Prompt Engineering Context Engineering Loop Engineering Graph Engeniering
alex
@alexvue.bsky.social
simple dev currently learning react native you can check out my blog https://alexop.dev/
Coding harnesses models are still really bad at understanding when they should use a subagent for a task or not, which is annoying.
What other way to start the day but joy and cringing at watching myself speak. @patak.cat and my talk is up now!! 🧵 www.youtube.com/watch?v=FqdJ...
Matias Capeletto & Willow (GHOST) - npmx: a fast, community-built browser for the npm registry
YouTube video by Local-First Conf
youtube.com
In a few years, most coding work will not happen on your machine but in the cloud.
Boris Cherny in a Interview gave a good tip that every 6 months you should delete all your skills hooks claude. md and then see how the new model behaves and only add things once it does it wrong for 3 times.
good talk about how you can improve your setup so that in development you are able to have 3x productivity and without slop. www.youtube.com/watch?v=Ib5...
Harness Engineering is not Enough: Why Software Factories Fail — Dex Horthy, HumanLayer
In July 2025 Dex Horthy turned the lights off: an agent software fa...
youtube.com
Do you still need Figma or can you not do everything with claude code or codex now? All those "old" apps where built in a world where Humans did most of the work those ai labs have the advantage to rethink the new way how we work from scratch.
With AI, anyone is a developer, designer, or business analyst. Here’s an example of how I use AI to help me with my workout tracking app: to come up with my own icon design system for exercises.
codex app is always the best agent app and much better then claude code you see here the new feature and how easy it is now to see what a subagent is doing
I think this is a important concept to understand when it comes to ai agents alexop.dev/posts/four-...
The Four Types of Memory for AI Agents (and How Claude Code Implements Each) | alexop.dev
Working, semantic, procedural, episodic. The CoALA framework splits agent memory into four kinds. Here is what each one is, and how Claude Code actually implements them on disk.
alexop.dev
good prompt to improve your tests if you use vitest clone this tmp vitest-dev/vitest.git and use subagents and check which patterns i can use to improve my own testing setup
why is the codex app so much better then the one from entropic I thought engineering is solved? @OpenAI @AnthropicAI
Nuxt 4.5 is out, and it's a big one. 🚀 - ⚡️ Vite 8 - 🦀 Rspack 2, now powered by Rsbuild - 🌊 experimental SSR streaming for faster TTFB - 🩺 a new stable error code system - 🎨 useLayout, named views, and more ... plus a lot of groundwork for Nuxt 5! 💚💚💚
Nuxt 4.5 · Nuxt Blog
Nuxt 4.5 is our biggest release in a while. Vite 8, Rspack 2 powered by Rsbuild, experimental SSR streaming, a stable error code system, a new useLayout composable, named views, and a lot of groundwor...
nuxt.com
the codex app is so beautiful . One example agent should generate a prototype about a feature idea I have my own workout tracking app I can see both his thinking but also how he interacts and creates the prototype view
codex app ux ui wise files miles ahead . Just nice how they display subagents now
I dont use the code editor anymore github app or claude code app is all you need which is crazy
With github actions and claude code you can automate everything - pr reviews - qa testing - fixing tests
Every team needs to change the workflow. The developer should not use an agent to implement a new feature; instead, the developer should create the software factory that your designers and business analysts are empowered to use to add new features without the need for any developer.
All devs that are good even without AI are also good with AI. Most of the time, only devs who were average or not really good before AI struggle now with AI. Good devs are just able to understand something complex fast.
We’ve also got a starter pack of speakers & sponsors to get you started with follows.
Had a nice time at @WeAreDevs as my first time as a speaker there talking about local first vue dexie offline apps sync engines. I also watched many interesting talks but more important meet many new people and had good discussions about all tech related topics.
I agree, if we always have to read the code, the productivity gains will never be so high.
Westworld was cooking when they had an entire plot arc about how giving your agents long term memory will inevitably drive them insane
best way to built codebases now with agents is to use functional programming style and monads