Modern Creator
Ben AI · YouTube

How the Top 1% of Claude Users Work Differently

Six lessons pulled from two Anthropic studies on 400,000 Claude sessions and on how Anthropic's own staff use Claude every day.

Posted
yesterday
Duration
Format
Essay
educational
Views
25.6K
459 likes
Big Idea

The argument in one line.

The best Claude users are the best managers: they brief clearly, correct by fixing the prompt, make output easy to review, and keep their setup lean.

Who This Is For

Read if. Skip if.

READ IF YOU ARE…
  • You use Claude or Claude Code daily for work and suspect other people are getting noticeably better results from the same models.
  • You run a team or business and want Claude to act on meeting decisions and action items without you relaying each one.
  • You keep re-prompting after bad outputs and want a faster way to correct course than long iteration loops.
  • You have accumulated skills, connectors, scheduled tasks and a long CLAUDE.md and wonder if they are slowing things down.
SKIP IF…
  • You want a deep technical walkthrough of the studies' methodology; this is a practical summary of the findings.
  • You have never used Claude Code or Claude's artifacts, since several tips assume those features exist.
TL;DR

The full version, fast.

Anthropic studied about 400,000 Claude sessions and its own staff, and found the best users get a verified result more than twice as often as beginners. Experts are defined by how precisely they state the job, the why, the guardrails and what done looks like, and they treat a bad output as feedback on their prompt, not something to patch with long iteration. They make output easy to review with output styles and visual artifacts, automate work that was never getting done, put decisions where Claude can read them, such as meeting notes, and regularly prune skills, context and scheduled tasks.

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 →
Chapters

Where the time goes.

00:00 – 00:46

01 · The two studies and the gap

400,000 sessions and Anthropic's own staff show top users getting working results more than twice as often.

00:46 – 05:09

02 · Lesson 1: Communicate clearly

A four-part brief: the job, the why, the rules, and what done looks like, with an example sales-report prompt.

05:09 – 08:34

03 · Lesson 2: Give better feedback

Treat bad output as feedback on your prompt, use slash rewind, ask for variations, and raise effort before switching models.

08:34 – 11:58

04 · Lesson 3: Make work faster to review

Set an output style and make outputs visual with Claude artifacts.

11:58 – 13:55

05 · Lesson 4: Automate work you never get to

27% of Anthropic's Claude-assisted work wouldn't have happened otherwise; keep a never-get-to list.

13:55 – 16:50

06 · Lesson 5: Give Claude your decisions

Move decisions into places Claude can read, use meeting notes, and run a scheduled task over action items.

16:50 – 18:18

07 · Lesson 6: Keep your setup lean

Prune skills, context files, connectors and scheduled tasks regularly; the detox skill audits five areas.

18:18 – 19:07

08 · Wrap-up and resources

The best AI users are the best managers; free resources and course links.

Atomic Insights

Lines worth screenshotting.

  • Anthropic defined an expert by how precisely they described the task and who corrected whom, not by job title or coding background.
  • Beginners got a verified result only 15% of the time; experts got there more than twice as often.
  • Managers scored highest of all, even on coding tasks, because they are used to explaining clearly what they want done.
  • The best Claude users make about 70% of the planning decisions but only 20% of the execution decisions.
  • Explaining why you want something lets the model make the small decisions the way you would on any long task.
  • Emphatic words like CRITICAL or you MUST make newer models overdo it; Anthropic's own guide says to soften them.
  • If you never define what done looks like, the model stops when it thinks it's done, which is a guess.
  • Beginners gave up after a failure 19% of the time; everyone else gave up only 5% to 7% of the time.
  • A bad output is feedback on your prompt: rewind to the first prompt and fix it instead of stacking corrections in the context window.
  • Switching models mid-chat means the whole chat is re-read at full price because you lose the cache; use a handoff prompt in a new chat.
  • Anthropic staff would need half their work week just to read what their agents produce, so the human becomes the bottleneck.
  • Around 80% of the time in Anthropic staff sessions, Claude sits idle waiting for a human.
  • 27% of Claude-assisted work at Anthropic wouldn't have been done at all without it.
  • Ask 'what have I never done because it never felt worth my time' instead of 'how do I do this faster'.
  • If Claude can only read what is written down, decisions made in private chats and meetings are invisible to it.
  • Anthropic's own standing instructions for Claude grew from a few hundred lines to 8,500 in a year, and its best users keep cutting back.
Takeaway

Great Claude use is great management

WHAT TO LEARN

The gap between beginners and experts is communication and oversight, so brief clearly, fix prompts instead of patching outputs, and keep Claude's context and setup tidy.

