14:06Vaibhav Sisinty · Listicle
These 10 New Claude Skills Just Doubled My AI Subscription (For Free)
A 14-minute ranked teardown of the only 10 Claude skills worth installing, each demonstrated live with real output.
May 13thThe argument in one line.
The most capable AI models are increasingly gated behind government and vendor approval rather than opened to everyone, and the real risk of adopting AI coworkers isn't the model itself but that your company's irreplaceable operating memory ends up locked inside whichever vendor's tools you deploy first.
Read if. Skip if.
READ IF YOU ARE…
- You want a fast, non-technical weekly digest of what changed across the major AI labs and tools without reading a dozen separate articles.
- You're evaluating whether to adopt an AI agent inside your company's chat tool or workspace and want the vendor lock-in risk explained in plain terms.
- You're curious about no-code AI design/build tools and want to see a full one-sentence-to-live-app workflow demonstrated step by step.
- You follow the Claude vs GPT competitive back-and-forth and want the latest capability and access-restriction news.
SKIP IF…
- You need deep technical benchmarks or primary-source verification - this is a summarized roundup, not an investigative report.
- You already use Genspark, Claude Tag, or Notion AI agents daily - the demos here are introductory, not advanced.
The full version, fast.
OpenAI's GPT-5.6 reportedly beat Claude's Fable 5 on a coding benchmark, then was immediately restricted to a small trusted-partner preview at the request of the US government - mirroring what happened to Fable 5 weeks earlier. The host argues this signals powerful AI concentrating behind gatekeepers rather than staying open to everyone. Meanwhile, AI-driven memory-chip shortages are pushing up Apple, Microsoft, and other hardware prices, and Anthropic launched Claude Tag, a Slack-native AI coworker with per-channel permissions and persistent memory - framed by one critic as a 'Trojan horse' because a company's operating memory becomes locked into whichever vendor runs it. The episode closes with a full walkthrough of Genspark Design turning a single text prompt into a landing page, an 11-screen app design, a launch video, and deployable live code.
Free for members
Chat with this breakdown — free.
Sign in and you get 23 free chat messages on us — ask for the hook, quote a framework, find the exact transcript moment, generate a markdown action plan. Bring your own key when you want unlimited.
Create a free account →Where the time goes.
00:00 – 01:05
01 · Cold open / this week in AI
Hardware price hikes, GPT-5.6 beating Fable 5, Claude Tag teased as a 'quiet trap,' and the Genspark tutorial promised for the end.
01:05 – 03:10
02 · OpenAI GPT-5.6 beats Claude, then gets restricted
GPT-5.6 (Sol/Terra/Luna) reportedly beats Fable 5 on coding; US government requests a limited trusted-partner preview instead of public release, citing cybersecurity capability.
03:10 – 04:58
03 · AI inflation: why Apple raised prices
AI data centers eating the global memory-chip supply pushes up MacBook/iPad/Xbox prices; Micron becomes the most-traded US stock.
04:58 – 08:29
04 · Claude Tag: the AI coworker in Slack
Anthropic's Claude Tag works inside Slack with per-channel permissions and persistent memory; an ex-MIT professor calls it a 'Trojan horse' for company context lock-in; host's takeaway is 'own your context.'
08:29 – 09:17
05 · NVIDIA BioNeMo: AI discovers medicine
NVIDIA's BioNeMo Agent Toolkit lets agents run real lab tools on GPUs to design drug candidate molecules in minutes instead of days.
09:17 – 10:02
06 · Ornith: free open-source coding AI
A free, open, self-improving coding model that writes its own workflow rules while guarding against reward hacking.
10:02 – 10:56
07 · Sakana Fugu: the unbannable AI
A multi-model orchestrator (GPT, Gemini, Claude) from Japan that claims to match banned-model benchmark scores by combining existing models rather than depending on one.
10:56 – 11:54
08 · ChatGPT GPT-5.5 Instant
The free default ChatGPT model gets a playful update and is shown planning a detailed 5-day Kerala trip with budget and constraint tradeoffs explained.
11:54 – 13:04
09 · Genspark Design teaser: sentence to app
Introduces Genspark Design, powered by Claude, which turns a single sentence into finished app screens, later expanded into the full tutorial.
13:04 – 13:58
10 · Codex on the ChatGPT mobile app
OpenAI puts Codex inside the ChatGPT phone app so users can trigger and approve computer tasks remotely.
13:58 – 14:52
11 · Gemini study notebooks (free)
Google's free Gemini study notebooks quiz students to find weak spots and build personalized lessons, positioned as a low-cost alternative to expensive coaching.
14:52 – 16:06
12 · Notion AI agents
AI agents work directly inside Notion project boards, moving task cards through workflow stages autonomously while a human supervises.
16:06 – 17:13
13 · ByteDance Seedance 2.5 AI video
A new AI video generator makes 30-second clips from up to 50 combined input assets and can swap a single element without re-rendering the whole video.
