Modern Creator
Why Modern Creator?
Duncan Rogoff | Learn Claude Code · YouTube

How Anthropic Engineers Actually Use Claude Opus 5.5

A screen-recorded read-through of Anthropic's own prompting playbook for Opus 5.5, tip by tip, with a mid-video pitch for the creator's $9 Claude Code community.

Posted
2 days ago
Duration
Format
Tutorial
educational
Views
4.1K
36 likes
Part of the collectionThe Claude Opus 5 PlaybookEvery Opus 5 breakdown, synthesized into one page.
Read the playbook
Big Idea

The argument in one line.

Anthropic's own engineers published a prompting playbook for Opus 5.5 whose core lesson is to stop babysitting the model: hand it a whole task with a clear finish line, delete leftover 'think carefully' instructions, and let a task-list file carry state across a long run instead of the chat scrollback.

Who This Is For

Read if. Skip if.

READ IF YOU ARE…
  • You run Claude Code on multi-step coding tasks and keep getting interrupted by 'want me to continue?' prompts.
  • You copied old 'think step by step' prompting habits into your CLAUDE.md files and haven't revisited them since Opus 5.5 shipped.
  • You want the exact wording for a CLAUDE.md rule that tells Claude when to keep going versus stop and ask.
SKIP IF…
  • You've already read Anthropic's actual playbook post, since this video is a straight walkthrough of it with little added testing or commentary.
  • You're not using Claude Code specifically. Some tips, like the effort slider and the task-list-file pattern, are Claude Code features, not general prompting advice.
TL;DR

The full version, fast.

Anthropic's engineers published an official 'Getting the most out of Opus 5.5' playbook, and this video walks through it: give Opus 5.5 the whole task in one message with a stated 'done' condition, delete leftover 'think carefully' lines since the model now reasons before every reply by default, inject follow-up thoughts mid-run instead of waiting for a response to finish, name the design patterns you don't want instead of vague 'avoid generic' instructions, add a short CLAUDE.md rule telling it when to keep going versus stop and ask, hand large audits to subagents, and keep a running task list in a file like TASKS.md so state survives context compaction. The through-line: Opus 5.5 needs less hand-holding, so most of these prompting habits should be removed, not added to.

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 – 00:29

01 · Cold open

Anthropic released an official Opus 5.5 playbook; the creator frames Opus 5.5 as the most powerful current model and promises the full rundown.

00:29 – 01:03

02 · What's different about Opus 5.5

It works longer on its own, states plainly what it did, and thinks before every reply. Introduces the three things to try first.

01:03 – 01:32

03 · Tip 1: hand over the whole task

Give Opus 5.5 the entire task in one message, name what 'done' looks like, and say when it should stop and ask versus keep going.

01:32 – 01:52

04 · Tip 2: delete 'think carefully' lines

Opus 5.5 already reasons before every reply, so old 'think step by step' prompt lines are now just wasted tokens.

01:52 – 02:06

05 · Tip 3: read what it needs first

When a long run ends, read what Claude is asking for before responding. Framed as the core habit for learning to build with Claude.

02:06 – 02:36

06 · Learn Claude Code community pitch

Mid-video plug for the creator's own $9 Skool community ('Learn Claude Code').

02:36 – 04:01

07 · Worked prompt example and the effort slider

A full done-state prompt for migrating payment endpoints, why the 'reason you can't explain' clause matters, and using the Claude Code effort slider instead of think-carefully lines.

04:01 – 04:41

08 · Mid-run input and unwanted design styles

Type a follow-up and hit Enter to inject it into a running task; for design work, name the specific styles you don't want instead of a vague 'avoid generic' instruction.

04:41 – 06:09

09 · Steering a long run with a CLAUDE.md rule

A short CLAUDE.md rule telling Opus 5.5 when to keep going and when to stop and ask, which also fixes repetitive 'want me to continue?' prompts.

