I’ve been trying to like them for the last year since migrating to wayland, but I really don’t. I’m about to go back to herbstluftwm.

I suppose that’s a solution, but you’re also doubling your responsibilities, and the price of the groceries still stays too damn high.

My only complaint were the ones I experienced in Japanese konbini, which has you scan your purchases, but you had to wait until after payment to bag them. Once this really patient, wonderful local couple showed me what I was doing wrong, it was okay. But still an unfortunate hiccup in their process.

That’s sometimes part of the awkwardness. My additional gripes are completely pedantic, but hear me out:

None of the grocery chains in my area offer bagging services, so I have to wait for the attendant to scan my items, bag them myself, and then shuffle back to the register to pay. None of this is a problem, except that the human attendant rarely ever scans the items in the order that makes sense to bag them. So there’s the weird dance of waiting, pushing items aside, etc.

If I’m doing self-check, I can scan and bag using my own patented autistic process ™ that is quick, makes sense, and doesn’t crush things, and I can do it in one shot.

Weird brag to admit you can’t figure out the self-checkout and Fisher Price level UI that audibly shouts at you what to do next.

As someone who finds interacting with register attendants painful, and doesn’t steal, this is unfortunate news.

Their spicy chicken is decent by fast food standards.

If you’re gonna talk shit about my skills you could at least @ me. I’ve worked decades in the industry, with a lot of that time knee-deep in dev tooling and devops. It’s a cake walk keeping an LLM on guard rails producing well-tested code if you’re a seasoned and sane architect, compared to say, humans. Sorry you can’t seem to wrap your head around it.

I have hand-rolled agents/skills that run through a standardized process:

1. investigate
2. plan
3. execute
4. self-review
5. open PR

I leverage bash and python scripts whenever possible within harnesses, to try to avoid Claude injecting flair where steps and processes should be as repeatable as possible. I also write a lot of supplementary skills for it to use, with least-privilege scoping (important!), to give it access to do things like read work request tickets, run staging smoke tests, etc.

You can get to a point where you can basically say something like, “Investigate and plan SWE-1773” and it’ll do all the shit, spit out a human-readable markdown plan complete with mermaid diagrams. You then okay to execute it, and at the end do an earnest PR review, and some manual tests.

Each brrr is gonna depend on how thorough you need to be, feature sizes, and if it shits the bed during planning. But honestly, the likelihood of your LLM shitting the bed is heavily dependent on how good of a harness engineer you want to become. If you shape and mold the environment and workflow like someone who say, works in developer tooling, then output quality can become surprisingly reliable.

Professional coder turned vibe coder here. My approach has been to lean in really heavy on creating deterministic harnesses and guardrails, that can also be templated for bootstrapping new slop projects quickly and easily.

Then I end up just getting paid to press button make feature go brrrr, and on my other monitor I work on personal projects that I actually care about, while simultaneously being paid because I press button make feature go brrrr.

Spending the time curating my harnesses and workflows, including pretty rigorous automated testing of what Claude shits out, is what keeps it all together. You can’t just set it and forget it. Something about that is fun to me, though, because I am masochistic devops man.