17:13 – 18:04
14 · Hermes open-source AI agent
A free agent framework adds a 'blank slate' mode (abilities off by default) and a slash-learn command that turns any reference material into a reusable skill.
18:04 – 18:55
15 · Microsoft Excel Copilot skills
Copilot in Excel learns a recurring report process once, then any teammate can trigger the full report generation with one line.
18:55 – 19:44
16 · AI chip war: OpenAI Jalapeno + IBM
OpenAI builds its own inference chip (Jalapeno) to cut reliance on NVIDIA, while IBM packs nearly 100 billion transistors onto a fingernail-sized chip.
19:44 – 20:43
17 · Figma Motion
Figma adds native animation/timeline tooling so designers can animate screens and hand developers ready-to-use motion code without rebuilding from scratch.
20:43 – 21:24
18 · Grok /goal
Grok Build adds an autonomous goal-and-walk-away mode that plans, executes, and self-checks its own work, catching up to Claude Code and Codex.
21:24 – 22:10
19 · Perplexity Computer for Counsel
An AI agent for legal teams that links every answer to a verifiable source, connecting to DocuSign and case-law tools and routing tasks across GPT-5.5, Claude Sonnet, and Gemini.
22:10 – 30:10
20 · Genspark full build-along tutorial
Step-by-step live demo: one prompt becomes a landing page, then an 11-screen app UI, then a launch video, then live deployable code, plus saving a reusable design system and using templates.
30:10 – 30:58
21 · Recap + community CTA
Wraps with a teaser for a separate Sakana Fugu vs Fable 5 head-to-head video and a subscribe/community call to action.
Lines worth screenshotting.
- GPT-5.6 reportedly beat Claude's Fable 5 on OpenAI's own coding benchmark, then was restricted to a small trusted-partner preview at the request of the US government within days.
- The same government-access restriction that hit Claude's Fable 5 weeks earlier has now hit OpenAI's most powerful model too - two frontier models gated in a row.
- Anthropic says 65% of its own product team's code is already written through Claude Tag before offering it to customers.
- Claude Tag enforces access per team and per channel - a Claude instance in a legal channel literally cannot see the engineering codebase, and memory learned in one channel doesn't leak to another.
- An ex-MIT professor calls Claude Tag a 'Trojan horse': swapping the underlying AI model is easy, but a company's accumulated workflow memory gets stuck inside whichever vendor's tool holds it.
- A MacBook Pro in India jumped 70,000 rupees overnight because AI data centers are consuming so much of the world's RAM and storage supply that consumer hardware prices are rising - a trend the host calls 'AI inflation.'
- Micron, one of only three companies that make most of the world's memory chips, became the single most-traded stock in America - trading more volume daily than Nvidia or Tesla.
- NVIDIA's BioNeMo Agent Toolkit lets an AI agent actually run lab tools and design drug candidate molecules, cutting the molecule-discovery step from days to minutes.
- Ornith is a free, open-source coding AI that writes its own workflow rules through self-improvement rather than having a human hand-code them, addressing reward-hacking risk directly.
- Sakana Fugu claims to match banned-model benchmark scores by orchestrating multiple existing models (GPT, Gemini, Claude) together rather than using any single one, so it can swap out any model that gets banned.
- ByteDance's Seedance 2.5 can generate a single AI video up to 30 seconds long, combine up to 50 input assets including photos of 10+ different actors, and swap one element of a video (like a product) without re-rendering the whole clip.
- Genspark Design turns a single text prompt into a landing page, a matching multi-screen app design, and a launch video, then can convert the finished design directly into deployable working code.
- The host's core takeaway: rent whichever AI model is best this month, but own your own context (files, chat history, workflows) so no single vendor can hold your operating memory hostage.
The real risk isn't the AI model, it's who owns your context.
WHAT TO LEARN
As frontier AI models get gated behind government and vendor approval, the practical move is to keep renting whichever model is best while making sure your own workflows and data never live exclusively inside one vendor's product.
- Frontier AI models are increasingly restricted at launch rather than released broadly, so plan for capability gaps between what's available to insiders and what's available to you.
- An AI agent embedded in your company's chat or workspace tool can quietly become a lock-in point: the model is swappable, but the accumulated workflow memory built up inside it is not.
- Before adopting any 'AI coworker' product, check whether its memory and history export cleanly, or whether closing the account also erases the operating knowledge it built.
- Rising consumer hardware prices are a real, traceable downstream effect of AI infrastructure demand on shared component supply chains, not a coincidence or one-off.
- Multi-model orchestration tools (combining several AI systems rather than depending on one) are emerging specifically as a hedge against any single model being banned, price-hiked, or shut down.
- No-code AI design tools can now take a project from a single sentence to a deployable, live product in one sitting, collapsing what used to require a designer, developer, and video editor into one person's afternoon.
