I built self-learning-skills because I noticed my agents often spent time poking around and guessing at things I had already solved in previous runs. I used to manually copy-paste those fixes into future prompts or backport them into my skills.
This repo streamlines that workflow. It acts as a sidecar memory that:
- Stops the guessing: Records “Aha moments” locally so the agent doesn’t start from zero next time.
- Graduates knowledge: Includes a CLI workflow to Backport proven memories into permanent improvements in your actual skills or docs.
It works with Claude Code, GitHub Copilot, and Codex and any other system that implements the https://agentskills.io specification.
At first pass this seems 1) incredibly useful for me 2) incredibly expensive for them, but after using it a bit I'm thinking it might be incredibly valuable for them because once I review and approve one of the options, they're essentially getting preference data on which of the approaches I felt was "best".
A minimalist, text-based memory system designed to naturally store and recall important events. It emphasizes simplicity, portability, and human-friendly structure by using six optional fields: who, what, when, where, how, and thing. These fields capture factual context clearly, deferring interpretation for later use or analysis.
Happy to clarify. The parent comment tackles a macro-economic utopia I never proposed.
My post was about individual level gains: using AI to automate routine work, offload mental load, and free time to think, create, or just live.
I’m not claiming to fix the global economy, nor denying real risks like job loss or scarcity.
Labeling me a "summer child" assumes I am naive about those challenges...another projection.
In short, I described a practical benefit available today, not a perfect future.
A thoughtful reply would engage with those points instead of refuting a position I never took.
I believe there's still plenty of margin to capture beyond merely overseeing AI. Could we reach a point where humans add no marginal utility? Maybe. I hope not, but we can't discount the possibility.
Totally hear you. Reinvesting saved time in higher-value / more interesting work doesn’t remove the structural-equity risk, which arguably might be the most challenging problem to solve.
In the end there are plenty of stories, but they're ones that are relevant. The story that the LLM gave feedback on was about flipping a raft on the Grand Canyon, the LLM's advice was that it felt unrelated to the point I was trying to make. That made me realize I had it in there more because I wanted to talk about the rafting Grand Canyon, vs. it being useful and entertaining to readers.
I think it was faster in that I would have never written the book without the LLMs. Essentially they unlocked the swirl of thoughts and notes that lived somewhere between my head, TextEdit, emails to myself, and anywhere else I stashed things.
It's like it unblocked the "hard part" (getting the words into a coherent form for others), while letting me focus on the "value parts" (my unique perspective / ideas).
It might not be that overall it saved me time, but it made it a hell of a lot more fun, so in the end I completed it - and maybe AI helping us see things through to completion is where we'll see a big unblock in human potential.
Even in that it likely depends on what you're measuring for waste. Is it wasted electricity, or is wasted productivity/opportunity time waiting for your machine to boot up?
The email is a good callout, chat would feel the same. What's interesting is the nuance in those channels, i.e. someone saying "hi" by itself in a work chat seems rude to me... just get to the point. But if it was switched in a real conversation, it'd feel rude without.
That was my gut too, but in general it's a question I've wondered about. i.e. what are the signals we send in our usage that will be beneficial to improving the models.
One of the things I found helpful about getting out of the specific / formulaic feedback was asking the LLM to ask me questions. At one point I asked a fresh LLM to read the book and then ask me questions. It showed me where there were narrative gaps / confusing elements that a reader would run into, but didn't realy on the specific "answer" from the LLM itself.
I also had a bunch of personal stories interwoven in and it told me I was being "indulgent" which was harsh but ultimately accurate.
Anthropic analyzed 700k real-world Claude chats to figure out what values it expresses naturally.
One particularly interesting finding was that nearly half of Claude's real-world conversations involve subjective content...not just factual Q&A. From over 700,000 analyzed chats, ~44% include interactions where Claude had to express judgments or preferences.