Lightnews — Scholar-powered news

Victor Savkin @victorsavkin.bsky.social · 19h

Sometimes Nx's Self-Healing CI just fixes your failed builds. Other times it gives you a solid starting point, giving you the right context or helping you to seed a convo with an AI agent.

We’ve been iterating on it. Here’s a quick video of where we’re at.

@nx.dev

Nx @nx.dev · 1d

Sometimes our AI fixes nail it...well almost.
Now you can apply these fixes locally in your editor, make that final tiny tweak and then commit it yourself.

Curious? Check out @juri.dev's full walkthrough 👇
youtu.be/37q9O-PYPlY

1

Victor Savkin @victorsavkin.bsky.social · 6d

AI agent configs are the wild west right now.
So we added a command that sets it up for you.

No need to upgrade Nx. Just run:

npx nx@latest configure-ai-agent

@nx.dev

Nx @nx.dev · 6d

We just introduced a new "configure-ai-agents" command in Nx 21.6.2 to ensure Nx operates optimally with AI agents.

👉 Configures the Nx MCP
👉 Sets up (and updates) rules for AI agents and IDE assistants

Give it a try!

Reposted by Victor Savkin

Nx @nx.dev · 8d

State of JS 2025 is live and also has a monorepo tools section. Make your voice count!

survey.devographics.com/en-US/surve...

2 8

Victor Savkin @victorsavkin.bsky.social · 13d

10/10 Summary:

AI agents are very useful. But creation itself is irreplaceable.

If you always delegate creation, you lose mechanical sympathy, and eventually effectiveness.

Question: how to preserve that effectiveness while still getting the most from agents.

2 2 8

Victor Savkin @victorsavkin.bsky.social · 13d

9/10 The longer-term risk is erosion of expertise.

If engs mostly review and rarely create, they become managers.

And managers, no matter how capable, eventually lose the ability to intuitively understand the system. They lose mechanical sympathy.

1 2 6

Victor Savkin @victorsavkin.bsky.social · 13d

8/10 Question: When should AI drive, and when should it only review?

The boundary is fuzzy and domain-specific.

I think there’s no downside to having AI review your work, no matter how complex the domain.

1 1

Victor Savkin @victorsavkin.bsky.social · 13d

7/10 It's because I wasn't able to engage with it deeply when producing it and I didn't have a good understanding of different options and edge cases.

Several times I just rewrote things from scratch and produced both a better solution and a clearer mental model.

1 1

Victor Savkin @victorsavkin.bsky.social · 13d

6/10 In complex domains, (1) isn't an option.

(2) can be an option, but problems surface. And they don't surface right away. They appear months later.

The issue isn't that the written code is broken, but it isn't exactly right. The solution doesn't fit well.

1 1

Victor Savkin @victorsavkin.bsky.social · 13d

5/10 I use AI agents in three ways:
- The agent works independently (vibe).
- The agent drives, I navigate (short leash).
- I drive, the agent reviews.

1 3

Victor Savkin @victorsavkin.bsky.social · 13d

4/10 Observation 2
Sometimes that deep understanding isn’t essential (e.g., demos, scripts, CRUD).

But in other contexts (distributed systems, performance-critical work), it’s vital. The tradeoffs depend on the domain, and your approach to AI has to adapt.

1 1

Victor Savkin @victorsavkin.bsky.social · 13d

3/10 Observation 1:
The act of writing produces a much deeper understanding than reviewing, no matter how careful the review. This applies to both human-written and AI-written code.

That’s why we ask authors (not reviewers) to debug. Writing forces you to internalize the system.

1 4

Victor Savkin @victorsavkin.bsky.social · 13d

2/10 Many of us have observed the same challenge:

An incomplete understanding of AI-written code that is carefully reviewed can become problematic over time. On several occasions we had to rewrite it from scratch.

1 1

Victor Savkin @victorsavkin.bsky.social · 13d

1/10 A thread on using AI agents well, mechanical sympathy, and why you should not become a manager.