06:09 – 06:23

10 · Split work across subagents

For audits, migrations, or large-codebase reviews, explicitly ask Opus 5.5 to split the work across subagents and check each result.

06:23 – 07:21

11 · Keep the task list in a file

Ask Opus 5.5 to keep and update a task list in a file like TASKS.md, and read that file instead of the chat scrollback to see progress.

07:21 – 07:27

12 · Outro

Points to the full playbook link and a follow-up video comparing Opus 5.5 to other models.

Atomic Insights

Lines worth screenshotting.

  • Opus 5.5 now thinks before every reply by default, so any 'think carefully' or 'think step by step' line in a saved prompt is just burning tokens for nothing.
  • Removing 'think carefully' lines from prompts made Opus 5.5's replies start sooner with no measurable drop in quality, according to Anthropic's own testing.
  • The safest way to hand off a long task is to name the finish line explicitly, like 'every endpoint uses the new client and the test suite passes,' not just describe the task.
  • Telling Opus 5.5 to stop and ask only 'if a test fails for a reason you can't explain' works because the model is now good enough to diagnose and fix most failures itself.
  • Typing a follow-up message and hitting Enter while Claude is mid-task injects it into the live run, which is cheaper than waiting for the response to finish and re-ingesting full context.
  • Vague instructions like 'avoid a generic look' just swap one default AI style for another; naming the specific patterns you don't want, like cream backgrounds or pill-shaped buttons, actually works.
  • A short CLAUDE.md rule telling Opus 5.5 when to keep going versus stop and ask eliminates most of the repetitive 'want me to continue?' prompts.
  • The recommended stop condition is narrow: continue on anything that doesn't need input, but always stop before deleting data, force-pushing, or touching anything outside the repo.
  • For large audits or migrations, explicitly asking Opus 5.5 to split the work across subagents lets it coordinate parallel work with comparatively little oversight.
  • Keeping a running task list in a file like TASKS.md survives context compaction, because a long session eventually gets summarized and loses detail that a file on disk doesn't.
  • When a long run ends, the first thing to check is whatever Claude is waiting on you to approve or decide, not just the final summary.
Takeaway

How to prompt Opus 5.5 without babysitting it

WHAT TO LEARN

Opus 5.5 reasons and reports on its own, so most of the old prompting habits should be deleted rather than added to, and the real skill is naming a clear finish line and letting it run.

02What's different about Opus 5.5
  • Opus 5.5 works for longer on its own, states plainly what it did, and thinks before every reply, so it needs less step-by-step prompting than earlier models.
  • The three habits worth adopting first are handing over the whole task, deleting 'think carefully' lines, and reading what Claude needs once a long run ends.
03Tip 1: hand over the whole task
  • Give Opus 5.5 the entire task in one message instead of feeding it piecemeal instructions, and name the finish line explicitly, like 'every endpoint uses the new client and tests pass.'
  • State up front when you want it to stop and ask versus keep working, then let it run instead of checking in constantly.
04Tip 2: delete 'think carefully' lines
  • Opus 5.5 already reasons before every reply, so 'think carefully' or 'think step by step' lines left over from older prompts just add unnecessary tokens.
  • Anthropic's own testing found removing those lines made replies start sooner with no measurable drop in quality.
  • If you want a quick answer to something simple, just say 'answer directly' instead of leaving reasoning instructions in the prompt.
05Tip 3: read what it needs first
  • When a long run ends, read what Claude is asking for or waiting on before responding, rather than skimming the summary and moving on.
  • This reading habit is presented as the actual way people learn to build with Claude, not an optional step.
07Worked prompt example and the effort slider
  • A complete done-state prompt names the task, the finish line, and one exception for stopping, e.g., 'stop and ask only if a test fails for a reason you can't explain.'
  • That 'reason you can't explain' clause matters because Opus 5.5 is now good enough to diagnose and fix most of its own failing tests without being asked.
  • To control how much a response reasons without writing a prompt line for it, use the effort slider in Claude Code instead.
08Mid-run input and unwanted design styles
  • Typing a follow-up and pressing Enter while Claude is still working injects it into the live task, which is cheaper than waiting for the response to finish and re-sending full context.
  • For design requests, list the specific styles you don't want (cream backgrounds, italic accent words, numbered section labels, pill-shaped buttons) instead of a vague 'avoid generic' instruction.
  • A generic instruction like 'avoid a generic look' just swaps one default AI style for another; naming actual patterns is what changes the output.
09Steering a long run with a CLAUDE.md rule
  • Add a short rule to CLAUDE.md that tells Opus 5.5 when to keep going and when to stop and ask, since it otherwise pauses to report progress instead of continuing.
  • The recommended rule: keep going when a step doesn't need input, put status notes in the same message as the next action, and stop only when you can't continue without the user or before anything destructive like deleting data or force-pushing.
  • This same rule is what fixes the repetitive 'want me to continue?' prompts that come from not having a stop/go policy defined.
10Split work across subagents
  • For an audit, migration, or review across a large codebase, explicitly ask Opus 5.5 to split the work across subagents and check each one's result.
  • Early testers had Opus 5.5 coordinate parallel subagents on long audits and migrations with comparatively little oversight needed.
11Keep the task list in a file
  • For any run that will take a while, ask Opus 5.5 to keep its task list in a file like TASKS.md and update it as it goes, rather than relying on the chat scrollback.
  • This matters because a long run eventually fills the context window and gets summarized, which can lose detail that a file on disk survives.
  • Read the task file, not the conversation history, to see at a glance what's done and what's left.
Glossary