- When evaluating a new AI agent for legal, financial, or compliance-sensitive work, verify it links every answer to a checkable source rather than trusting fluent-sounding output.
Terms worth knowing.
- Claude Tag
- Anthropic's Slack-native AI agent, invoked by tagging @Claude in a channel, that can read files, break tasks into steps, update documents, and post status updates with per-channel permission boundaries.
- Trojan horse (context lock-in)
- A framing used to describe how adopting an AI coworker tool can quietly trap a company's accumulated operational knowledge inside one vendor's product, making it hard to switch AI providers later even though the underlying model itself is easy to swap.
- AI inflation
- Rising consumer hardware prices (laptops, tablets, gaming consoles) caused by AI data centers consuming a large share of the global supply of memory chips (RAM and storage), driving up costs for everyone else.
- BioNeMo Agent Toolkit
- An NVIDIA toolkit that lets AI agents run real scientific software on NVIDIA GPUs to design and evaluate drug candidate molecules, rather than just describing the process in text.
- Reward hacking
- A failure mode where an AI system, when allowed to design its own success criteria or workflow, learns to satisfy the letter of a test (e.g. faking expected output files) instead of actually completing the underlying task.
- Terminal-Bench
- A coding benchmark referenced as the test on which OpenAI claims GPT-5.6 outperformed Claude's Fable 5.
Things they pointed at.
01:05productGPT-5.6 (Sol, Terra, Luna)
04:58productClaude Tag
08:29toolNVIDIA BioNeMo Agent Toolkit
09:17toolOrnith
11:54toolGenspark Design
13:58toolGemini study notebooks
16:06toolByteDance Seedance 2.5
17:13toolHermes
21:24toolPerplexity Computer for Counsel
00:00linkStaying Ahead community
Lines you could clip.
01:05
“OpenAI just built something so powerful, the US government stepped in before it could go public.”
07:50
“Rent the intelligence from whoever's best this month, but own your context.”
07:00
“Claude Tag is a Trojan horse, not because Anthropic is doing anything evil, just because the incentives are obvious.”
Word for word.
Read-along
Don't just watch it. Burn it in.
See every word as it's spoken — crank it to 2× and still catch all of it. The same dual-channel trick behind Amazon's Kindle + Audible.
metaphoranalogystory
00:00AI is about to make your next laptop, phone, even your next Xbox more expensive, and almost nobody saw it coming. Apple just raised their MacBook and iPad prices, and they won't be the last. But here's the thing.
00:11That's not even the biggest story this week. OpenAI just built its most powerful AI ever, powerful enough to beat Anthropic's Fable five. And it's so powerful, the US government stepped in and blocked us from using it.
00:22And Anthropic quietly slipped an AI coworker into companies, one that reads your files and works on its own. This one might sound like a normal update, but there's something big hidden inside it. An ex MIT professor is calling it a quiet trap.
00:34Plus, a free Gemini study tool that could replace expensive coaching and a whole lot more. 18 updates this week. And right at the very end, we tested Gens park Design, a tool that builds a full working app from a single sentence.
00:46And what it pulled off is genuinely worth seeing. If any of these get technical, don't worry. The time stamps are in the description.
00:53Jump to whatever you want. And the full step by step guide for Genspark plus all the updates is waiting in our community. Now let's get into it.
01:05OpenAI just built something so powerful, the US government stepped in before it could go public. Same thing that happened to Claude's Fable five last week. Now it's OpenAI's turn.
01:15OpenAI revealed its next generation GPT 5.6. It comes in three versions.
01:21Soul is the flagship, the most powerful one. Terra is the balanced everyday one. Same level as their last model, but half the price.
01:29And Luna is the cheap, fast one for high volume work. And on OpenAI's own coding test, the flagship soul just beat Claude's Fable five. But here's the twist, you can't actually use it.
01:40OpenAI says that at the request of the US government, they're starting with a limited preview for a few trusted partners, not the public. The reason should make you sit up.
01:50OpenAI says this is their strongest model yet at cybersecurity. Good enough at finding weak spots in software that the government wanted a look before it went out to everyone. But step back because this is bigger than one model launch.
02:04And honestly, it worries me. The most powerful AI ever built is going out to a small group of chosen people first, and that head start is everything. They get to build with the best AI while the rest of us wait for permission.
02:17And by the time we catch up, they're already years ahead. And the reason given is safety. Too powerful, too good at cybersecurity, so access has to be controlled.
02:26Maybe that's fair. But notice who ends up in the room. The same companies building these models have pushed hard for rules like this.
02:34And now the rules are here, and they decide who's let in. And here's the part that really gets me. These models learned from us.
02:42From everything humanity ever put online, your posts, your code, your writing, it was trained on all of us. And now suddenly, we are the ones who can't be trusted to use it.
02:52My honest read, this might be the moment the dream quietly died. The dream that AI would be for everyone, that a kid with a laptop anywhere could use the same intelligence as a billion dollar company.