1 1 5

Reposted by Victor Savkin

Nx @nx.dev · 21d

Waiting for @biomejs.dev support in Nx? Why not do it yourself?

⏱️ 20 minutes
✅ Caching
✅ Inferred Tasks

Looking to integrate a different tool? We'll show you how! 🚀

1 6 11

Victor Savkin @victorsavkin.bsky.social · 22d

We completely reworked the @nx.dev graph visualization. It’s night and day compared to what we had before.

🎥 Watch the video below and provide feedback.

Nx @nx.dev · 23d

People love our visual project graph. However, in larger workspaces, the visualization was often unusable.

That's about to change soon. @nartc.bsky.social from our team has dedicated some of his time to the graph viz👇

www.youtube.com/watch?v=xVo...

Updates to the Nx Graph by Chau Tran

https://nx.dev/features/explore-graphLearn more:- https://nx.dev- AI-Powered Self-Healing CI: https://nx.dev/ci/features/self-healing-ci

www.youtube.com

2

Victor Savkin @victorsavkin.bsky.social · 22d

We’re fine-tuning @nx.dev Self-Healing CI to ensure it only proposes changes when it’s confident they’re correct.

For example, Nx has awareness of what’s happening in the repo, so it can recognize when the issue isn’t related to your PR.

Nx @nx.dev · 22d

Integrating AI agents into your workflow isn't just about implementing a solution at any cost; it's about guiding agents to know when to step back and provide clear guidance to the user.

Saves time, reduces costs, and minimizes annoyances.

1 5

Victor Savkin @victorsavkin.bsky.social · Aug 22

A lot to like in this release, especially the long-requested Docker support.

Nx @nx.dev · Aug 22

📣 Nx 21.4 is here! We've got some exciting changes here, so let's get into it. 🧵

1

Victor Savkin @victorsavkin.bsky.social · Aug 21

3/ The original simplicity that made the idea valuable gets buried by clever engineering.

Victor Savkin @victorsavkin.bsky.social · Aug 21

2/ You can see it very clearly in AI tooling.

Cursor, Claude Code--a bit of complexity for a lot of value. Kind of magical.

6 months later: dozens of tools, swarms of agents using different models with separate config files. Nothing composes.

1 1

Victor Savkin @victorsavkin.bsky.social · Aug 21

1/ Programmers are clever, which gets us into trouble.

Every new idea starts simple (TDD, Agile, JS frameworks). Then comes a sweet spot: just enough complexity for big value.

But we race WAY past it into long books, certification programs, and diagrams of diagrams.

1 2

Victor Savkin @victorsavkin.bsky.social · Aug 18

6/ In some ways it's like the web: for most things I’d rather use a web page than install yet another app I’ll barely touch.

It’s also cross-platform. Switching platforms becomes trivial when fewer things are apps and are just systems.

Victor Savkin @victorsavkin.bsky.social · Aug 18

5/ I know some folks will immediately want to vibe an “app” on top of this setup. I disagree. The absence of an app is the entire point.

Moving away from apps towards systems connected to multiple data source is what makes it so enjoyable.

1

Victor Savkin @victorsavkin.bsky.social · Aug 18

4/ Why it’s better:
• more enjoyable to edit text
• zero lock-in (swap editors/agents anytime)
• workflow experimentation is cheap & fast

In spirit, this feels like paper notebooks: you have total freedom in how your setup works.

1

Victor Savkin @victorsavkin.bsky.social · Aug 18

3/ Now my “PKM app” is: a folder of text files (Dropbox), edited via Helix/Neovim, with Claude Code handling retrieval, association, and bulk refactors.

In other words, I replaced an app with a system (folder + editor + conventions + retrieval + sync).

1

Victor Savkin @victorsavkin.bsky.social · Aug 18

2/ Example: PKM. I’ve always felt a tension between:
• ergonomics of editing text
• ergonomics of retrieval & association
Claude Code erases this tension completely.

1