It's better if you think that the only thing that is really there is a context window.
They can add complex concepts and tools on top, but all that is is a different way to put things in the context window. Even the chat history on the web... You are not sending a message every time... It's not really a chat... the model is writing what it predicts will come next, like autocompleting a Word document that is written in a chat-like format.
So agents are like you, opening a new window and having the chat there, so you don't pollute the current window with all the tokens needed to process that question, and to use only the output here.
This is important bc of the effective context window problem. Models are more accurate the smaller the context is.
Hence, MCP tools are problematic. If you have registered many of them, the rules for using each one are added to your context, even if you don't use them.
Having a very extensive Claude.md file is also problematic.
You can use skills to instruct the model on which agents to use when requesting a specific thing. Antrophic says they have trained the model to discover on its own when to read the skill and follow the instructions you picture there, which can include Python scripts to run.
So yeah, agents help the model save context window for your current problem, skills help the model follow your instructions better, and instructions can include agent calling, and MCP is crap, you'd better ask the model to generate code to make that call
Oh, there are also slash commands. I don't really use them... if someone has a success story for them, I would love to know about it.
To solve it, you just don't allow your current context to use more than 50% of the total window size
To do that in Claude code, you have to use subagents and design small enough agents
Then you can use skills to make it remember every time the little details or the steps
More effectively, you use skills to tell the main thread when you go to use which agent.
If you don't understand anything I said, try to restate the important things to the model periodically, and keep your tasks small.
Use plan mode and make the model store, keep track of the progress on a markdown file, and when context is polluted, call /compact and then make it re-read the context from the files created
You can prompt it as simply as:
First, understand the login feature on the repo using subagents and create a document on docs/ for future reference. Then, understand the task at hand and create an implementation plan.
<task>
blah blah
</task>
Also, using XML tags makes the attention remember easily
father time? time by itself does not explain anything... is the new evidence that happens over time the one that teaches us. We are just doomed to a mostly monotonic growing time function. So we need action not that invitation to sit down and wait for the next time chunk to hit.
Usually this means you may probably get distracted easily. So you are good at maths and physics, but then there is this new concept on software engineering and what you do? maybe you switch and forgot for a while about the last topic you where becoming an expert.
There is one thing you are doing constantly and that is: Learning. You are becoming one of the best at learning a new topic quickly.
So is not really as in susans blog post that we are never going to have a single destiny... our destiny is to be the best at learning new topics in record time. This is certainly useful at a research team and new knowledge areas can really laverage someone like that. It's like the category theory expert who can found relations between mathematical objects.
My mindset towards this comes from an stoic maybe buddhist kind of perspective mainly influenced by Thich Nhan Nhat: So you want to be an expert in many things but there is no time for that? cool, embrace it, hug that feeling, accept it you don't need to suffer about it and maybe you already know this and you are looking for practical advice to get more done right?
So that is what I do, I embrace this fact, breath it out, and continue with the topic at hand.
Now getting back at the distractibility-factor... the real problem is when this starts making us not deliver important things here i'm with the person who commented about systems. Atomic Habits is a good book to make those systems and make it simpler for our biased existence to follow a predesigned path daily just as if you were your own parent and designed systems for the little you, the one that have to make decisions daily.
Honestly some psilocybin have helped me understanding this... and micro-doing it seems to add clarity and helps on the letting go.
This seems to be the opposite advice from what pg is talking about. The very last paragraph closes saying don't worry about the todo just follow your muse.
If you are forcing yourself that much to get it done maybe it's not the right thing to be working on anyway?
Report to IRS is not a bug is a feature you have to pay taxes somewhere and there are plenty of non double taxation agreements within the US and Latam.
It is art and years of preparation right there performing real-time.
So that guy criticizing is only auto declaring he is incapable of improvising on this matter.. and it's ok it's only for the best ones after all. It is called a challenge for a reason.
Yeah could be a vm fullscreen but have anyone thought this could be just random screenshots stolen from somewhere? I can make up a very nice story about a flying spaghetti monster given google images index.
Totally. Indeed the times i have been more driven and successfull at hacking something, has been on the most restricted environments looking for freedom. bypassing the proxy for example trough icmp tunnels... when you already have the freedom there is less passion to hack something
at least for common use cases like curl and wget you probably need a wrapper. I mean i love the idea of never leaving my cli, but... as developer and bash user im lazy... so please don't make type long curl commands. (Yeah I know I can add some aliases, but a wrapper could be handy). Anyway good job :)
Definitively +1 on the desktop app. And more that just bc that fits better on some workflows is bc the kind of features that a desktop app leave you to implement. Currently my team is using timedoctor, i hate its ui/ux but they have this feature: "track keyboard activity" that lone is not impressive but their desktop app using this feature after a configurable period of time can alert me to ask if im still working on the selected task and if i don't answer he fallback to a "time break" activity. That is great bc right now with timely i can forget about a timer and let it grow for ever and later will be hard to remember when did I really ended doing that.
Great ui/ux I'm on trial now.. and i love what you got... but w/o this feature my team will keep using ugly timedoctor, nothing personal btw ;)