02Lesson 1: Communicate clearly
  • Brief Claude like an employee: describe the whole job, give the reason, list only the guardrails that matter, and define what done looks like.
  • Decide what to do but leave how to do it to the model; top users make about 70% of plan decisions and 20% of execution decisions.
  • Soften emphatic language such as CRITICAL or MUST, since newer models follow it so closely they overdo it.
03Lesson 2: Give better feedback
  • When output goes wrong, rewind and fix the original prompt with the missing context and a check, instead of stacking corrections in the chat.
  • Ask for several variations, then raise the effort level one step before changing models, and use a handoff prompt in a new chat if you do switch.
04Lesson 3: Make work faster to review
  • Set an output style that leads with what needs your attention, whether the goal was met, a short path summary, and anything that went wrong.
  • Ask for outputs as one-page designs, sheets, whiteboards, dashboards or docs so review and feedback are faster and shareable with a team.
05Lesson 4: Automate work you never get to
  • Keep a never-get-to list of neglected jobs and review it weekly with Claude to find skills or scheduled tasks worth building.
06Lesson 5: Give Claude your decisions
  • Put decisions where Claude can read them, add a CLAUDE.md line to read today's meeting notes, and use a scheduled task to draft work from action items.
07Lesson 6: Keep your setup lean
  • Audit skills, context files, connectors and scheduled tasks every couple of weeks, and delete rules that no longer match how the model behaves.
Glossary

Terms worth knowing.

Verified result
An outcome from a Claude session that was checked and confirmed to actually work, which Anthropic used to compare beginners with experts.
Output style
A Claude Code setting that changes how every response is structured, for example the built-in concise style that leads with the result and drops recaps.
Slash rewind
A Claude Code command (also double Escape) that returns the conversation to an earlier prompt so you can edit it and rerun from there.
Artifacts
Live Claude outputs such as docs, dashboards, whiteboards, sheets, slides and motion that you can edit, comment on and share instead of reading plain text.
CLAUDE.md
An instruction file Claude reads at the start of every session, used to store standing rules and context.
Scheduled task
A recurring Claude job that runs on a timer, such as checking hourly whether a meeting ended and working through its action items.
Handoff prompt
A summary prompt written by the current model so you can paste it into a fresh chat instead of switching models mid-chat.
Resources

Things they pointed at.

05:22toolInterview me skill
06:03toolFireflies
10:40productClaude artifacts (docs, dashboards, whiteboards, sheets, slides, motion)
14:45productClaude meetings
18:10toolDetox skill
Quotables

Lines you could clip.

18:38
“The best AI users are the best managers. Whatever job position you're in with AI, you're now a manager.”
clean one-line thesis→ TikTok hook↗ Tweet quote
07:30
“The best users don't get it right every single time. They're just much better at correcting and giving feedback.”
counterintuitive and self-contained→ newsletter pull-quote↗ Tweet quote
The Script

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.

