Загрузка видео...
Не удалось загрузить видео
I don't understand why everyone isn't doing this yet. Anthropic released a guide for Claude that makes work much more efficient while burning a fraction of the tokens Main Idea: Agent Teams in Claude Code Instead of burning the expensive Opus 5.5 context on every line of code, tasks... show more
295,908 просмотров • 1 день назад •via X (Twitter)
Комментарии: 54
This is like a workshop with specialists: parallel worktrees reduce collisions, while a dedicated critic catches integration failures before they reach the main branch.
A very accurate comparison, brother
@Bober_smart what's often overlooked: define clear task boundaries for each agent. Avoids overlap and wasted resources. Hit this wall myself while building!
Yes, that’s the obvious thing
@Bober_smart haha, right? Sometimes the obvious stuff gets overlooked. Anything else you think people miss?
As a rule, people pass up many opportunities because they think implementation is too complex, in reality, however, it takes just five minutes to grasp the situation, set it up once, and automate the processes
@Bober_smart exactly! I think people get intimidated by the setup more than the actual process. Ever had that happen?
In reality, people make things complicated for themselves
@Bober_smart right? we love to overthink things!
That is simply how human thinking is wired
Claude handles the team well but costs less than rivals do. This saves time and money for builders who value efficiency over raw speed alone
You have a big advantage here
I’ll do it today, thanks
Always happy to help
And indeed, why don't people do that?
Many people might not take this seriously
Giving each model a specific role makes sense
After all, each model is responsible for a separate process
All that remains is to teach this swarm to trade shields on their own.
That would be great
excellent recommendations
I'm glad you liked it
Agent teams are becoming the real productivity multiplier
Yes, that is absolutely true
You found the real cheat code.
Yes, indeed, that is how it looks
damn, this is really, really cleverly done it's weird that I haven't used this before
I hope your workflows will move along better now
Curious how you measure if the swarm is actually cheaper. Do you track tokens per merged PR vs. a single Opus session? That's the number I'd want to see before switching
Yes, the costs really are lower, in the post, I included a link to an official source that clearly shows the price difference
less context waste, more focused work
And more positive results
it really helps avoid wasting expensive intelligence on routine tasks
this is a much smarter way to use agent teams
Actually, it is real gold
is sonnet 5.5 safe for prod code?
Yes, of course, it is designed for that
'不用Opus烧每一行代码'——这话每半年被重新包装一次。Agent Teams确实有用，但别指望配好就自动省token，还是得看你怎么切任务
this is very helpful dude, thanks for sharing!
what specific parts of the guide do you find most impactful for efficiency?
saving time is the real feature people should use guides like this more
good work by the team #XCFDX
GLiDE is better than Jev on Decision Index: 64.81 vs 57.91, and 83.5 vs 75.1 in tools. I’d test it in that file/tool-selection layer before making Jev the default.
i’ve been wasting tokens apparently
Real substance in every post, keep them coming
Yes, they did, but they still lack deep dream memory and governance. You get that here
This highlights an important shift.
Great structure, I like this better than using a third party tool. So do you start all your sessions in the /second-brain folder? Where do you store your local project data where agents actually write code? Outside this folder? Asking as I will typically start the session inside the project folder that I am working on /dev/project and /second-brain may sit in next to it at the root of my user folder …especially if I want to use it with multiple agents (i.e. Claude, Codex, Grok, etc.) How would I reference or use the brain or specific project folder? Hope this makes sense Looking for ideas how to effectively implement. Thank you!
Which open-source project did you use to generate this video?
多智能体可以并行做任务，但各会话都要消耗上下文，还会产生协调开销。Claude Code文档提示，团队模式比单会话用更多 token；把部分任务交给较便宜的模型可能降低费用，不能据此推断总 token 也会减少。
That's really cost-effective, task division cuts the cost a lot. Gotta try it later.
why people not making it
The 16ms script layer is the part I’d steal first. No reason to waste an LLM call on mechanical tasks
why isn't everyone doing this?
![i finally mastered how to maximise my opus 5.5 usage limits... the trick: let jev choose which subagent gets each task and how much effort it should use. [here’s how i’d wire it:] claude breaks the project into tasks. jev selects from predefined worker profiles. claude applies the selected settings and dispatches the work. → main session, medium: clarify the requirements, define what “done” looks like, and prepare the tasks → builder, low: small, clearly defined tasks with existing examples or patterns → builder, medium: tasks that connect multiple parts or need decisions within the approved plan → verifier, high: check requirements, probe edge cases, and report problems for the builder to fix jev gets the task’s scope, what’s uncertain, and the consequences of failure. it chooses from the profiles allowed for that task. your approval checkpoints stay in place. paste this into your next planning session: “use opus 5.5 with jev selecting the worker and effort profile for each task. first, check that a working jev integration is available and that this environment supports separate effort settings for subagents. check for configuration or environment overrides that could prevent those settings from taking effect. if anything is missing, explain what needs wiring before proceeding. break my request into tasks with clear ownership, dependencies, relevant context, and acceptance checks. keep small related tasks together when a separate subagent would add unnecessary overhead. keep the main session at medium effort. offer jev these worker profiles: builder at low effort for small, clearly defined tasks using existing patterns; builder at medium effort for tasks that connect multiple parts or require decisions within the approved plan; verifier at high effort for checking requirements and edge cases. give jev each task’s scope, uncertainties, dependencies, and consequences of failure. only offer profiles appropriate to the current stage. validate its selection before dispatching. use the actual jev integration; don’t simulate its decisions. if it abstains or returns an invalid choice, stop that handoff and ask me. show me the task plan and proposed assignments before starting. after approval, launch the selected workers with their assigned effort settings, relevant context, file ownership, and completion checks. let me review the result before verification. the verifier may add tests but must leave implementation code unchanged. have it report what passed, what failed, and what remains uncertain. send implementation fixes back to the builder, then recheck the affected parts. if a task repeatedly fails, examine the requirements and approach before increasing effort. report available total usage, including jev calls, worker calls, retries, and verification. don’t invent missing data. compare similar completed tasks before claiming savings.” steal this 👇](https://image.24vids.com/tw-2103852115341049897/amplify_video_thumb/2103851843214614528/img/QwYdKECdjfJZgMMK.jpg)
