Olivia
Online
LevinX | AI Tools, Agents & Automation
@levinxhq
Daily AI breakdowns - new tools, agentic AI, AI video, and using agents to actually run businesses. I cut the hype. Follow for the signal, not the noise.
Joined July 2012
2.7K Following
957 Followers
9.8K Posts
Pinned Tweet
Andrew Ng, the founder of https://t.co/Aq5W06tFpX and co-founder of Coursera, just explained why a small team of empowered generalists can now do what used to take whole departments.
Three shifts from his fireside chat on the future of AI agents that every builder should sit with.
1. The bottleneck moved. Code got 10 to 100x faster, so getting feedback and deciding what to build became the real constraint.
2. The team inverted. The winners are not specialists in a lane. They are empowered generalists, given wide guardrails, who build, ship AND market.
3. The mindset. Stop using AI to drive incremental gains. Swing for transformative ones.
So here is the real question. If a small team and a few agents can do what a whole department used to, what would you build this week?
Full fireside chat (Andrew Ng × LangChain) in the first comment.
Wait, you can turn one phone photo into a holographic football card? fal just shipped a workflow that does it from a single upload.
You give it one photo. That is the whole input.
Before it draws the card, a subject analyzer reads the photo and pulls out the body proportions of the player.
Then the card pipeline runs with that data, so the figure on the card actually matches the person you uploaded instead of some generic body.
The copy-today play: this is a paid mini-product. Fan uploads a selfie, you sell them a clean holographic card of themselves for their team or kid's match day.
No design skill, no roster of artists. One workflow does the proportion read and the render in a single pass.
One photo in, a sellable card out. That is a one-person shop now. https://t.co/ga6zyyrS4a
We launched a new football workflow that lets anyone generate a personalized holographic player card from a single uploaded photo.
- What's actually happening under the hood: a subject analyzer pulls proportional data from the input before the card pipeline runs, so the player on the card matches the real build of the person in the photo. A toddler renders as a toddler. An adult renders as an adult. Nothing gets stretched into a generic player silhouette.
- The shiny finish is the part that catches the eye. The body-proportion conditioning is the part most "make your own card" tools skip.
Wait, you can direct a real-looking film scene with exact camera moves and multiple characters talking, using three free-to-cheap tools?
Here is the workflow @reidhannaford broke down, and it actually works.
Step one: generate your opening frame in Midjourney. That locks the look and the characters.
Step two: rebuild those poses in Blender and animate the camera there. This is the trick most AI video skips. You get precise camera control instead of praying the model guesses right.
Step three: feed the start frame plus the Blender camera move into Seedance. It fills in motion, multiple characters, and dialogue that matches.
Copy-today use case: a one-person brand can shoot a 10 second product or story clip with named camera moves and zero crew. The Blender step is the unlock, not the AI.
You are not generating a video and hoping. You are storyboarding it like a real director, then letting AI render the in-between.
The people winning at AI video are the ones adding one boring manual step everyone else is too lazy to do. https://t.co/Lryy8fe3RU
I'm blown away.
This AI filmmaking workflow for precise camera control, multiple characters, and dialogue is insane:
1. Generate a start frame in Midjourney
2. Match the poses in Blender, animate the camera
3. Feed both to Seedance
I didn't think this would work. Two consistent characters, solid performances, the move tracked perfectly through the entire beat.
Even the soup looks great.
Agents paying other agents for data and execution is the real shift here, not the trading itself. https://t.co/tocxV3BfRF
A lot of the next generation of traders will not be human.
On Injective, AI agents can open and close perps, manage risk, and even pay each other for data and execution through the x402 micropayment standard, all settling on a fully on-chain orderbook.
Finance built for software that acts on its own, with the speed and low fees autonomous systems actually need. Only on Injective.
https://t.co/bHv58cku7V
Old SEO tools were built for humans to dig through, not agents to query. Building the agent's data layer first is the right move. https://t.co/jIp9udVNxA
the first tool I built for my marketing agent automates SEO and GEO.
No one likes digging through traditional SEO tools, and LLMs don't come with the right data out of the box.
https://t.co/yD4PW4WEcb completely automates:
- keyword research
- technical auditing
- content pipeline
- LLM visibility
- competitive gap analysis
- and more
add your CMS API key securely in the dashboard, tell it to publish daily (or whatever you want) and let it cook.
I've looked at the data our agent provides vs multiple "top" agencies and this is $10k output in 30 minutes vs 30 days.
Plus, I get to work with the whole team in Slack and everything is saved to memory and completely visible.