Entropic recently published two studies on how people actually use Clod. One looked at around 400 ,000 Clod sessions from 235 ,000 people and the other looked at how their own staff use Clod every day. And I learned that the gap is huge.
Entropic's own staff now have AI agents working six times their own hours and the best Clod users get a working result more than twice as often as normal users and get more than 50 % worked on. So in this video, I'll break down the six main lessons from these studies on what the best Clod users do differently. exactly how to apply each one today so you can get the most out of ai and i'll give away all of the prompts and skills i use in this video for free in the first link in the description and if you're new here i'm ben i'm a three -time founder and i run an ai agency an ai second brain app and one of the largest ai communities where i help professionals and business owners get real results with ai So the first lesson and takeaway from these studies, and probably the most important one, is learning how to communicate with AI clearly.
Now, in the study of 400 ,000 cloud code sessions, one of the clearest differences between people who got great results and people who didn't was how clearly they told cloud what they wanted. Beginners got a verified result, so something that actually worked only 15 % of the time, while experts got there more than twice as often.
And the interesting thing is, Entropic didn't define someone as an expert based on their job title or something. defined an expert based on how precisely they described what they wanted what they asked Claude to check and whether they were the one correcting Claude or Claude was correcting them. So in short how clearly they communicated.
Now some of your ability to communicate a task well is of course related to how much experience and knowledge you have around it but it does have a lot to do with how clearly you can communicate. For example managers actually scored the highest of all even on coding tasks versus actual coders and Entropix thinks this might be because managers are used to explaining clearly what they want done.
And with AI being able to automate more and more work and new models like Opus and Fable being able to do tasks end to end, knowing what to work on and the skill of clearly communicating that to cloud will become one of the most important skills of the future across any job. And you can already see this inside of Entropic.
One of their staff said the strategic part of their job went from about 20 % to 90 % of the job. So what can really separate you from most people when using AI and in the future is by actually putting effort into getting clear on what you want and writing clear prompts not only because you'll get far better outputs but also because you're consciously training yourself to become a better communicator and honestly therefore a better manager and there's a simple mental model i've been using which you can go through for every single prompt in order to know you've communicated everything you should communicate to get the best possible outcome and it's very similar to how a manager would brief an employee so i recommend using a voice transcription tool and going to the following four points for every task you give cloud The first part is describing the job it needs to do.
Now with these newer models, you want to describe the entire task it needs to do, not just one step or a list of steps. And Tropic found that the best cloud users make about 70 % of the plan decisions, but only 20 % of the execution decisions. So in other words, you want to decide what to do, but give cloud room to figure out how to do it because these new models are often better at finding the path than we are, especially for complex tasks like research.
Now, the second part is the why, just like with you. Every instruction works better with a reason behind it. Now, Entropic's own research finds that explaining the intent behind a request produces far better outputs with these new models because on any long task, a model will run into small decisions it needs to make.
And when it knows why, it would make them in a similar way you would. Now, the third part is the rules or guardrails. And I like to think about this as just thinking quickly for a minute on what potentially could go wrong when giving it this task and writing those things down as rules or guardrails.
But only write down the ones that really matter. Like what clods shouldn't touch or shouldn't scent. And you want to write them like you talk to a colleague, ideally, again, with a reason behind it.
And while many people are used to using capital letters to emphasize, like very important, Entropic's own guide says to soften this kind of language because new models listen so closely that words like critical or you must actually make them overdo it. Now, the fourth part is describing what done looks like. Newer models like Fable and Opus stop only when it thinks it's done.
So if you've never said what done means, it just guesses. So getting in the habit of actually describing or showing what an end result should or could look like makes a huge difference. Ideally, you can give it a couple of simple checks or examples.
For example, every point should link to a source, the summary fits on one page, or it should look like this report we made last month. You can then also tell it to not stop until every check passes. Boris Cherny, who created Cloud Code, says giving the model a way to verify its own work is probably the single most important thing that people do not get right.
So for example, if you work... on a sales report you can just say build me a one -page summary of last month's sales calls because i want to see where deals stall in the sales process only use the calls in fireflies because that's the tool we use for our sales calls and done means every reason actually links to a call with quotes and applying this framework doesn't mean your prompts have to become really long necessarily but just getting into the habit of covering these four points for every prompt will get you far better outputs and will help you build that communication skill that is becoming more and more important in the future of work and sometimes you'll notice you can't explain a task clearly yet through those four points and that usually means you don't know exactly what you want yet and that's usually when i use my interview me skill because it keeps asking me questions until i'm clear on what i actually want and then writes the prompt for me using these four parts now i've put that skill and the full prompting framework in a free resource in the first link in description below but even when you communicate clearly cloud will still sometimes get things wrong and how you respond to that is the second lesson from these studies getting good at giving cloud feedback now in that same study when something went wrong
what they defined as beginners, just gave up 19 % of the time, while everyone else gave up only 5 % to 7 % of the time. And more importantly, when things went wrong, experts still got it to a desired output four times as often as beginners. So the best users don't get it right every single time.
They're just much better at correcting and giving feedback. And Entropic says exactly why. Part of the value of expertise is the ability to steer the agent in the right direction.
In fact, one of the things they used to rate someone as an expert was... Simply this, do you correct Clot or does Clot end up correcting you? And while many people tend to correct Clot with what's wrong right away and often get into long, sometimes frustrating iteration loops, Entropic found that their best users treat a bad output more as feedback on their own prompt.
