Artem Dudarev

Making information useful

[Artem Dudarev @dudarev]

Fable 5 is most interesting not as a demo machine, but as a tool builders are already using to ship useful work: https://t.co/yKvQZj7Mju

[Artem Dudarev @dudarev]

So I built a small tool for myself - a focused feed where I follow specific channels, bookmark videos with votes, and can actually find them later.

If that sounds useful, have a look: https://t.co/RjfLzDAZ2T

[Artem Dudarev @dudarev]

AI-assisted software development is moving fast.

I try to keep up with podcasts, interviews, conference talks, and the rest.

But YouTube feed mixes AI coding stuff with everything else I watch.

[Artem Dudarev @dudarev]

You can read the full roundup and related notes over at:

🔗 https://t.co/lM025TpcbJ

[Artem Dudarev @dudarev]

My AI coding YouTube picks this week are about making coding agents truly operational.

A few good ideas:

Pick the right harness boundaries.

Raise your verification frequency.

Treat context and guardrails as first-class artifacts.

🧵👇

[Artem Dudarev @dudarev]

🔗 https://t.co/rvGcThk39b

[Artem Dudarev @dudarev]

One example how I use that extension is to curate my weekly AI coding YouTube picks.

Here’s this week’s page: https://t.co/cRcC8zRZRZ

[Artem Dudarev @dudarev]

I made a tiny Chrome extension: YouTube watch history → Markdown (titles + links).

Perfect input for AI assistants to enrich the list and help me dive deeper. Super useful for me.

https://t.co/9frjO1bLKC

[Artem Dudarev @dudarev]

Zevi believes that titles and responsibilities are going to change significantly and everyone will become a builder.

His key advice: Don't try to be a 10x PM or dev. Be a "10x Learner."

Full episode: https://t.co/bFib4ooUqI

[Artem Dudarev @dudarev]

I just listened @ArnovitzZevi on @lennysan's podcast.

He describes a "Multi-Model Peer Review" workflow. I use a similar parallel-agent approach during code reviews and found his observations on the models really resonate with me.

Some of his model characterizations:

[Artem Dudarev @dudarev]

The core idea is basically getting a second opinion.

Every model has different blind spots. When you make them review each other's work, you surface issues that a single model might miss just to "please" you.