Mario Vellandi

Anyone want to playtest a game about dev culture? winky.vellandi.net Feedback appreciated 😀

Updated the 1st ending to better lean on collaborative software dev while slightly hinting at semantic ui for humans and agents. Multiple people join a new 'committee' the antagonist started

Anyone want to playtest a game about dev culture? winky.vellandi.net Feedback appreciated 😀

My editor Eric updated the story today, and am now looking to refine the first 2 endings. Design systems are adapting to serve humans & agents. The protagonist needs insight and a glimpse into how his role will need to change.

This week I've been learning where agent skills/plugins are installed, optimizing them, and reverse-engineering an image annotation skill from screenshots + descriptions with nano banana pro (very cool!). Now doing various remotion + agent tuts, optimizing github.com/remotion-dev...

I considered adding beads, but that seems more appropriate at a project-level if doing long term iterative development

This week I've been learning where agent skills/plugins are installed, optimizing them, and reverse-engineering an image annotation skill from screenshots + descriptions with nano banana pro (very cool!). Now doing various remotion + agent tuts, optimizing github.com/remotion-dev...

Asking the agent to analyze each skill, plugin, mcp has been enlightening to understand how much is being plugged into the context window and how often (system vs. each turn).

Am enjoying this interview with Cursor's design lead, demoing an OS research project using agents and the internal browser, along with software and interaction design. www.youtube.com/watch?v=bdh8...

Makes me wish Claude Code plugin had a simple "ask" mode as well, as well as stopping after a plan is accepted instead of jumping right into implementation. The editor layout and usage made me realize it's an adaptive interface for different users including designers.

Gotta say, SponsorBlock is a neat example of crowdsourcing. It's a browser plugin for YT where people can easily submit time segments of a video to a central DB that's then synced to all other users, allowing the playhead to auto skip noted segments. sponsor.ajay.app Obv not fab for producers

What's interesting is that it actively encourages categorizing a segment into different types, so that it's not necessarily just about ads, but a variety of other segment types. That way, viewers can whitelist/filter segments. Submitters see an attribute checklist to review before sent.

following this anthropic article on iterative agent dev www.anthropic.com/engineering/..., I'm trying to build a a meta-scaffold for greenfield projects, system isolated using local containers and mounts, used for pre-planning and subsequent dev

Following @codestirring.bsky.social thoughtful post on dev with AI www.chrisgregori.dev/opinion/code... I thought: "instead of building software components and sub-systems from scratch, how might we better take advantage of existing and new OSS software to build upon? Especially within an

recent content, for another agent to query against. The ex-community manager slash librarian in me, marvels how the entire process could be so much more efficient and humane if tech communities centrally shared common resources like events, jobs, profiles, usage info that could queried or

Following @codestirring.bsky.social thoughtful post on dev with AI www.chrisgregori.dev/opinion/code... I thought: "instead of building software components and sub-systems from scratch, how might we better take advantage of existing and new OSS software to build upon? Especially within an

A part of me thinks it just comes down to great metadata, indexing, and search capabilities provided by github, npm, hex.pm, elixir-observer (multi source), and agent.md rules to use these services. This at least helps with all the projects that may never be considered if they didn't get

Following @codestirring.bsky.social thoughtful post on dev with AI www.chrisgregori.dev/opinion/code... I thought: "instead of building software components and sub-systems from scratch, how might we better take advantage of existing and new OSS software to build upon? Especially within an

existing language/vm ecosystem"? That's a discovery and trial concern, which aside from searching directories, word-of-mouth, and community media & peer dialogue, now includes asking an agent "find me options for X" or it doing it auto sometimes with a wary web query "X best 2026". While discovery,

I'm creating/generating a documentation site on TanStack Router so I can better understand its mechanics for myself (bc the official docs are good but the design is busy with little contrast). In the generation process, it's funny and annoying how verbose and repetitive the content can get.