Because again, in most cases, the bottleneck is our ability to communicate. And besides that, with newer models, what often gets you to a good output far faster and more reliably than long iteration loops is fixing your initial prompt. Now, many people don't know this, but in the Clot code tab, you can just type slash rewind or press escape twice.
It can go back to an earlier prompt, add what was missing and run it again from there. So let's say you ask Cloud to reply to a client email and it's far too stiff. Instead of typing, make it friendlier, then not too friendly, then you forgot the deadline, filling the context window with wrong outputs.
Just do slash rewind or double escape and add what was missing to your first prompt. For example, the deadline and two past emails for tone of voice. And even better, add in a check for Cloud.
for your show me, check it mentions the deadline, and sounds like my past emails. As Entropic puts it, Cloud performs best when it has a clear target to check its work against. And besides this being often far more efficient, it's another opportunity to analyze and practice your own communication skill.
And if you really can't tell why Cloud went the wrong way, you can just ask it. Amanda Askall from Entropic does exactly this. When Cloud does something unexpected, she just asks it why, and then she fixes that in the initial prompt.
Now, sometimes, of course, you do need to iterate with Cloud because you only... know what you want once you see a first version or you want to improve a specific part but again you want to avoid the temptation of only telling it what it did wrong good feedback says what's off why and what good looks like again it's the difference between a good and a bad manager and a simple trick i use to speed up my workflows a lot is to ask for a few variations instead of one because the chance that one of them is closer to what you want is far higher which then makes it much easier and faster to steer cloud in the right direction and only after you've tried doing this and dropping recommends raising the effort level by one level so opus on medium would become opus on high and only if that's still not enough you probably need to use a better model but you want to make sure you do that in a new chat because switching models mid -chat costs you a lot of money you lose the cash so the entire chat gets re -read at full price which is why i would recommend asking your current model to create a handoff prompt which you can then paste into a new chat instead of switching a model mid -chat the third lesson is to make clouds work faster to review now the best users use cloud for most
of their day -to -day work. Entropic's own staff now have agents working six times their own hours. But all of that work still has to be reviewed by a human and Entropic found that their staff would need half their work week just to read it all.
And besides that around 80 % of the time in Entropic's staff sessions Claude is just sitting idle waiting for a human to respond. Interestingly what they defined as experts got also far longer answers back. About 3200 words per message against 600 for beginners.
In short the better you get the more you become the bottleneck. not clot and that will only get worse as agents take on more and more work so the answer isn't to read more it's to get clot to respond in a way that is more human digestible and there are two main ways to do this the first thing is to set an output style now you can set a custom output style that makes clot respond in a specific way each time clot for example has a built -in one called concise which makes clot lead with the result and leave out the fluff narration and recaps and to turn it on you can just type slash output style concise in clot code but you can also set up a custom one i for example i've built a custom one so that every response starts with four things in that order first what needs my attention or input by default cloud often puts that at the very end and since most people read top down it speeds things up second whether it actually achieved the goal or not because it's one of the first things i want to know is it actually done or not third a short summary of how it got there so i can spot a wrong turn without reading every step and fourth anything that went wrong and why because problems are exactly what i can't afford to
myth and cloud tends to not mention it when it wasn't able to actually read an article or something like this or it often gets lost in a long answer now it also instructs to keep outputs brief so i'm never reading a long wall of text you can copy my output style from the screen right now or copy it from the free resource link in the first link in the description now you can also of course make your own just ask cloud to create an output style with your instructions then you run slash output style to switch it and then the second which has been a game changer for me and my team is to make outputs visual andre karpathy put it very nicely visuals are like a highway for your brain and just asking ai to output things in a one -page html can really speed up the way you work with ai by a lot now the nice thing in cloud is that we now have these live artifacts cloud docs dashboards whiteboards sheets slides design and motion that cloud can use to make things visual instantly me and my team now use artifacts for almost anything we do anything i need to read a review like a research or a report i ask a lot to output it in a one -page design or in a slideshow for brainstorms i ask a lot to
lay it out on a whiteboard. For tables, I ask Cloud to put it in a sheet. For any analysis use case, I ask Cloud to create a Cloud dashboard.
And any copy I need to iterate on, I just ask Cloud to output it in a Cloud Docs, which means you can iterate and edit that copy instantly. And for complex concept or report, you can even ask Cloud to output it as a short explainer video with Cloud Motion. And besides this being far faster to digest information, it also makes it faster to iterate and give feedback because of Cloud's comment feature in these artifacts and because you can make the final edits manually yourself and another big advantage is that your whole team can open up and collaborate on these artifacts instantly now before continuing very quickly if you enjoy this type of content you'll really enjoy my ai accelerator where you get access to my 30 -day cloud course that takes you from beginner to pro in less than one hour a day you also get access to a full course on growing an ai business based on everything me and my team have learned over the last three years and unlimited one -on -one live ai consulting with my team and a big library of skills templates and resources across all business departments so
If you want to dive deeper into AI, definitely check it out in the second link in the description below. Now, the fourth lesson is to automate the work you never get to, not just your day -to -day tasks. Because in one of Entropic's internal study, their staff estimated that 27 % of their cloud -assisted work wouldn't have been done without it.
So more than a quarter of what they use cloud for is work that simply wasn't happening before. And in their latest report, their non -technical staff went from two coding sessions in a month to 90 in less than a year. Now, to be clear, automating your common day -to -day tasks is probably...
where you want to start but when they become more efficient it also frees up time to do new things and then expanding the question from how do i do this faster to what have i never done because it never felt worth my time is what the best users do differently because some of the biggest wins come from using ai for work that wasn't happening at all alex albert from entropic described his own habit as shutting down that initial dismissal he has when a task feels not worth his time we do this a lot internally too for example following up with leads in our crm that nobody followed up with over the last year, getting a detailed customer report by letting Cloud go through all of our support calls or opening and starting new marketing channels because our content processes are becoming far more efficient.