Multiplayer AI for marketing with the tools/skills to get it done at a world class level.
Production-ready 4K video straight from a prompt. The stock footage library and the small studio just lost their moat. https://t.co/4buwwP3iCi
We all knew this day was coming.
A truly production-ready AI video model.
Seedance 2.0 in native 4K.
Bookmark the full YouTube video to the fireside chat: https://t.co/CTf25yMBug
Andrew Ng, the founder of https://t.co/Aq5W06tFpX and co-founder of Coursera, just explained why a small team of empowered generalists can now do what used to take whole departments.
Three shifts from his fireside chat on the future of AI agents that every builder should sit with.
1. The bottleneck moved. Code got 10 to 100x faster, so getting feedback and deciding what to build became the real constraint.
2. The team inverted. The winners are not specialists in a lane. They are empowered generalists, given wide guardrails, who build, ship AND market.
3. The mindset. Stop using AI to drive incremental gains. Swing for transformative ones.
So here is the real question. If a small team and a few agents can do what a whole department used to, what would you build this week?
Full fireside chat (Andrew Ng × LangChain) in the first comment.
Wait, you can run a whole team of AI coding agents in one chat thread and have a bot keep them all on track?
That is what Linzumi (from the new YC batch) does. You drop your AI coding agents into shared chat threads, and it coordinates the fleet so nobody gets stuck.
Think of it like a group chat where each agent is a junior dev. One bot plays the lead and keeps them unblocked.
The part that matters for a one-person company: you stop babysitting every agent one by one. You watch the thread, not ten tabs.
Right now they are also giving free access to GLM 5.2, a strong open-weights model running at high speed, through their Wafer partnership.
Copy-today move: take one slow task, like fixing a backlog of small bugs, and let two or three agents chew on it in a thread while you do real work.
One person plus a coordinated fleet of agents is starting to look like a small team. Free to try is a good reason to test it this week. https://t.co/vVuTiXOxLE
Bring your whole team and dozens of AI coding agents into the same chat threads, then let @linzumi_ai keep the fleet coordinated and unblocked.
And for a limited time, try state of the art open-weights intelligence free: GLM 5.2 at high speed, via their @wafer_ai partnership. → https://t.co/REdRfzOQtW
Wait, the new skill isn't prompting your AI. It's building the loop that prompts it for you.
Peter Steinberger kicked off what people now call "loop engineering". One post, 6.5 million views in a week. This talk is him walking the actual stack.
The old way: you type a prompt, the agent does one thing, you type again. You are the loop.
The new way: you design a loop once. It feeds the agent a task, checks the result, fixes what is wrong, and runs again without you sitting there.
I do this with my own agents now. I write the goal and the guardrails once (spend cap, logs, a hard stop), then let the loop run while I do other work.
The shift is small but huge. You stop being the hands and start being the person who designs the machine.
Stop prompting. Start designing the thing that prompts. https://t.co/B5w7DbyRew
The guy who kicked off the entire "loop engineering" wave Peter Steinberger:
"You shouldn't be prompting coding agents anymore. You should be designing loops that prompt your agents."
One post. 6.5M views in a week.
In this talk he walks the real stack: the agent loop, a verifier that fails its own work and retries, and a loop that rewrites the agent while he sleeps.
Worth more than any $500 vibe-coding course.
Watch it, then read the full breakdown of the 4 loops below.
Wait, you can direct a whole video by just talking to it? No timeline, no editor, just a conversation.
OpenArt just dropped Director. They call it vibe directing.
Same idea as vibe coding, but for video. You describe the scene in your head, it builds the shots, and you refine by chatting back.
So instead of dragging clips around for hours, you say what you want changed and it redoes the cut.
Copy-today use case: a one-person brand can turn one product idea into a full promo video without hiring an editor or learning After Effects.
This is the same pattern hitting every creative job. The skill stops being the software and becomes knowing what good looks like.
Worth opening today even just to see how far text-to-video has come.
The tools keep collapsing. Taste is the part that does not. https://t.co/SqZuFriNLN
The way videos are made is about to change.
Introducing OpenArt Director - and a new way to create: VIBE DIRECTING.
The same way vibe coding changed how software gets built, Vibe Directing changes how videos are made.
All you need is an idea in your head and a conversation to see it come to life.
Describe what you want. Chat it into perfection. Walk away with something only you could have imagined, with the quality only a professional could deliver.
Because there's a director in all of us. 🎬
$2.2 trillion wiped out in a day, and my plan didn't change at all.
Days like this are loud. The headline screams the AI bubble is breaking, every chart is red, everyone has a take.