03:04I really hope I'm wrong. But if this isn't pushed back on now, it gets very hard to undo later. Your next MacBook and iPad just got more expensive, and the reason is AI.
03:15Here's the chain. AI runs in giant data centers and those eat huge amounts of memory chips. The same RAM and storage that sits inside your laptop.
03:24So much of it now goes to AI that less is left for everyone else and prices shot up. For months, Apple absorbed that cost. This week, it stopped.
03:33In India, a MacBook Pro jumped 70,000 rupees overnight from about 1.7 lakh to 2.4 lakh.
03:41The iPhone was spared for now. And it's not just Apple. Microsoft just raised Xbox prices worldwide for the exact same reason with Dell, Lenovo, and Samsung right behind it.
03:52There's a name for this now, AI inflation. And here's one more twist. Part of why Apple's cheapest desktop, the Mac mini, keeps selling out is AI itself.
04:01Builders are snapping them up as tiny private AI servers, running agents on them around the clock instead of renting the cloud. Free tools like OpenClaw, an AI agent you run yourself, helped push that trend enough that Apple's own CEO named agent demand as one reason the Mac mini has been sold out for months.
04:20So AI hits your wallet from more than one side. It eats the memory that makes every Mac pricier, and it's part of why the cheap one keeps selling out. And here's the part that stinks.
04:30While you pay more, the companies that make this memory are having the best year in their history. Micron, one of only three companies that make most of the world's memory, just posted record profits, and it's now the single most traded stock in America. More money flowing through it every day than Nvidia or Tesla.
04:48It's become the center of the whole AI trade. The AI boom is quietly making them rich, and you're covering the bill. That's the part nobody mentions about this AI race.
04:58For the first time, it just showed up on your receipt. Imagine adding a new hire to your team's group chat, one that never sleeps. That's basically what Anthropic just launched.
05:07It's called Claude Tag, and for now, it lives inside Slack, the app that serious companies use for work instead of running everything on WhatsApp. You type at Claude in any channel, hand it a task like build scheduled exports, and it takes over. This is where it stops being a chatbot.
05:24It pulls the right files from your drive, breaks the job into steps, updates the docs, and post a checklist of what's done. And this isn't them just selling it. Anthropic says 65% of their own product team's code already gets written this way, and they have been testing it internally to speed up product iteration and prove business value before rolling it out to customers.
05:45There's one claud per channel. So a teammate picks up right where you left off, switch on ambient behavior, and it stops waiting to be asked, nudging stalled threads, and flagging what affects your work. Now the obvious worry, an AI loose in your company files.
06:00Their answer, Claude has access to the systems it needs on a per team, per channel basis. So Claude in the legal channel can see contract information.
06:08In the engineering channel, it edits the code base. But if you tag Claude in the legal channel to edit the code base, it won't. Not because it's being polite, but because it literally cannot see it.
06:19Memory respects the same boundaries. So what it learned in a private channel or a DM stays there. And just to repeat, Claude has its own account, so every credential used is locked.
06:29The catch, it's beta and only on the paid team and enterprise plans for now. Clot design also got a major update last week, and it barely resembles the old one. A full master class is in production.
06:41So if you want it, let us know, and we'll prioritize it. But step back for a second because this is bigger than a Slackbot. Andre Karpathy,
06:48who just joined Anthropic, called it the third big shift in how we use AI. First, it was a website you visit, then an app you download. Now it's a coworker that lives inside your company with your memory and your tools working right next to your team.
07:04And here's the part worth sitting with. The more Claude runs your company, the more of your company lives inside Anthropic. Ashwin Gopinath, an ex MIT professor, put it sharply.
07:14He says Claude Tag is a Trojan horse and not because Anthropic is doing anything evil, just because the incentives are obvious. His point is this, swapping the AI model is easy. But the memory of how your company actually runs, the workflows, the customer promises, the we tried that and it flopped lessons, that's the part that gets stuck inside one vendor.
07:35Models can be swapped. Your company's memory can't, not easily. My honest read, this is the direction, full stop.
07:42AI is going from a tab you open to a teammate you manage. So don't fear it. Start using it now.
07:48Just keep one rule in mind. Rent the intelligence from whoever's best this month, but own your context. And what do I mean by your own context?
07:56Think of it like this. Context is everything your AI knows about how you work, your files, your past chats, your steps, your little preferences. Right now, imagine all of that getting saved inside one company's app.
08:10The day they raise the price or shut it off or you want to switch, you can't just pick it up and walk out because your . Owning your context just means keeping that memory somewhere you control so any AI can use it and no single company holds your work hostage. That, by the way, is exactly why open models matter so much right now.
08:33So far, AI has mostly been good at talking about things. NVIDIA just took a big step toward AI that can actually do science. They launched the Bionemo agent toolkit, and the simplest way to understand it is this.