These are things we know we should do, are nice to have, but never had the time for or weren't feasible at all before we had AI. And what I've noticed is that a lot of what feels like a big project isn't one anymore. For example, I always put off doing shorts because it felt like a lot of work until I sat down for two hours building a skill that turns my long form scripts to short scripts and a skill to automate some of the video editing.
So what I recommend is keeping a simple never get to list with all of the jobs you keep putting off or things you should probably do. And then once a week or month, go through it with Cloud and ask which of these can we do and can any of them become a skill or a scheduled task? Because often all it takes is sitting down for an hour or two, thinking through the process and automating it.
And often once it's set up, it just runs. Now, the more work Cloud takes on, the more it needs to understand your context and decision. Which brings me to the fifth lesson.
which is to give cloud your decisions. Because as cloud takes over more and more of the doing, what's left for us is mostly the deciding. Now in the cloud code study, again, people made about 70 % of the planning decisions, but only 20 % of the execution decisions.
And those decisions mostly happen in meetings and in chats. At Entropic, the median employee now spends 20 % more time in meetings than a year ago. But cloud only knows what it can read.
And if it's working on your day -to -day tasks, it also needs context on your day -to -day decision -making. users move their decisions to where cloud can actually see them and tropics own report says when ai does more and more of the work making your decisions where everyone can see them and not just in your head or in private chats is critical and for most businesses the place where most decisions get made is in meetings for example we have a one hour call every day where we decide on new videos we give feedback on work and set the tasks for the day or week now of course many people have been using meeting transcription tools like fireflies which their ai can use but with cloud's new meetings feature it's even more powerful because cloud can join your calls take notes and write down every decision and use this context directly in the same tool you already use to automate your work and this means cloud can use that context feedback and decisions from your daily or weekly meetings to give you better outputs but it also means cloud can instantly start executing on action items on one of our calls for example it found 62 action items across my team and 48 of them cloud could actually do directly so i highly recommend exploring cloud meetings if you haven't yet but at the moment is limited
google meet so you can also use a meeting note taker like fireflies of course and to make sure cloud actually uses those notes in your day -to -day sessions you can just add a line to your cloud .md which is the instruction file cloud reads at the start of every session that sounds something like before any work read the meeting notes of all of the calls that happen today that way every every task automatically takes your latest context and decisions into account now you can just update your cloud .md by just telling cloud to update it but even more powerful it can also start executing on any action item that relevant right away so instead of you and your team telling claw what to do after the meeting you just set up one simple scheduled task that checks every hour if a meeting ended goes to the action items and starts working on every single action item it can work on and shows you the drafts to check before anything actually gets sent or published because then five or ten minutes after a meeting you walk into a review queue with cloud not a to -do list now you can get the full prompt to set up that scheduled task i use by taking a screenshot here or get it in the free resources below and because the entire team
cloud reads the same notes, feedback also reaches everyone else's agents. So when I give our editor feedback on a call, his cloud picks it up straight away. In one case last week, for example, his scheduled task immediately made adjustments to the video based on my feedback from the call alone.
Some people believe this is where work is heading. Meetings become the place where you give feedback, set direction and make decisions and AI executes on the work. Now I have a full video on cloud meetings where I show you exactly how to set this up.
I'll link it in the description below too. Which brings me to the last lesson and the one almost nobody talks about which is keeping your setup lean because the more you use cloud the more you're going to build more skills more connectors more context more scheduled tasks and hundreds of chats so your chats become impossible to find your skills power up cloud still follows rules that changed months ago and scheduled tasks fail in the background without you noticing this even happens at anthropic where the standing instructions how to call it their staff wrote for cloud grew from a few hundred lines to 8500 in just one year so what sets the best users apart is that they don't just keep adding they also keep cutting back boris journey who created cloud code says his own setup might be surprisingly vanilla he even recommends deleting your cloud .md your skills and your hooks every six months just to see what the new model can do without them now i personally don't go that far but i do clean up my own setup regularly and the first time i did cloud actually found 239 things that were old duplicated or never used again it's the same as managing a team a good manager doesn't just add new processes they also drop the ones that no longer work and the
best cloud users do exactly that with their own cloud setup and a big part of the uploaded setup also loads into every single chat so instead of getting better the more you add cloud actually gets slower messier and more expensive so i've actually built a skill to make this really easy for you which is called the detox skill it audits and cleans up the five parts of your cloud setup your interface your skills your memory and context files your connectors and your scheduled tasks it puts everything it finds on one page with a suggested fix for each item and once you approve it cleans up and organizes your old chat, flags duplicated or outdated rules in context and fixes them, flags failing connectors and scheduled tasks, and much, much more.
I recommend running the skill on a schedule every two weeks to keep your cloud setup lean. And you can also download it for free in the first link in the description below. In essence, it really comes down to this.
The best AI users are the best managers. Whatever job position you're in with AI, you're now a manager. Now, again, you can get all the free resources I mentioned in this video in the first link in the description.
And if you want to get access to unlimited one -on -one AI... consulting with my team my full 30 -day beginner to advance cloud course and a community with professionals and business owners at the forefront of ai definitely check out my ai accelerator in the second link in the description as always thank you so much for watching if you enjoyed this video i'd highly appreciate a like and subscribe and if you want to learn more about working better with cloud you can check out the video here above
The Hook

