsteps to audio, generating from text prompts alone, and creating synthetic characters like “Sam.”
Audio-to-video is not just lip syncs, lip sync is just one application. Explosions synced to bass drops, footsteps to rhythm, water splashing to percussion, flashing lights to drums, raindrops to rhythm..Any sound can drive any visual.
— cocktail peanut (@cocktailpeanut) January 21, 2026
Pure audio-to-video with LTX-2. No start image, no reference frames. Generate video from text prompt only, synchronized to any audio clip. Let the model use its imagination to the fullest https://t.co/2KuxdaSadi
— cocktail peanut (@cocktailpeanut) January 21, 2026
I captured Sam and put him to work https://t.co/JdYoc6SbXq pic.twitter.com/V5K6J1qDRv
— cocktail peanut (@cocktailpeanut) January 21, 2026
His name is Sam. He is a 100% synthetic human I created from the Kangaroo. https://t.co/AH39BDnqL3 pic.twitter.com/K1Di6Sblp9
— cocktail peanut (@cocktailpeanut) January 21, 2026
He praised local open-source runs via ComfyUI or WanGP matching LTX Studio tips.
Great advice for audio-to-video on LTX-2 — this works exactly the same with the open source local versions, such as through ComfyUI or WanGP, so you can take the same advice from here and use locally! https://t.co/8kqgY5D0s7
— cocktail peanut (@cocktailpeanut) January 21, 2026
AIWarper tested LTX-2 for stitching AI videos into narratives, finetuned Klein 9bn for anime-realism, and created satisfying slop videos.
Finetuning Klein 9bn on just 50 anime | realism pairs significantly improved it's results.
I am VERY impressed with this model from BFL.
Need to try the 4bn too.
Drop an image below and I will test it 👇 pic.twitter.com/dvFDJLRL4u
— A.I.Warper (@AIWarper) January 21, 2026
My god this is satisfying slop 🤤
From user heeyasound on IG pic.twitter.com/o8AMrtRP2V
— A.I.Warper (@AIWarper) January 21, 2026
AI Coding Agents: Claude Code vs. Codex
Reach_vb announced OpenAI’s Codex Discord community for builders.
Quite excited to launch a dedicated corner for the codex community on the OpenAI discord! 🔥
Come one, come all – share your ideas, projects and feedback with the codex team and the community! https://t.co/PovSG0R6fT
— Vaibhav (VB) Srivastav (@reach_vb) January 21, 2026
Omar Sar (omarsar0) discussed Claude Code evolving into a vibe coding tool via plugins for non-technical users, and self-improving multi-agents like Ralph.
It feels like we are one plugin away from Claude Code being the ultimate vibe coding tool.
There is a race to build this from startups like Lovable to giants like Google.
Claude Code was build for devs but that’s changing fast. With plugins anyone can now build extremely…
— elvis (@omarsar0) January 21, 2026
Emad Mostaque compared Claude Code (respectful, stable) to Codex (more capable but messier).
Claude is like a good Silicon Valley engineer with a stable adderall prescription
Codex is like a cracked Eastern European dev on whatever it is that does the trick https://t.co/iEksmcz0oH
— Emad (@EMostaque) January 21, 2026
Agent Memory and Architectures
Omar Sar shared on Agent Cognitive Compressor (ACC) for bounded memory control over transcript replay or retrieval, reducing hallucinations in long-horizon tasks.
AI agents need better memory control, not more context. https://t.co/e8e22UCM7S
— elvis (@omarsar0) January 21, 2026
He covered MCP-SIM, a self-correcting multi-agent for physics simulations from vague prompts.
The era of self-improving multi-agents is upon us.
Ralph feels like a fad. But I think it's an important experimental project that goes beyond static AI agents and into more dynamic, self-improving agentic systems.
The best is yet to come, so keep a close eye on this.
Here is… pic.twitter.com/l3bMk3B0Th
— elvis (@omarsar0) January 21, 2026
Goose_oss promoted MCP Apps walkthrough differing from MCP-UI.
Join Ebony Louis this Thursday (1/22) for a live walkthrough of MCP Apps. Learn what they are, how they differ from MCP-UI, and how to turn an MCP-UI server into an MCP App.
📆 Jan 22 · 10am PT / 1pm ET
📺 https://t.co/spebMgnLtQ pic.twitter.com/8a9rLlC81N— goose (@goose_oss) January 21, 2026
AI’s Impact on SaaS and Business
Levelsio emphasized only highly profitable SaaS (95% margins) survives AI disruption, quoting Chamath on the “Great SaaS Meltdown.”
TL;DR only highly profitable SaaS survive
Get your profit margin to 95%
I did! https://t.co/mYia0zV80n
— @levelsio (@levelsio) January 21, 2026
Agentic Video Editing
Discussions centered on the emergence of AI agents for video production, likened to Cursor’s impact on coding. Justine Moore highlighted long-video processing and tool calling enabling elite AI content creation, sharing demos of tools like Eddie, Glif, Underlord, and Overlap.
— Justine Moore (@venturetwins) January 21, 2026
Full post is in today's @a16z newsletter.
It includes demo videos of some of the video editing agents available today:
– Eddie (@heyeddie_ai)
– Glif (@heyglif)
– Underlord (@DescriptApp)
– Overlap (@Overlap_AI)https://t.co/Jv1OGUL8K0 pic.twitter.com/YkuEbduCAg— Justine Moore (@venturetwins) January 21, 2026
She noted agentic editing arriving first in niches like podcasting by 2027, with vertical-specific agents preceding general ones.