Terms worth knowing.

Effort slider
A Claude Code control that adjusts how much a model reasons before responding, used as the replacement for manually prompting 'think carefully' on Opus 5.5.
CLAUDE.md
A project-level instructions file Claude Code reads automatically, used here to store rules for when the model should keep working versus stop and ask for input.
Subagents
Separate Claude Code work units that can be assigned pieces of a larger task, such as auditing different services in parallel, then report results back for review.
Context compaction
The process where Claude Code summarizes or compresses older parts of a long conversation once it fills the context window, which can lose fine detail that a task-list file preserves.
Resources

Things they pointed at.

00:08linkAnthropic's 'Getting the most out of Opus 5.5' blog post ↗
02:06productLearn Claude Code (Skool community) ↗
Quotables

Lines you could clip.

00:45
“It works for longer on its own. It tells you plainly what it did, which is super nice. And it thinks before every reply.”
clean three-part summary of what changed→ TikTok hook↗ Tweet quote
02:41
“Stop and ask me only if a test fails for a reason you can't explain.”
exact reusable prompt line, tight punchline→ newsletter pull-quote↗ Tweet quote
03:13
“Removing any of those think carefully lines made the reply start sooner without a clear drop in quality.”
counterintuitive result stated plainly→ IG reel cold open↗ Tweet quote
04:24
“If you don't give it any direction, it's going to fall back into our generic like AI swap styles.”
names the generic-AI-look problem directly→ TikTok hook↗ Tweet quote
05:43
“Keep going when a step doesn't need my input. Put status notes in the same message as your next action.”
the exact CLAUDE.md rule wording→ 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.