The bait, then the rug-pull.

Two Anthropic studies, one covering 400,000 sessions and one covering its own staff, point to a large gap between ordinary and top Claude users. The video turns the gap into six habits you can apply today.

Frameworks

Named ideas worth stealing.

02:43list

The four-part brief

  1. The job
  2. The why
  3. The rules
  4. What done looks like

A manager-style brief for every prompt: describe the whole task, give the reason, list only the guardrails that matter, and say how to check it is finished.

Steal forany Claude or Claude Code prompt
09:32list

Four-point output style

  1. What needs my attention
  2. Whether the goal was achieved
  3. Short summary of how it got there
  4. Anything that went wrong and why

A custom output style that puts the most review-relevant information first and keeps responses brief.

Steal forreviewing long agent runs
CTA Breakdown

How they asked for the click.

VERBAL ASK
18:57link
“Get all the free resources in the first link in the description.”

Free prompts and skills offered throughout, with a paid AI accelerator pitched twice.

Storyboard

Visual structure at a glance.

open
hookopen00:00
four-part brief
valuefour-part brief02:43
review faster
valuereview faster08:29
lean setup
valuelean setup16:50
CTA
ctaCTA18:57
Frame Gallery

Visual moments.

One-click upgrade to your Google

Get more breakdowns in your search results

Add Modern Creator as a preferred source and Google shows you more of our breakdowns in Search, Top Stories, and AI Overviews. It only changes what you see, and you can undo it in your Google settings anytime.

Add to Preferred SourcesOpens your Google source preferences with us pre-loaded. Tick the box and you're done.
Watch next

More from this channel + related breakdowns.

13:15
Ben AI · Tutorial

Anthropic's 7 New Rules for Prompting Claude 5 Models

A YouTuber pulls Anthropic's own internal prompting guides for Opus 5 and Fable 5, cross-references them against Claude Code creator Boris Cherny's Y Combinator talk, and distills them into seven concrete habits.

September 2nd
13:40
Ben AI · Tutorial

Claude 2.0 is Insane (All Updates Explained)

A walkthrough of the rebuilt Claude Code desktop app: new layout cards, projects with nested sessions, side chats, split view, the docs and design tools, and four slash commands worth knowing.

October 25th