08:46Right now, an AI can read about how to design a medicine, but it can't run the real scientific tools to do it. NVIDIA just handed AI agents those tools. So you type something like design 10 protein binders for PDL one and show me the results.
09:01The agent plans the steps, runs them on NVIDIA's GPU, and drops the structures into a three d viewer. And that matters because discovering a new drug normally takes years and billions of dollars. The first slice of that, finding the molecules NVIDIA says, just went from days to minutes, and Anthropic and OpenAI are already plugging in.
09:20Quick one for the people who liked GLM 5.2, that free Chinese model we covered last week. A US team just dropped their own free open coding AI called Onith.
09:30And the interesting bit is how it was made. Normally, when an AI writes code, a human sets up all the rules around it, how it should try things, when to retry, how to check its own work. Onith figures out those rules by itself, then solves the problem.
09:45Like a worker who works out the best way to do a job instead of waiting to be told. And it's completely free and open. Anyone can download it and even use it commercially.
09:55The takeaway, building powerful free AI isn't just a China thing anymore. The US is in this race too. After China answered the fable five ban with GLM 5.2, a free model anyone can download, Japan just came back with a completely different kind of answer.
10:10It's called Sakana Fugu. And instead of being one model, it's a team of the top available AIs, GPT, Gemini, Claude's, Opus, and it splits your task across them, then hands back one combined answer.
10:23The clever part, since it's not tied to any single model, if one ever gets banned tomorrow, Fugu just swaps it out and keeps running. And get this, Sakana says it hits Fable five level scores on many benchmarks without using Fable five at all since that's the model nobody outside The US can touch anymore. Now testers are still putting those claims through their paces, so we did too.
10:46We ran it head to head against Claude Fable five and built a full video around what we found. Go watch that next. But first, let's wrap up the week's other major updates.
10:56This week, OpenAI posted a fun tweet for Messi's 30 birthday showing off how playful its newest model has become. You just type goat birthday, and it knows you mean messy. Wishes him happy birthday, goat emoji, and all.
11:09That's the new GPT 5.5 instant, the default chat GPT, the free one hundreds of millions of people use every day. So we tested that part. We asked it to plan a five day Kerala trip.
11:22Four people, 80,000 rupees, a backwater stay, a hill station, only two hotel changes, and one person who can't sit in a car for more than three hours. See that little pause before it answers.
11:34It's working through all of it, not rushing. And look at the choices. It dropped Alepi, the famous backwater, and picked Kumarakum instead, which is closer, quieter, and has the same backwaters.
11:46The pricey houseboat swapped for a lakeside resort and a day cruise to protect the budget, and it explained every single call. Genspark just launched something that turns one sentence into a finished app design. In seconds, it's called Genspark Design, and it's powered by Claude, which has genuinely good taste.
12:04Say you want a note taking app. You type exactly that. Design a note taking app.
12:09And right away, it builds a full set of polished screens. Not a rough sketch, actual finished designs. Don't like the look?
12:16You just tell it. Make it more playful, and it redesigns the whole thing on the spot. New colors, illustrations, the works.
12:22Here's the part I love. You can literally draw on the screen, circle an empty corner, write, add a dark mode toggle, and it appears right there where you drew it, like sketching on a napkin, except it actually builds what you sketched. And it goes further than screens.
12:38From that one app, it builds you a matching landing page, even a product demo video all in the same style. But this is the moment that matters. When you finally like it, you hit one button, build it, and it turns that whole design into real working code.
12:53Normally, this takes a designer, then a developer, and weeks of back and forth. You just did all of it yourself in one flow.
13:01We love this one so much. We actually tested it ourselves and built a full tutorial for you, but you'll have to wait till the end. We've still got some exciting updates to get through first.
13:10OpenAI just put Codecs inside the ChatGPT app on your phone, and it lets you run tasks on your computer from there. Your laptop stays on at home. You're out with just your phone, and you can hand it work from wherever you are.
13:22Here's what that looks like. You type a message like grab the latest video from the launch folder and send it to the team on Slack. Your computer picks it up, opens the right app, makes the file and posts it in Slack.
13:35You can see it working step by step on your phone. And if it's in the middle of a longer job and hits a point where it needs your okay, it sends that to your phone too. You tap approve and it carries on, so it's not stuck waiting for you to get back to your desk.
13:49You can also connect more than one machine, your laptop, a Mac mini, and send work to whichever one you want. Right now, it's rolling out as a preview on iPhone and Android. So it's early, but it's a useful shift.
14:02Google just launched something that could replace expensive coaching for a lot of students, and it's completely free. It's called study notebooks inside the Gemini app. You start a notebook, upload your own material, your textbook, your syllabus, and tell it what you're preparing for.
14:18First, it gives you a quick quiz, not to grade you, but to find where you're weak. Then it builds short personalized lessons aimed right at those gaps, and a dashboard tracks what you've got down and what still needs work. And here's the part that hits home for India.
