Dmitry Kandalov

@dkandalov
194 Followers
127 Following
575 Posts
I have witnessed programming for DOS and spent the last 20 years in the Java lands, most recently working with server-side Kotlin.
Websitehttps://dmitrykandalov.com
GitHubhttps://github.com/dkandalov
Dmitry KandalovMay 21
As always, a great talk by Ross Tate at #KotlinConf26 about local lifetimes in Kotlin. You can watch it on the livestream https://www.youtube.com/live/MmwBJbzWbV0?t=18159s (while it's still there).
KotlinConf ’26 Day 1 Livestream: Keynote Included

YouTube
Dmitry KandalovMay 20
New code smell: "AI" tool is consistently good at predicting/writing the next piece of code.
Dmitry KandalovMay 19
Marius Gundersen - mdg 🌻May 19

I am moving to a usage-based billing

Dear employer, I will no longer charger a flat monthly salary, instead I will move to token based billing. Included will be every token I receive as input, including but not limited to emails, slack messages, spoken words and mandatory multi-modal meetings that could have been emails. Included will also be output tokens in the form of code, documentation and small talk. Tokens will also be consumed by my internal monolog as I reason about problems.

#ai

Dmitry KandalovMay 15
Here is a gist with a plugin for OpenCode to commit modified files after each edit tool execution https://gist.github.com/dkandalov/84d1720684e15fdc9508c25915ef2cb0. Because it's easier to review (or undo) changes in Git history.
Plugin for OpenCode to commit modified files after executing edit tools (see https://opencode.ai/docs/plugins)

Plugin for OpenCode to commit modified files after executing edit tools (see https://opencode.ai/docs/plugins) - auto-commit-hook.ts

Gist
Dmitry KandalovMay 8
Is that a thing to have a "script" that self-bootstraps via LLMs? For example, start with a minimal script that calls LLM to "write" itself, always leaving some gaps to fall back to LLM.
Dmitry KandalovApr 21
How come code coverage tools for LLM agents is not a thing? I understand LLM providers might want to just burn more tokens, but everyone else should be interested right? 🤔 (Even if they never run code coverage manually.)
Dmitry KandalovApr 16
Given that LLMs running in a loop (aka agents) are not too bad at searching/analysing code, why not let them access the source code of your project dependencies by default? More often than not, the code is already downloaded by the build tooling.
Dmitry KandalovApr 8
I find it ironic that the same companies that use glorified O(n) complexity in their interviews can be so into the "AI" will replace people theme without specifying the complexity of the problems they want to solve 🧐
Dmitry KandalovApr 5
Another thing that surprises me about LLM agents is the lack of version control integration. Committing on a successful edit would be one of the first features I'd go for (with an option to squash commits later). And obviously include chat history in some form.
Dmitry KandalovApr 4
Considering the "AI" agents era we're in, I'm surprised how compile-and-run-tests in the project before/after making a change is not easily pluggable into agents or even encouraged during setup. Maybe I miss something, and begging LLM to do it in a Markdown file is fine.