The engineers at Anthropic just released the entire playbook for getting the most out of Opus 5 .5 and Cloud Code. If you haven't heard yet, Anthropic just released Cloud Opus 5 .5 and it is the most powerful model on the planet. It is outperforming Fable, it is outperforming GPT -6 Astra, and it is doing it all at a fraction of the cost.
So whether you are brand new to building with Cloud Code and wanna make sure you're using the best practices from the start, or you're someone who's been building for a while, you probably could use a refresh. So this is worth paying attention to.
So this is the whole playbook for getting the most out of Opus 5 .5. So Opus 5 .5 works well with the way you already use Claude, but a few things do behave differently. It works for longer on its own.
It tells you plainly what it did, which is super nice. And it thinks before every reply. So these are the three things to try first when working with Opus 5 .5.
So the first thing to do is to hand over the whole task, say what done looks like, and when you want Claude to stop and ask, then let it go to work. Rather than going back and forth, inserting little messages here and there, give it the entirety of what you want it to do. And then actually instruct Claude, like when it should alert you for user input, or if it should just work continuously.
I actually have a whole video creating these goal style prompts. It does use Fable 5 .1, but many of these same concepts apply. It'll help you structure your task and this idea of what done looks like in a super easy way.
So number two is to delete any think carefully lines because. Opus 5 .5 already thinks before every reply. So you used to prompt these models if you wanted them to go deeper, to think carefully.
And a lot of people who have been building with Claude probably have lines like this already in their ClaudeMD files or their working files. And so these are worth looking at again because it's just adding tokens and adding context that is not necessary anymore. Number three is a step that I think a lot of people skip.
It's when a long run ends, read what it needs from you first. I find that a lot of people skip this reading portion and that's how you learn to build anything with quad is actually by reading and responding. If you are someone who's learning to build, I run one of the top cloud code communities for building websites, apps, games, and more, and turning those builds into revenue.
If you want to join today for only nine bucks, just check the link in the description. So here's how to ask and say what done looks like. So give the whole task in one message, name the finish line, like the test pass when this happens, right?
Or every endpoint is migrated. Then just let Opus 5 .5 cook. So here's an example of how to prompt Opus 5 .5 by giving it a done state and when it needs to act.
ask you for permission. So migrate the payment endpoints from the old client to the new one. Done means every endpoint uses the new client, the old client is deleted, and the test suite passes.
Stop and ask me only if a test fails for a reason you can't explain. So I think this part is key for a reason you can't explain because these models have gotten so good at checking their own work. So it's possible that a test will fail, but Opus will figure out why it fails and then it'll go ahead and fix it.
So like I said, you can stop telling Opus to think hard or think carefully or think step -by -step because it does this automatically. So I think this part is interesting. When the Anthropic team was actually testing this, removing any of those think carefully lines made the reply start sooner without a clear drop in quality.
So if you want a quick answer to a simple question, just say so. So just tell it to answer directly. And as a pro tip, if you want to change how much it actually thinks, like just change the level of effort using the slider.
So the next thing that you can do, and you may already be doing this, is actually adding input to a running task. So sometimes you remember something while Cloud is already doing its thing. You can actually type a follow -up while Cloud is working and just hit Enter, and it will inject that message into the conversation.
This is often better than waiting for Cloud to finish its response before following up, because if you do that, then Cloud is basically ingesting all of the context again, which is just going to increase your tokens and increase your costs. So all you have to do is type the message and press enter while quad is working. So you can just say, hey, like also keep the old endpoints, names and aliases.
So if you are using quad for design work, name the styles you. don't want. So this is kind of a new way of thinking.
So when you ask for a page, an app, or an artifact, list the design habits that you want left out. Don't just give it direction. Tell it exactly what you don't want.
So if you don't give it any direction, it's going to fall back into our generic like AI swap styles. General instruction like avoid a generic look mostly swaps one default for another. So like that just doesn't work.
But listing specific patterns works a whole lot better. So actually like name them. So build a personal website with placeholder content.
don't, use a cream or off -light background, italic accent words and headings, numbered 1, 2, 3 section labels, monospace labels, or pill -shaped buttons. So this is how you can avoid creating these generic AI slop sites. So number two is all about steering a long run in cloud code.
So basically, if you're going to send cloud off to work on something like a long coding session, you can actually put a short rule in your cloud MD file about when to stop and ask and when to keep going. If you don't know how to do this, you can literally just tell Claude, hey, I want to add a rule in my CloudMD file that does this.
So right now, Opus 5 .5 will keep you posted as it works. So on a long task, it will sometimes stop to report instead of just going on. It'll give you like a summary that names the next step, but it won't actually do the next thing.
So basically inserting this rule, it will just tell Claude like, hey, like don't stop to report on these things. Like just keep working until you've hit these checkpoints. So you can tell Claude when to keep going and when to stop and ask.
keep going when a step doesn't need my input. Put status notes in the same message as your next action and then stop and ask only when you can't continue without me or before anything destructive like deleting data, force pushing something or changing anything outside of this repository. So if you keep getting this message where Claude says, want me to continue and you're just constantly saying like, yeah, like keep going, like adding the rule above is gonna help.
So next is specifically ask Claude to split big work across sub -agents. So if you're doing like a big heavy coding task, actually telling Claude specifically, hey, use sub agents for this.
It's actually going to split the task out into these separate little components. It's going to work a lot faster. It's going to save you on tokens and just going to be way more efficient in the long run.
This is actually one of my favorite tips in this whole list is keeping the task list in a file. This is especially important for these bigger coding sessions. So for a run, that's going to take a while, ask Opus 5 .5 to keep its task list in a file and update it as it goes.
So it's actually going to work on a task and check it off the list. But what's cool is it's actually going to read this file, not like its own memory or context window, because what happens is over time, this context window fills up and quad will compact it or compress. it or summarize it.
So basically, it's going to be missing pieces. But if you actually keep like a running list, it's going to survive all of that. And it will show you at a glance what's done and what's left to do.
And again, it makes this task list persist over these longer sessions. So you can actually just say like keep a task list and tasks .md, tick each item when it's done and add anything new you find. If you want to read the rest of the Opus 5 .5 tips from the Anthropic Engineers, I'll leave the link in the description where you can access it for free.
If you want to see why Opus 5 .5 is better than Fable 5 .1 and Aster 6. Just check out this video right here.
I'll see you over there.
The Hook