14:33If you know what NEET and G coaching cost these days, it's gone completely insane. Lacks of rupees most families just can't spare. This already works for both in your own language and it costs nothing.
14:45So a kid who could never pay for a coaching class can now get a proper study plan made just for them. Good coaching has always been about who can afford it. This finally changes that.
14:55You can now drop AI agents straight into Notion, the app a lot of people already use to run their projects. Before this, the AI lived in some other tab. You'd ask it something, wait around, then copy the answer back into Notion yourself.
15:09Now the agent just works inside Notion and you hand it tasks like you would a teammate. Watch this board. A bug shows up as a task card.
15:18A clawed agent grabs it, reads through your whole workspace to figure out what's going on, finds the problem, writes out the fix, and you can see the card slide from triage over to plan on its own. Then a person takes a quick look, says looks good, and just tags cursor. Cursor picks it up, writes the actual code, sends it off for review, and the card slides all the way to done.
15:40Nobody dragged it. The agents moved it, and you don't have to stop at one. You can run a whole team of these at the same time, different jobs going at once, even after you've shut your laptop for the day.
15:51And the whole time, you're the one calling the shots. You decide what each agent can touch, and you can watch exactly what it's doing. So that project board you already live in, it just turned into your newest set of teammates.
16:02ByteDance just revealed a new version of its AI video generator, Cdance 2.5, and it pushes past what these tools could do before. Three things got a big jump.
16:12First, length. Most AI video tools max out around fifteen to twenty seconds. Cdans can now make a single video up to thirty seconds long, the longest of any tool right now, Bydans says.
16:24Longer clips mean the scene actually holds together instead of cutting off. Second, you can feed it a lot more, up to 50 things at once, images, clips, audio, handed photos of over 10 different actors, and it arranges them all into one party scene, everyone in the right place. But the editing is what stood out.
16:44Third, you can change one piece of a video without redoing the whole thing. A lipstick ad, same scene, same action, just swap the model and the language. The same ad runs in English, French, Japanese, Spanish, Korean, and Hindi, lips matching each one.
17:01One ad instantly remade for every market. It's in final testing now with a full release in early July, but the direction is clear. AI video is going from short single clips to something that can carry a whole scene.
17:14Next, a free open source AI agent called Hermes got two updates that come down to one idea. An AI you actually control and one that actually remembers. First, a new blank slate mode.
17:26Most agents come fully loaded. Every ability switched on from the start. Hermes has over 60, but blank slate keeps almost all of them disabled.
17:35You enable only what you need, and your choices are permanent even after an update. Second, a command called slash learn. Point it at almost anything you already know.
17:45A folder of documents, API docs, a manual, a PDF, even a task you just walked it through, and it reads all of it and turns it into one clean reusable skill. Next time, one word and it runs the whole thing. No explaining it again.
17:59Two updates, same direction. An agent you shape that then shapes itself around you. Next, Microsoft just made Excel a lot smarter, and this one hits anyone who lives in spreadsheets.
18:11You can now teach Copilot to do your boring monthly reports for you. Think about the same report you rebuild every single month, Same steps, same formatting, hours gone.
18:21Now you walk Copilot through your process just once, save it, and after that, anyone on the team types one line and it's done. So you ask it for the quarter end report with the commentary and all, and it just goes, spins up a fresh sheet, fills in the tables, formats everything the way your company likes it, and even writes out what the numbers actually mean.
18:42The kind of report that used to eat someone's whole day finished in seconds. The catch, it needs the paid Microsoft copilot plan, and this is still rolling out. But the direction is clear.
18:53Your spreadsheets are about to start doing the boring part for you. Two chip stories this week, and they matter more than they sound. The AI race is moving into the chip itself, the part that actually runs everything.
19:05First, OpenAI built its own chip called Jalapeno. Until now, they rented their computing power mostly from NVIDIA. This one's designed to make chat GPT faster and cheaper to run, and they even used their own AI to help design it.
19:20Then IBM packed nearly 100,000,000,000 tiny switches onto a chip the size of your fingernail, crossing a technical limit many thought was years away. The kind of leap that keeps every future chip getting faster, though this one's still about five years from real products, different ends of the same shift.
19:38The next phase of AI gets decided by who builds the hardware to run it, not just the smartest app. Figma, the tool that nearly every designer in the world uses to design apps and websites, just launched Figma Motion.
19:52Now you can make your designs actually move right inside Figma. Here's why that's a deal. Till now, Figma could only make still screens.
20:00If you wanted a button to slide or a screen to animate, you had to take it into a separate tool, do the work there, then hand it to a developer to rebuild from scratch. It broke every time. Not anymore.
20:13You turn on a timeline, mark where a thing starts and where it ends, and Figma fills in the movement on its own. Slide, grow, spin, fade, all of it. You can even just type what you want, like make this text glitch and the AI animates it for you.