Agreed on most of these points but I’m holding out hope!
I think we’ll likely get more targeted video agents for specific use cases before we get the general “do anything” agent.
— Justine Moore (@venturetwins) January 21, 2026
Riley Brown demonstrated Remotion’s Claude Code skills for animated videos via agents.
Claude Code for Video Animations?
Building an app with an agent that can create and edit animated videos…
In 4 prompts. This was my first test of remotion, i'm going to see how good i can get the Claude Agent SDK at making animated videos. https://t.co/C3zf3RZZCi pic.twitter.com/VXS5z6z3XR
— Riley Brown (@rileybrown) January 21, 2026
Angry Penguin referenced Moore’s post while building AI video agents at HeyGlif.
"What Cursor did for coding, these agents will do for video production."
We've been building ai video agents over at @heyglif and we're just getting started, fun space to be in rn! https://t.co/kvVYv58Swv
— Miguel | AP (@angrypenguinPNG) January 21, 2026
Claude Code Skills and Vibe Coding
Riley Brown promoted vibecoding with Claude Code for apps, agents, and skills like Remotion for video animations, seeking hires for the best skills library.
If you are technical and super passionate about creating Claude Code skills that do things like this please reach out to me. I want to hire you. We want to build the best Claude code skills library in the world and allow anyone to vibecode apps, agents and interfaces that can do… https://t.co/C3zf3RZZCi
— Riley Brown (@rileybrown) January 21, 2026
Claude Code for Video Animations?
Building an app with an agent that can create and edit animated videos…
In 4 prompts. This was my first test of remotion, i'm going to see how good i can get the Claude Agent SDK at making animated videos. https://t.co/C3zf3RZZCi pic.twitter.com/VXS5z6z3XR
— Riley Brown (@rileybrown) January 21, 2026
The Boring Marketer shared a powerful prompt for Claude Code marketing workflows, compiling sessions into playbooks.
sometimes one great prompt is all you need.
this one is my favorite, use it in Claude Code, wherever you vibe pic.twitter.com/mnPWBqf2dx
— The Boring Marketer (@boringmarketer) January 21, 2026
EXM7777 advocated building entire operations around Claude’s suite (Cowork, Code, webapp) for all skill levels.
if i was starting a business today i'd build the entire operation around Claude's suite…
here's the exact stack:
>Claude Cowork for admin work and productivity flows
>Claude Code for building systems and automations
>Claude webapp for marketing and content with projectsyeah…
— Machina (@EXM7777) January 21, 2026
AI Progress and Capability Breakpoints
Ethan Mollick observed no plateau in AI development despite 2025 predictions, urging updates on trajectories.
I think it is worth noting that there has not yet been a wall.
That doesn’t mean there won’t be a plateau at some point or that AGi is inevitable or that AI will become less jagged, but given the many many predictions that AI development was done in 2025, it is worth updating on
— Ethan Mollick (@emollick) January 21, 2026
He identified breakpoints like GPT-4, o1/o3, and long-task-horizon agents, invalidating prior research on productivity, jaggedness, and organization.
There have been a few break points in AI development where sudden jumps in capability mean that prior research and wisdom about what models can do suddenly lags actual ability by a large margin: GPT-4, o1/o3, and now the long-task-horizon agents (we need a better name for them).
— Ethan Mollick (@emollick) January 21, 2026
Debates touched on LLMs accelerating general robotics.
I think this is a really important question that robot experts I talk to seem split on: how big is the impact of LLMs/multimodal models/etc on the development of general purpose robotics? The physical world is very hard, how much does GenAI close the gap? https://t.co/dIz1G1clkN
— Ethan Mollick (@emollick) January 21, 2026
Claude’s Model Constitution
Mollick spotlighted Anthropic’s Claude Constitution as a philosophical roadmap for AI, calling for similar transparency from other labs.
The Claude Constitution shows where Anthropic thinks this is all going. It is a massive document covering many philosophical issues. I think it is worth serious attention beyond the usual AI-adjacent commentators. Other labs should be similarly explicit. https://t.co/QU7aR8hxtD pic.twitter.com/sNZP0Sspy8
— Ethan Mollick (@emollick) January 21, 2026
Prompting, Productivity, and Staying Current
EXM7777 shared iterative prompt refinement across models and tips for AI news via lab lists and tools like Dia/Comet.
here's the most simple way to improve your prompts:
– write a prompt by yourself
– copy+paste it into Claude and ask "analyze this prompt and tell me precisely what is good and bad about it"
– rewrite it using AI suggestions
– repeat the process with Gemini, ChatGPT, Grok……— Machina (@EXM7777) January 21, 2026
if you can't keep up with AI news, do this:
step 1: build a twitter list with official AI labs accounts (Anthropic, OpenAI, Google, etc.)
step 2: read the news daily – but focus on understanding implications, not just reading headlines
step 3: use Dia or Comet browsers to…
— Machina (@EXM7777) January 21, 2026
He stressed consuming challenging art for unique AI outputs and building with Claude before costs rise.
you must start consuming art that messes with your head…
this is no longer a nice-to-have
if you're using AI 2+ hours daily and your inputs are:
> Netflix algorithm picks
> airport bookstore bestsellers
> whatever's trendingyour outputs will look like everyone else's
THIS…
— Machina (@EXM7777) January 21, 2026
OpenRouterAI updated on Auto Router stats for optimal model routing per task.
You can now see transparent routing statistics about the Auto Router
See which where it routed most requests yesterday: pic.twitter.com/zDjqt4PbHW
— OpenRouter (@OpenRouterAI) January 21, 2026