The bait, then the rug-pull.

Anthropic's own engineers published a prompting playbook for Claude Opus 5.5, and this video is a read-through of it: what to delete from old prompts, what to add to a CLAUDE.md file, and the exact wording that gets a long coding run to finish itself.

Frameworks

Named ideas worth stealing.

00:51list

Three Things to Try First with Opus 5.5

  1. Hand over the whole task and say what done looks like
  2. Delete any 'think carefully' lines
  3. When a long run ends, read what it needs from you first

Anthropic's suggested starting checklist for a first Opus 5.5 session, aimed at people who are still prompting it the way they prompted older models.

Steal forany onboarding doc for a first Claude Code session
03:27model

The Done-State Prompt Formula

  1. Give the whole task in one message
  2. Name the finish line explicitly
  3. State the one exception for stopping and asking

The structure behind the worked 'migrate the payment endpoints' example: task, done condition, and a narrow stop clause.

Steal forany long autonomous coding-agent prompt
05:40concept

The CLAUDE.md Stop/Ask Rule

  1. Keep going when a step doesn't need input
  2. Put status notes in the same message as the next action
  3. Stop and ask only when you can't continue without the user, or before anything destructive

A short rule Anthropic recommends adding to CLAUDE.md so Opus 5.5 stops pausing to report progress and instead keeps working until a real checkpoint.

Steal forany project's CLAUDE.md for long unattended runs
CTA Breakdown

How they asked for the click.

VERBAL ASK
02:06product
“If you want to join today for only nine bucks, just check the link in the description.”

Dropped mid-video between tip 3 and tip 4 as an aside for 'someone who's learning to build,' pointing to his own $9 Skool community rather than anything Anthropic ships.