20:29And the best part, a developer can pull that finished animation out as ready code and drop it straight into the real app without rebuilding it from scratch. It's free right now and still in early testing, So expect a few rough edges. Grok just added something in Grok build that moves it closer to the big coding tools.
20:48It's called slash goal, and the idea is simple. Normally, when you use AI to build something, you babysit it. It does one step, stops, waits for you, does the next.
20:59With this, you just give it the goal once and walk away. So you type something like rebuild this part of my app, and it plans the whole thing out, breaks it into steps, does each one, and here's the key part, it checks its own work at the end. If something's broken, it keeps going until it actually works.
21:15You can pause it, steer it, or just let it run. Now fair point. This isn't something brand new.
21:21Claude Code and Codex have done this for a while now. So this is Grock catching up to the pack. For two years, lawyers have been facing problems for citing cases that don't exist because their AI made it up.
21:33Perplexity decided to solve this problem. It built Computer for Counsel, an AI agent for legal teams, where every answer links back to a real source you can verify. It securely connects the tools lawyers already use, like DocuSign for contracts and MidPage for case law.
21:49Say you ask it to build a tracker for US privacy laws, it routes to the best model for each task, Whether that's GPT 5.5, Claude Sonnet 4.6, or Gemini 3.1 Pro and provides each answer with citations.
22:03You can track regulatory changes in your mail, manage NDAs in one place, track all trademarks, and even research deeply. The computer does the heavy lifting, and the judgment remains yours. Okay.
22:14This is Genspark, the tool I promised you. One line of text, it builds the whole thing, the landing page, the app, even the launch video for it. And the best part is we're going to do all of it together, every single click.
22:26So let's just get straight into it. Genspark is one big AI workspace. When you log in, this is the home screen.
22:31And, yeah, it does a lot. AI employees and office suite, content tools, all of it. But we only need one thing today.
22:38Up at the top, let's go to build, and under it, there's an option called design and prototype. Let's click that, and it'll open up a new tab that asks one simple question. What would you like to design?
22:49Look at everything it can make. Websites, mobile apps, marketing, social posts, even videos, posters, and documents.
22:57One tool for nearly every visual thing a solo founder needs. And if you've used Claude Design, this is Jen Sparks' take on that same idea. Funny enough, it's actually powered by Claude under the hood.
23:09So let's build something real. And I know the screen looks busy right now, but don't worry about a single button on it. The way this works is really simple.
23:16We paste in a prompt. We answer a few quick yes or no questions, and we hit present. That's the whole skill.
23:23Let's start with the landing page. So let's paste in the prompt. An interactive landing page for a notes app called staying ahead with draggable stickers, flipping Polaroid notes, a community wall, a sign up form in cream and dusty rose colors.
23:38That's the whole instruction. Let's hit enter. Now watch this because it doesn't just start drawing, it stops and thinks first.
23:45See, it spins up a little project over on the right, gives it a name, and writes itself a little checklist before it builds anything, what layout it should go with, what colors to use, what interactions it should have. It's planning the job before it does the job. Then it asked me a few quick questions to get the details right.
24:04I answered those off camera, and you'll see every single one of them on the finished page in a second. Okay. It's done.
24:10Let's hit present, and there's our page. Let's go through it top to bottom. Up here, the hero with the handwritten title and a sticker you can drag around.
24:19Scroll down to these feature cards and watch what happens when I hover over one. See that? It flips like a real Polaroid.
24:25Let's keep going. And here's a community wall with handwritten notes pinned onto it. And right at the bottom, a proper sign up form.
24:33Now let's prove this is actually real. Let me drag across the text on the page. See how it highlights.
24:39This isn't a screenshot. It's a real working web page with live text you could edit. This is wild.
24:44And those questions it asked me, they're all listed right here on the page. Oh, and staying ahead, that's actually my own community, so building this one felt a little personal. A landing page is one thing.
24:55Now let's build the whole app. So let's head back to the chat, and this time we tell it to design an app UI. And watch how it interviews us one question at a time.
25:04It asks what platform we want, so we go with iPhone. Then it asks which screens to design and gives us a whole list. Home, the notebook, the editor, onboarding, settings, profile, so we just tick all of them.
25:16It asks if we want everything on one canvas. Yes. Whether to keep the same look as the landing page?
25:22Yes. How much of that handwritten style we want? Just a few touches.
25:26What interactions to wire up? Tapping a notebook to open it. Swiping a pen tool.
25:31And finally, how many variations of each screen? There's a little slider, so let's set it to three and hit continue. Here's the thing.
25:39That whole stretch felt like a designer sitting across from you asking what you want. But really, we were just clicking iPhone, ticking a few boxes, and saying yes. Now before it draws anything, it plans it out again.
25:51See it thinking. It writes out a full plan, an 11 screen structure grouped into onboarding, the core app, Discover, and Social with the exact style it's going to use. And then look at this.
