[flagged]
Agentic Context Management: Memory and Cost as Architecture Problems
31–40 of 42 posts
Re: Agentic Context Management: Memory and Cost as Architecture Problems
#32Context pollution and rot are probably more important than memory, because facts can usually be retrieved if the agent is good at following breadcrumbs. What's also the biggest killer is code rot. Agents are particularly good at death by thousand cuts. They implement something poorly, or incorrectly, or introduce a bad pattern into the project. Then they continue to amplify that badness over time, as they continue to…
Yes so regular human and agentic evaluation of the coding agent output, scoring it on specific criteria?
Re: Agentic Context Management: Memory and Cost as Architecture Problems
#33Im wondering how silent information loss is detected later and what exactly the validation score measures.
Re: Agentic Context Management: Memory and Cost as Architecture Problems
#34Earlier quoted context omitted.
It goes for all spheres of life, full stop. I’m not sure if agents struggle with this because they learned it from humans, or if they struggle with it because it’s a universally challenging problem, but it’s something we share with them.
The comment highlighted how LLMs exacerbate the issue by entrenching the preexisting issues.
Indeed, and reply to comment seemed a "yes and" -- As with humans.
It's curious how much of these could apply to either:
https://en.wikipedia.org/wiki/Reconstructive_memory
https://en.wikipedia.org/wiki/Misinformation_effect
And many mechanisms exacerbate issues by entrenching preexisting issues.
Re: Agentic Context Management: Memory and Cost as Architecture Problems
#35Earlier quoted context omitted.
> They implement something poorly, or incorrectly, or introduce a bad pattern into the project. Then they continue to amplify that badness over time, as they continue to copy from it on subsequent work. It spreads like a virus. > Keeping these seeds out of the project is very difficult, and cleaning up the rot is very difficult. My "aha" moment was when I realized this goes for all spheres of life where this tech is/…
It goes for all spheres of life, full stop. I’m not sure if agents struggle with this because they learned it from humans, or if they struggle with it because it’s a universally challenging problem, but it’s something we share with them.
"Out of the crooked timber of humanity, no straight thing was ever wrought"
Re: Agentic Context Management: Memory and Cost as Architecture Problems
#36Earlier quoted context omitted.
The comment highlighted how LLMs exacerbate the issue by entrenching the preexisting issues.
Indeed, and reply to comment seemed a "yes and" -- As with humans. It's curious how much of these could apply to either: https://en.wikipedia.org/wiki/Reconstructive_memory https://en.wikipedia.org/wiki/Misinformation_effect And many mechanisms exacerbate issues by entrenching preexisting issues.
I don't understand. The comment said if humans want to change the route, this tech makes it more difficult. Human inertia is X, inertia with this tech is X ^ Y. The Y is the issue being discussed.
Re: Agentic Context Management: Memory and Cost as Architecture Problems
#37I've found that a simple markdown knowledgebase (with some useful extensions like semantic search & git context) is all I need to improve the memory of my agents. Even my non-coding agents have a memory repo.
Here's my implementation: https://hraness.com/kb
Re: Agentic Context Management: Memory and Cost as Architecture Problems
#38[dead]
Re: Agentic Context Management: Memory and Cost as Architecture Problems
#39[flagged]
Re: Agentic Context Management: Memory and Cost as Architecture Problems
#40Earlier quoted context omitted.
> They implement something poorly, or incorrectly, or introduce a bad pattern into the project. Then they continue to amplify that badness over time, as they continue to copy from it on subsequent work. It spreads like a virus. > Keeping these seeds out of the project is very difficult, and cleaning up the rot is very difficult. My "aha" moment was when I realized this goes for all spheres of life where this tech is/…
It goes for all spheres of life, full stop. I’m not sure if agents struggle with this because they learned it from humans, or if they struggle with it because it’s a universally challenging problem, but it’s something we share with them.
You full-stopped too soon: It may happen everywhere, but it doesn't happen the same way everywhere or for the same reasons.
LLMs will create different kinds of corruption than humans, because the underlying mechanisms are different. Our ability to manage type of corruption will depend on how whether they can be predicted by math or intuition.