[cobolexpert]

This is something that continues to surprise me. LLMs are extremely flexible and already come prepackaged with a lot of "knowledge", you don't need to dump hundreds of lines of text to explain to it what good software development practices are. I suspect these frameworks/patterns just fill up the context with unecessary junk.

[vidarh]

You get to 80% there (numbers pulled out of the air) by just telling it to do things. You do need more to get from 80% there to 90%+ there.

How much more depends on what you're trying to do and in what language (e.g. "favourite" pet peeve: Claude occasionally likes to use instance_variable_get() in Ruby instead of adding accessors; it's a massive code smell), but there are some generic things, such as giving it instructions on keeping notes and giving them subagents to farm out repetitive tasks to prevent the individual task completion from filling up the context for tasks that are truly independent (in which case, for Claude Code at least, you can also tell it to do multiple in parallel)

But, indeed, just starting Claude Code (or Codex; I prefer Claude but it's a "personality thing" - try tools until you click with one) and telling it to do something is the most important step up from a chat window.

[cobolexpert]

I agree about the small tweaks like the Ruby accessor thing, I also have some small notes like that myself, to nudge the agent in the right direction.

[TeMPOraL]

There's no such thing as universal "good software development practices". There's only lots of opinions. Some are more popular, some less; some are language, domain or company-specific (or even tool-specific - see various webshit frameworks whose idiosyncrasies spill over to significantly alter coding style), and many exist only as historical baggage - but they all largely conflict with each other. And LLMs have seen them all.

Consider as an example, that "Clean Code" used to be gospel, now it's mostly considered a book of antipatterns, and many developers prefer to follow Ousterhout instead of Uncle Bob. LLMs "read" both Clean Code and A Philosophy of Software Design, but without prompting they won't know which way you prefer things, so they'll synthesize something more-less in between these two near-complete opposites, mostly depending on the language they're writing code in.

The way I think about it is: "You are a staff software engineer with 15 years of experience in <tech stack used in the project>" is doing 80% of the job, by pulling in specific regions in the latent space associated with good software engineering. But the more particular you are about style, or the more your project deviates from what's the most popular practice across any dimension (whether code style or folder naming scheme or whatnot), the more you need to describe those deviations in your prompt - otherwise you'll be fighting the model. And then, it's helpful to describe any project-specific knowledge such as which tools you're using (VCS, testing framework, etc.), where the files are located, etc. so the model doesn't have to waste tokens discovering it on its own.

Prompts are about latent space management. You need to strengthen associations you want, and suppress the ones you don't. It can get wordy at times, for the same reason explaining some complex thought to another person often takes a lot of words. First sentence may do 90% of the job, but the remaining 20 sentences are needed to narrow down on a specific idea.

[cobolexpert]

Maybe my initial message was overly harsh, I mostly agree with your points here. I think maybe the point of disagreement is exactly _how much_ extra prompt is necessary to approach 100% of the job, but this is quite hard to measure (obviously). Your point about latent space management is a good mental model to have IMO.

[raesene9]

I think avoiding filling context up with too much pattern information, is partially where agent skills are coming from, with the idea there being that each skill has a set of triggers, and the main body of the skill is only loaded into context, if that trigger is hit.

You could still overload with too many skills but it helps at least.

[epolanski]

> I suspect these frameworks/patterns just fill up the context with unecessary junk.

That's exactly the point. Agents have their own context.

Thus, you try to leverage them by combining ad-hoc instructions for repetitive tasks (such as reviewing code or running a test checklist) and not polluting your conversation/context.

[cobolexpert]

Ah do you mean sub-agents? I do understand that if I summon a sub-agent and give it e.g. code reviewing instructions, it will not fill up the context of the main conversation. But my point is that giving the sub-agent the instruction "review this code as if you were a staff engineer" (literally those words) should cover most use cases (but I can't prove this, unfortunately).

[vidarh]

I do think you're right that you should be cautious about writing too convoluted sub-agents.

I'd rather use more of them that are brief and specialized, than try to over-correct on having a single agent try to "remember" too many rules. Not really because the description itself will eat too much context, but because having the sub-agent work for too long will accumulate too much context and dilute your initial instructions anyway.

[Macha]

If I don't instruct it to in some way, the agent will not write tests, will not conform with the linter standard, will not correctly figure out the command to run a subset of tests, etc.