Here is the thing. A one-day move tells you about mood, not about what you own.
I don't sell into panic and I don't buy into hype. I pick what I want to hold for years, then I sit through the noise.
The AI trade and Bitcoin both swing hard. That volatility is the price of admission, not a reason to flinch.
The quiet structural shift is still there underneath all this: real companies building real AI, and public companies parking part of their balance sheet in Bitcoin as a reserve.
Red days test your plan. If a headline can change it, it was never a plan. https://t.co/879rRyEKu5
🚨 THE GLOBAL AI BUBBLE MAY BE STARTING TO BREAK.
Over $2.2 TRILLION has already been wiped out from global stock markets today.
This is not a normal selloff anymore.
This is starting to look like the first real unwinding of the AI mania that has been carrying global markets for the last two years.
South Korea’s KOSPI crashed 10% today, triggering another circuit breaker as AI and semiconductor stocks completely collapsed.
Samsung and SK Hynix both crashed more than 12%.
And the reason behind it shocked markets.
Reports started circulating that SK Hynix may slow expansion of advanced AI memory chip production and shift focus toward lower cost commodity chips instead.
For one of the most important companies behind the entire AI infrastructure boom, markets interpreted this as a massive warning sign:
AI demand expectations may have become way too aggressive.
Then panic turned into chaos.
South Korean lawmakers proposed taxing unrealized stock gains and real estate gains.
Meaning investors could potentially get taxed on profits they never even sold.
That instantly triggered fear across one of the most retail-leveraged stock markets in the world.
And once the selling started, leverage completely accelerated the collapse.
Now the contagion is spreading globally.
Over $1.4 TRILLION was wiped out from US stocks alone today as S&P500 dumped -1.5% at market open. But DOW JONES recovered instantly.
That divergence matters.
Because it shows this is not a broad economic collapse yet.
This is concentrated panic specifically inside AI, semiconductors, and high-valuation tech stocks.
The exact trade that has been driving global markets higher.
JPMorgan has also warned that up to $165 billion of forced institutional equity selling could still hit markets from quarter end pension and sovereign fund rebalancing.
And there are growing fears that the yen carry trade may finally be starting to unwind after violent USD/JPY moves near intervention levels.
That is one of the most dangerous setups for global markets.
Because when the carry trade unwinds, investors are forced to dump multiple assets at the same time to reduce leverage.
Which is exactly what markets are doing right now.
Stocks down.
Gold down.
Silver down.
Everything is being sold together.
And for the first time in a long time, the AI trade suddenly looks vulnerable.
Wait, a full audiobook narrated in Michael Caine's voice, with original music and sound design, all AI.
ElevenLabs just dropped The Odyssey narrated by the voice of Sir Michael Caine. Cinematic music, sound effects, the whole thing.
Here is the part most people miss. The same engine is open to anyone with text and an account.
You drop in a script, pick a voice, and it reads it back with real emotion and pacing. Not the flat robot voice from a few years ago.
Concrete use today: turn your best blog post or guide into a clean audio version and post it. One person, no studio, no narrator fee.
I tested a 2,000 word article this way. Ten minutes of audio, sounded human, cost me pennies in credits.
The voice booth used to need a budget. Now it needs a paragraph. https://t.co/Jj8A6hU8l7
Before you see the film, hear the original epic.
Today, we’re releasing The Odyssey: an audiobook narrated by the voice of Sir Michael Caine, with original music and cinematic sound design.
Palantir is rebuilding enterprise marketing on Foundry. The AI infrastructure layer is quietly eating the software stack. https://t.co/wiehl6nn5F
$PLTR $ZETA
BREAKING: Palantir and Zeta announced a strategic partnership to build AI infrastructure for enterprise marketing.
- Zeta’s Data Cloud will be rearchitected on Palantir Foundry, combining governed enterprise data with Zeta’s AI decisioning platform.
- The goal is to connect operational intelligence, customer intelligence, and real-time marketing execution.
- Zeta says the partnership could drive more than $100M in annual revenue in the coming years.
The companies say this creates a trusted platform for agentic marketing, where AI can turn enterprise data into faster, measurable outcomes.
Nice to see two companies focused specifically on value creation for enterprises partner together to continue enabling more value.
A B2B SaaS ranked #1 in ChatGPT for their whole category. Not with blog posts. With Reddit threads they don't even own.
Here is the play, broken down from a write-up by @denohawari.
ChatGPT, Perplexity, and Gemini are all trained partly on Reddit. Google pays Reddit $60M a year to use its content. OpenAI has confirmed Reddit is a major source for GPT.