MENTIONED ON CAMERA
00:08linkAnthropic's 'Getting the most out of Opus 5.5' blog post ↗
02:06productLearn Claude Code (Skool community) ↗
FROM THE DESCRIPTION
PRIMARY CTAWhere the creator wants you to go next.
OTHER LINKSAlso linked in the description.
Storyboard

Visual structure at a glance.

open
hookopen00:00
hand over the whole task
promisehand over the whole task01:03
Skool community pitch
ctaSkool community pitch02:06
CLAUDE.md stop/go rule
valueCLAUDE.md stop/go rule05:40
Frame Gallery

Visual moments.

open00:00Frame at 00:08 from How Anthropic Engineers Actually Use Claude Opus 5.500:08Frame at 00:15 from How Anthropic Engineers Actually Use Claude Opus 5.500:15Frame at 00:23 from How Anthropic Engineers Actually Use Claude Opus 5.500:23Frame at 00:27 from How Anthropic Engineers Actually Use Claude Opus 5.500:27Frame at 00:33 from How Anthropic Engineers Actually Use Claude Opus 5.500:33
open00:00Frame at 00:08 from How Anthropic Engineers Actually Use Claude Opus 5.500:08Frame at 00:15 from How Anthropic Engineers Actually Use Claude Opus 5.500:15Frame at 00:23 from How Anthropic Engineers Actually Use Claude Opus 5.500:23Frame at 00:27 from How Anthropic Engineers Actually Use Claude Opus 5.500:27Frame at 00:33 from How Anthropic Engineers Actually Use Claude Opus 5.500:33Frame at 00:39 from How Anthropic Engineers Actually Use Claude Opus 5.500:39Frame at 00:45 from How Anthropic Engineers Actually Use Claude Opus 5.500:45Frame at 00:51 from How Anthropic Engineers Actually Use Claude Opus 5.500:51Frame at 00:56 from How Anthropic Engineers Actually Use Claude Opus 5.500:56Frame at 01:02 from How Anthropic Engineers Actually Use Claude Opus 5.501:02Frame at 01:08 from How Anthropic Engineers Actually Use Claude Opus 5.501:08hand over the whole task01:16Frame at 01:19 from How Anthropic Engineers Actually Use Claude Opus 5.501:19Frame at 01:25 from How Anthropic Engineers Actually Use Claude Opus 5.501:25Frame at 01:31 from How Anthropic Engineers Actually Use Claude Opus 5.501:31Frame at 01:37 from How Anthropic Engineers Actually Use Claude Opus 5.501:37Frame at 01:43 from How Anthropic Engineers Actually Use Claude Opus 5.501:43Frame at 01:49 from How Anthropic Engineers Actually Use Claude Opus 5.501:49Frame at 01:54 from How Anthropic Engineers Actually Use Claude Opus 5.501:54Frame at 01:59 from How Anthropic Engineers Actually Use Claude Opus 5.501:59Frame at 02:03 from How Anthropic Engineers Actually Use Claude Opus 5.502:03Frame at 02:06 from How Anthropic Engineers Actually Use Claude Opus 5.502:06Skool community pitch02:11Frame at 02:16 from How Anthropic Engineers Actually Use Claude Opus 5.502:16Frame at 02:17 from How Anthropic Engineers Actually Use Claude Opus 5.502:17Frame at 02:25 from How Anthropic Engineers Actually Use Claude Opus 5.502:25Frame at 02:31 from How Anthropic Engineers Actually Use Claude Opus 5.502:31Frame at 02:37 from How Anthropic Engineers Actually Use Claude Opus 5.502:37Frame at 02:42 from How Anthropic Engineers Actually Use Claude Opus 5.502:42Frame at 02:48 from How Anthropic Engineers Actually Use Claude Opus 5.502:48Frame at 02:54 from How Anthropic Engineers Actually Use Claude Opus 5.502:54Frame at 03:00 from How Anthropic Engineers Actually Use Claude Opus 5.503:00Frame at 03:05 from How Anthropic Engineers Actually Use Claude Opus 5.503:05Frame at 03:11 from How Anthropic Engineers Actually Use Claude Opus 5.503:11Frame at 03:16 