John Riccardi

The marketing and grandiose claims that AI companies are making causes it to feel like anyone with a realistic view will be the one who we look back on as saying the iPhone won't be useful or the internet will never go anywhere. We need to have honest conversations like this. Kudos to @zed.dev

6/6

For some reason, a lot of pockets of the tech world right now make it feel like it's unthinkable to 'speak up' against infallible AI tooling. That's not healthy.

5/?

2. Why do I feel like it's incredibly brave for @zed.dev to post this? Companies SHOULD be willing to talk about their products strengths and areas for improvement.

4/?

- and let's be honest, AI demos are often only the first 10%. Watching it output a ton of code is impressive at first, but rarely continues on that way. That's only addressing the over-hype due to innocent awe, but often the hype is from purposeful marketing and manipulation

3/?

I'm watching the talk with Alberto now, and I'm really glad he's saying something I was already thinking. The last 10% of a project often takes 90% of the time. I think because of this, people get over-hyped on seeing any tool help you with that first 90%

2/?

Why I'm Dialing Back My LLM Usage - Zed Blog From the Zed Blog: Alberto Fortin shares his honest reflection on the reality of using LLMs in production code and why he's taking a more measured approach.

zed.dev/blog/dialing...

A few thoughts about this post.

1. I completely agree with the take that we really need a realistic view of what these tools (that's right, they're TOOLS) can and more importantly, cannot do.

1/?

albertofortin.com/writing/codi...

www.honest-broker.com/p/the-force-...

"The AI business model would collapse overnight if they needed consumer opt-in."

Arcane Grain John Riccardi | Lead Software Engineer specializing in React, TypeScript, and frontend development.

Need a little index card sized character sheet for OSE? Here ya go! 🧙🏼‍♀️ arcanegrain.dev/downloads/

Me: Let's do something productive!

Brain: Yeah... No...

Has anyone been hiring junior engineers lately? Or are there any junior engineers that have recently (last 6-12 months) got hired at their first job? I'd be curious to chat a bit about your experience if you'd be willing to share!

Someone put up a @meshtastic.org node on a weather balloon today and I was able to see nodes all the way in Michigan from Madison!

California Cops Investigate ‘Immigration Protest’ With AI-Camera System New data obtained by 404 Media also shows California cops are illegally sharing Flock automatic license plate reader (ALPR) data with other agencies out of state, who in turn are performing searches f...

California Cops Investigate ‘Immigration Protest’ With AI-Camera System

🔗 www.404media.co/california-c...

I bet someone's in trouble today at The Atlantic 😬

In fact, the context window for an AI chat bot is far larger than the context I can load into my brain if a coworker sends me a message asking for help/suggestions!

Similar to TDD, I'm finding more and more that being good at using AI means having skills that already benefit you without the AI

3/?

I'd argue that while this is an extremely important skill, it isn't new. Anyone who's worked on a distributed team has had to break down their problem into a succinct-enough message that a coworker will read it, while still getting the necessary context for the problem you're trying to solve.

2/?

Coding agents have crossed a chasm Coding agents have crossed a chasm Somewhere in the last few months, something fundamental shifted for me with autonomous AI coding agents. They’ve gone from a “hey this is pretty neat” curiosity to s...

From
blog.singleton.io/posts/2025-0...

> For complex problems, the bottleneck isn’t the AI’s capability to spot issues - it’s our ability to frame the problem in a way that enables their reasoning. This feels like a fundamentally new skill we’re all learning.

1/?

I'm finding that giving code reviews is taking me much longer, since I need to guarantee that the code we're allowing to get merged is not just AI generated stuff that _looks_ correct. I'm doing a much more intense code review now that CoPilot is turned on at my current client's org.

2/?

One thing I'm finding with devs who use AI a lot is they always say "You should always check the generated code before committing", but ... do you? We _should_ be writing tests, but do you? Double-checking generated code is one of the first invisible best-practices.

1/?

How do y'all keep your frontend types in sync with the data coming from API's (both internal / external).

I'd love to just consume types from an API's OpenApi spec, but I know that's not always feasible (especially at a large enterprise like the place I'm currently at).

Just got my first @meshtastic.org nodes and there are so many to connect to here!!

Wow, this is AMAZING.
Wow... this really doesn't work.

- Me every time I use A.I. to generate code.

VSCode Snippets VSCode Snippets. GitHub Gist: instantly share code, notes, and snippets.

Idk why it took me so long but I just set up some simple React and Storybook snippets. They're semi-specific to my current client's code style, but here ya go!

gist.github.com/riccjohn/8d8...

I think one part of the pro/anti LLM thing which needs to occur is this: pro-LLM arguments need to clearly take sides about the problems in the tools they're using. It gets elided too much in the service of "this is actually useful in these ways".

so tired of seeing elon musk’s clammy face on here, if ai was so good why can’t i mute faces yet

Looking back at code from 6 months ago and being embarrassed often means you've grown a lot since then. However, I LOVE finding my own code from months ago and actually being impressed too! I'm glad I took pride in my code at the time and wrote something that is maintainable and a good example!

Glue work is overhead. Much like operational labor, it does not add or create value, it ensures that value creation efforts land and can have an impact on quality.

Good glue work is therefore relentlessly efficient. Do as much as you have to, no more. Resist its tendency to sprawl and multiply.

TDD: The Missing Protocol for Effective AI Assisted Software Development | 8th Light Test Driven Development (TDD) provides a powerful framework for improving AI-assisted coding. Learn how writing tests first gives AI the clarity it needs to generate robust, reliable code—and boosts c...

Decided to expand on some of my posts here and turn it into a blog post!

8thlight.com/insights/tdd...

Exactly!!! I have an upcoming blog post that sort of touches on this.