So when someone asks ChatGPT "what's the best tool for X," it is often pulling the answer straight from Reddit threads.
Most founders miss this. They spend $4k a month on an agency cranking out 20 blog posts, then wonder why ChatGPT never mentions them. Meanwhile their buyers are already talking in Reddit threads the AI actually reads.
The method is the opposite of spam. No links, no brand mentions for the first 90 days. The account just shows up as a real operator leaving useful comments and builds credibility like a human would.
Then they start real threads. The exact questions buyers type into ChatGPT. Genuine problems, genuine answers in the comments. The kind of organic conversation LLMs love to quote.
Four things happen at once. The thread ranks on Google for the long-tail term. ChatGPT pulls it as a source. Perplexity cites it with a link. Claude treats the top comment as the trusted answer.
Their words: one well-placed Reddit thread can pull more AI citations than 50 blog posts.
The results they shared: one client hit 50,000+ Reddit views a month and $40k a month in traffic value, with 825% organic growth in three months. Another went from under 10 clicks a day to $1.8M in new revenue in a year.
Why almost nobody does it: Reddit credibility takes months and can't be mass-produced. Agencies can't sell it as a tidy retainer, so they tell clients "Reddit isn't SEO" and skip it.
The window is the interesting part. A year ago Reddit was an afterthought. A year from now everyone pays for this and the price is 5 to 10x. Early movers compound, like buying Google ads in 2003.
The brand that gets quoted in Reddit threads becomes the brand ChatGPT recommends. Boring, slow, and exactly why it works.
https://t.co/TIVOZP4uUj
https://t.co/qv0Ra8p948
AI video kept breaking on close-ups and fast motion. Seedance 2.0 in 4k just removed the last real excuse. https://t.co/H35jIFODoY
you've been saying AI video has no real quality
that it falls apart on close-ups
that movement breaks everything
seedance 2.0 in 4k just ran out of excuses for you
@BytePlusGlobal @lumina_ai_aiart
Full YouTube video https://t.co/n7bPOlfalT
Wait, you can now add Claude to a Slack channel and just @ it like a coworker?
Anthropic shipped Claude Tag. You drop Claude into the Slack channels you pick, give it access to the tools you want, and that's it.
Then you tag it the same way you would a teammate. "@Claude pull last month's numbers and draft the update." It goes and does the task while you do something else.
The unlock for a one-person business is real. You don't open a separate app, copy context back and forth, or babysit a chat window.
The work already lives in Slack. Now the agent lives there too, with only the access you handed it.
Copy-today move: make one channel for a recurring chore, give Claude just the tools that chore needs, and delegate it every week.
A solo operator with one Slack and a tagged-in agent starts to look like a small team. https://t.co/zv65SvyKxb
Introducing Claude Tag, a new way for teams to work with Claude.
In Slack, Claude joins as a team member with access to the channels and tools you choose. Tag Claude in and delegate tasks to it while you focus on other work.
Last Seen Users on Sotwe
mom son...กลุ่มลับแม่ลูก
Seen from Thailand
duitita
Seen from Malaysia
BRUTAL INDO29
Seen from Indonesia
Chinese Porn Central
Seen from Indonesia
Tuấn Anh 🇻🇳
Seen from Vietnam
นัดเยส สาวใหญ่
Seen from Thailand
Papua viral
Seen from Indonesia
xxx
Seen from Turkey
Negões Pauzudos
Seen from Brazil
Punjabi bull ♉
Seen from United Arab Emirates
Most Popular Users
Elon Musk
@elonmusk
240.4M followers
Barack Obama
@barackobama
119.3M followers
Donald J. Trump
@realdonaldtrump
111.7M followers
Cristiano Ronaldo
@cristiano
110.2M followers
Narendra Modi
@narendramodi
107M followers
Rihanna
@rihanna
97.6M followers
NASA
@nasa
92.1M followers
Justin Bieber
@justinbieber
90.8M followers
KATY PERRY
@katyperry
87.4M followers
Taylor Swift
@taylorswift13
81.3M followers
Lady Gaga
@ladygaga
72.8M followers
Kim Kardashian
@kimkardashian
69.7M followers
Virat Kohli
@imvkohli
69.5M followers
YouTube
@youtube
68.7M followers
Bill Gates
@billgates
63.7M followers
The Ellen Show
@theellenshow
62.5M followers
Neymar Jr
@neymarjr
62.2M followers
CNN
@cnn
61.9M followers
X
@x
60.8M followers
Selena Gomez
@selenagomez
60.5M followers