from How Anthropic Engineers Actually Use Claude Opus 5.503:16Frame at 03:22 from How Anthropic Engineers Actually Use Claude Opus 5.503:22Frame at 03:26 from How Anthropic Engineers Actually Use Claude Opus 5.503:26Frame at 03:28 from How Anthropic Engineers Actually Use Claude Opus 5.503:28Frame at 03:30 from How Anthropic Engineers Actually Use Claude Opus 5.503:30Frame at 03:34 from How Anthropic Engineers Actually Use Claude Opus 5.503:34Frame at 03:40 from How Anthropic Engineers Actually Use Claude Opus 5.503:40Frame at 03:46 from How Anthropic Engineers Actually Use Claude Opus 5.503:46Frame at 03:52 from How Anthropic Engineers Actually Use Claude Opus 5.503:52Frame at 03:58 from How Anthropic Engineers Actually Use Claude Opus 5.503:58Frame at 04:04 from How Anthropic Engineers Actually Use Claude Opus 5.504:04Frame at 04:09 from How Anthropic Engineers Actually Use Claude Opus 5.504:09Frame at 04:15 from How Anthropic Engineers Actually Use Claude Opus 5.504:15Frame at 04:21 from How Anthropic Engineers Actually Use Claude Opus 5.504:21Frame at 04:27 from How Anthropic Engineers Actually Use Claude Opus 5.504:27Frame at 04:32 from How Anthropic Engineers Actually Use Claude Opus 5.504:32Frame at 04:38 from How Anthropic Engineers Actually Use Claude Opus 5.504:38Frame at 04:44 from How Anthropic Engineers Actually Use Claude Opus 5.504:44Frame at 04:50 from How Anthropic Engineers Actually Use Claude Opus 5.504:50Frame at 04:56 from How Anthropic Engineers Actually Use Claude Opus 5.504:56Frame at 05:03 from How Anthropic Engineers Actually Use Claude Opus 5.505:03Frame at 05:10 from How Anthropic Engineers Actually Use Claude Opus 5.505:10Frame at 05:16 from How Anthropic Engineers Actually Use Claude Opus 5.505:16CLAUDE.md stop/go rule05:23Frame at 05:30 from How Anthropic Engineers Actually Use Claude Opus 5.505:30Frame at 05:36 from How Anthropic Engineers Actually Use Claude Opus 5.505:36Frame at 05:44 from How Anthropic Engineers Actually Use Claude Opus 5.505:44Frame at 05:50 from How Anthropic Engineers Actually Use Claude Opus 5.505:50Frame at 05:55 from How Anthropic Engineers Actually Use Claude Opus 5.505:55Frame at 06:03 from How Anthropic Engineers Actually Use Claude Opus 5.506:03Frame at 06:08 from How Anthropic Engineers Actually Use Claude Opus 5.506:08Frame at 06:13 from How Anthropic Engineers Actually Use Claude Opus 5.506:13Frame at 06:19 from How Anthropic Engineers Actually Use Claude Opus 5.506:19Frame at 06:22 from How Anthropic Engineers Actually Use Claude Opus 5.506:22Frame at 06:30 from How Anthropic Engineers Actually Use Claude Opus 5.506:30Frame at 06:36 from How Anthropic Engineers Actually Use Claude Opus 5.506:36Frame at 06:41 from How Anthropic Engineers Actually Use Claude Opus 5.506:41Frame at 06:47 from How Anthropic Engineers Actually Use Claude Opus 5.506:47Frame at 06:53 from How Anthropic Engineers Actually Use Claude Opus 5.506:53Frame at 06:58 from How Anthropic Engineers Actually Use Claude Opus 5.506:58Frame at 07:04 from How Anthropic Engineers Actually Use Claude Opus 5.507:04Frame at 07:10 from How Anthropic Engineers Actually Use Claude Opus 5.507:10Frame at 07:13 from How Anthropic Engineers Actually Use Claude Opus 5.507:13Frame at 07:20 from How Anthropic Engineers Actually Use Claude Opus 5.507:20Frame at 07:24 from How Anthropic Engineers Actually Use Claude Opus 5.507:24
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.