Another evening, another feature for Jarvis. Let's see if we can talk to it over IM.
It looks like people are going for tokenmaxxing, I'm trying request-maxxing.
I'm trying to see how much I can squeeze into a single GitHub Copilot request.
Can I do it with two (1 Planning, 1 Implementation)?
What better way to spend a couple of hours on Easter Monday than testing the new Gemma-4 with Claude Code via LM Studio?
The performance isn't quite there yet, especially on my M2 Pro with 32GB of RAM, but it's still interesting to test the model's capabilities.
#Gemma4 #LMStudio #LocalLLM #GenAI
You can now run Gemma 4 locally on your phone!
The Edge Gallery App by Google allows you to run some models on your phone, and the performance is decent even with some older phones.
This is my Galaxy S22 Ultra running Gemma 4 4B.
I now want to test one of the most recent phones, maybe a Pixel 😏.
I can see that 100%, the crazy thing is that I've learned about it while using Agents to write the code.
Tonight, I'm learning something new by watching Claude Sonnet fix my code after bumping the copilot-sdk/go version.
Using go doc <library> can be faster than checking the web documentation.
Also, it is good to see Claude Sonnet help fix its friend, GitHub Copilot 🫶
Jarvis just got an upgrade.
What do you need to feel comfortable reviewing this PR? 😬
As I said, Jarvis, my personal toy agent, can now create subagents.
How do you prevent subagents from spawning other agents? 🤔
Easy! You don't pass the run_agent tool to them. 👏
Writing code in 2026 in 3 easy steps:
0. Explain your intent to the Agent
1. Let it write the code & documentation
2. Review the documentation
3. Go back to step 0 🔁
Let me chain that for you
Friday night update: Jarvis is leveling up. 🛠️
I’ve been working on this agent (built on my langchain-go experiment) for a few weeks, and today’s milestone is a big one: Task Delegation.
I added a subagent tool so Jarvis can now spin up dedicated workers for specific sub-tasks.
I’m not really an "orchestrate multiple AI agents in parallel" kind of person.
Unless my GitHub Copilot quota resets in two days and I still have 42% left 👀.
Then I suddenly become an absolute system-architecture visionary.
I might agree with you there what I really hate though is when companies communicate this over email, often hiding the change in the body of the email.
This is about YOUR data and AI.
This is the message I just got from GitHub. I don't have much to say aside from "make your choice, I made mine".
At least GitHub has been transparent enough to make the pop-up very visible.
Full blog post here github.blog/news-insight...
I understand that the AI propaganda might be stressful for some, but I love how this TOOL allows me to improve some of the flows and deliver more value without much extra effort.
The Cursor composer-2 vs fast saga continues... 🐛
Yesterday: fast is default while composer-2 is disabled.
Today: Even if you EXPLICITLY disable fast in settings, new chats STILL default to it.
Note: fast is 3x more expensive! 🤑
Creating it has never been so easy.
Another (AI-assisted) late-evening project is out.
A VS Code/Cursor BMAD extension and I would love your feedback on it or even feature requests 👇
marketplace.visualstudio.com/items?itemNa...
There has never been a better time to build a software company than today.
pragmatism is an underrated quality
I've been experimenting with Claude code, but I'm still unsure about the user experience.
The thing I'm sure though is that this seems like a big waste of space.
Anyone else feel the same?
AI experience [easy mode]
Using Docker Compose with an Nginx container to reverse proxy my app, the app's port changes with the PORT variable, but Nginx had it hardcoded.
I could've solved it myself through trial and error, but at 22:04, Copilot did it in a minute!
Please Stop Force-Feeding AI FOMO to Others!
Not too long ago, I wrote a post about the AI Hype Burnout and how it was harmful to the whole industry.
Natural follow up here: lanziani.com/posts/2026/0...
#AI #ArtificialIntelligence #Tech #FutureOfWork
A new claw is born every other day, so I've decided to build my own.
Taking inspiration from Kelsey Hightower, I introduce you Noclaw.
The smaller it can be, PRs are welcome.
github.com/LucaLanziani...
LLM (Large Lanziani Models)
I’ve been heavily advocating for Agentic AI in Nearform; it got to the point where, during one of my company wide Spec Driven Development sessions someone redefined LLM as "Large Lanziani Models." 😂
#TechLeadership #Efficiency #AI #FutureOfWork
"Less is more" holds even truer in the AI-assisted coding era.
Everyone can now produce tens of thousands of lines of code.
The real skill isn't producing more, it's managing the entropy.
Aim to be the one who can
I mean mostly for people that don't code, they are missing out on all the innovations we have access to.
Although even the coding interfaces could do better IMHO, the fact that we see people resorting to multiple terminals with claude code it's already telling.
You either die a hero or you live long enough to be acquired by Bending Spoon
You are so nice to think I have a sleep schedule 😅