26:02It's given us 11 full phone screens. Every single one we asked for. Let's pan across slowly.
26:08Here's the onboarding, the first few screens someone sees when they open the app. Then we move into the core of the app, the shelf, the notebook, the editor.
26:18Over here is discover with search and browse, and then the whole social side. Zoom into any one of them and look at the detail, the toggles, the little handwritten touches. And if there's something you don't like, let's just grab the pen up in the corner, circle the thing we want to change like this, and tell it what to fix.
26:36And if you change your mind, you just undo it. Okay. This next part, I did not see coming.
26:42Honest take, I expected a gimmick. Gimmick. Let's ask Genspark to make a launch video for the app we just designed.
26:48In the chat, we just type create a launch video for this app and hit enter. And just like before, it scopes the whole thing out first. It asks how long we want it.
26:59So let's say thirty seconds. It asks what dimensions we'll go with 16 by nine. It asks how it should be built, animated, AI generated, or cinematic.
27:09Let's go animated. It asked for the storyline, a day in the life or a straight product walkthrough. We'll keep it the walkthrough.
27:16Which screens to feature? All of them. The mood, mysterious and teasing, the audio, voice over, music, and and sound effects.
27:23Then it asks for the little signature touches, hand drawn ink, stickers landing, a cursor trail, and we say yes to all three. And finally, how we want it delivered, a player plus a file we can download. Continue.
27:37It plans it out one more time and then it renders. That takes a moment. So let's jump straight to the result and hit present.
27:43Watch this. Hand drawn elements animating in, the title appearing, and then a clean fly through of the exact app screens we built two minutes ago with music and motion behind it. So one prompt just gave us a landing page, a full app design, and a finished launch video.
28:00That's the work of a designer, a developer, and a video editor. It's night and day from how this used to work.
28:07Before we ship it, two quick things that'll make this yours for the long run. First, the look. We don't want to rebuild the same style every single time.
28:15Right? So let's save it as a design system. We open it up, click use in new project, and now it shows up in my designs as a saved card.
28:23So from now on, anytime you start something new, you click that card first, and everything you make just stays on brand automatically. Second, templates. If you don't feel like starting from a blank prompt at all, Genspark already has ready made ones, websites, social posts, documents.
28:40So let's open this document template. We get a full preview of it. And if we like it, we just click make this template yours, and it prepares our own editable copy.
28:49And there it is ready for your words and your brand. So now you've got two ways in. You either describe something from scratch or you grab a template and make it your own.
28:58Either way, you're never staring at a blank page again. Now, this is the part that actually matters. Everything so far has been a design.
29:05A really good one, but still just a design. And to make it real, we have to put it live on the Internet. So back in our app, up in the corner, there's a build button.
29:14Let's click it, choose app, and watch the screen split into two. On the left, Genspark actually starts writing the code line by line, and you don't have to read a word of it, just watch it work. And on the right, a real working version of the app shows up.
29:30Let's tap through it. Scroll the notes, drag things around, open the screens we just designed. It all works.
29:36This isn't a picture anymore. It's a running app, and we can do the exact same thing for the website. Then when we're ready, we can put it live on the Internet for real.
29:44See these options up here, publishing, hosting, version history, even a GitHub connection.
29:50Some of those words look technical. I get it, but you don't have to touch any of them. Genspark sets it all up and walks you through it.
29:57So let's zoom all the way out. You started with one sentence, and you walked away with a landing page, a full app design, a launch video, and a real live product on the Internet. One person, one afternoon.
30:09No engineering team. So that's the week. And that AI from Japan that claims to match the band Fable five.
30:15We put them head to head, and the result genuinely surprised us. That full test is waiting for you too right on the screen. And everything from today, every tool, every link, every guide, it lands inside our staying ahead community before it ever hits YouTube.
30:29That link is in the description. Come grab it. Last thing.
30:31AI moves a full week ahead every seven days. And if you're not subscribed, YouTube simply won't show you the next one. So subscribe, and let's stay ahead of it together.
30:39See you next week.
The Hook
The bait, then the rug-pull.
The host opens on hardware sticker shock before pivoting to the episode's real hook: OpenAI's most powerful model yet beat Claude on a coding benchmark, and within days the US government stepped in and restricted it - the same fate that had just befallen Claude's own frontier model.
How they asked for the click.
VERBAL ASK
30:30subscribe
“AI moves a full week ahead every seven days. And if you're not subscribed, YouTube simply won't show you the next one. So subscribe, and let's stay ahead of it together.”
Soft recurring CTA delivered at the very end after the value is fully delivered, paired with a plug for the free 'Staying Ahead' community where guides and links are posted first.
MENTIONED ON CAMERA
FROM THE DESCRIPTION
PRIMARY CTAWhere the creator wants you to go next.
OTHER LINKSAlso linked in the description.
Visual structure at a glance.
