Luca - Dev <-> DevOps
@luca.lanziani.com
Hi, nice to meet you, follow me for Tech content, random jokes, food complains (I'm Italian).
Fulltime dad and husband (24/7), #remote Head of DevOps and Platform Engineering (8/5).
Fulltime dad and husband (24/7), #remote Head of DevOps and Platform Engineering (8/5).
What I could build before was limited by my time and knowledge; what I can build now is limited by my attention span and the amount of tokens I'm willing to burn.
February 8, 2026 at 11:16 AM
What I could build before was limited by my time and knowledge; what I can build now is limited by my attention span and the amount of tokens I'm willing to burn.
🎵 ooh It's just me myself and AI 🎵
January 25, 2026 at 11:16 AM
🎵 ooh It's just me myself and AI 🎵
Hear me out: AI Skills should include tools. 🛠️
Right now, we attach tools to Agents. This kills composability. (Why manually teach both a Frontend and Backend agent how to use the same linter?)
We need to flip the model. A Skill should describe the logic and hold the tool.
Right now, we attach tools to Agents. This kills composability. (Why manually teach both a Frontend and Backend agent how to use the same linter?)
We need to flip the model. A Skill should describe the logic and hold the tool.
January 24, 2026 at 3:25 PM
Hear me out: AI Skills should include tools. 🛠️
Right now, we attach tools to Agents. This kills composability. (Why manually teach both a Frontend and Backend agent how to use the same linter?)
We need to flip the model. A Skill should describe the logic and hold the tool.
Right now, we attach tools to Agents. This kills composability. (Why manually teach both a Frontend and Backend agent how to use the same linter?)
We need to flip the model. A Skill should describe the logic and hold the tool.
Me doing a code review on the legacy system I wrote five years ago.
January 24, 2026 at 8:25 AM
Me doing a code review on the legacy system I wrote five years ago.
Me explaining the inference speed of my local LLM running on a consumer GPU.
January 23, 2026 at 7:55 PM
Me explaining the inference speed of my local LLM running on a consumer GPU.
"Me justifying my Agentic workflow results from local LLMs."
January 23, 2026 at 5:55 PM
"Me justifying my Agentic workflow results from local LLMs."
Prediction: By late 2026, we won't be impressed by AI that writes poetry ✍️
We will be impressed by AI that can fix a bug, test the code, and deploy it while we sleep 😴
Generative = Creating
Agentic = Executing
What task would you pay to never do again? 👇
#TechPredictions
We will be impressed by AI that can fix a bug, test the code, and deploy it while we sleep 😴
Generative = Creating
Agentic = Executing
What task would you pay to never do again? 👇
#TechPredictions
January 22, 2026 at 8:32 PM
Prediction: By late 2026, we won't be impressed by AI that writes poetry ✍️
We will be impressed by AI that can fix a bug, test the code, and deploy it while we sleep 😴
Generative = Creating
Agentic = Executing
What task would you pay to never do again? 👇
#TechPredictions
We will be impressed by AI that can fix a bug, test the code, and deploy it while we sleep 😴
Generative = Creating
Agentic = Executing
What task would you pay to never do again? 👇
#TechPredictions
The barrier between idea and reality has never beer
thinner.
thinner.
January 22, 2026 at 7:32 PM
The barrier between idea and reality has never beer
thinner.
thinner.
Sorry I have to do it…
“AI talking about Software Engineers”
“AI talking about Software Engineers”
January 22, 2026 at 3:56 PM
Sorry I have to do it…
“AI talking about Software Engineers”
“AI talking about Software Engineers”
“Me explaining why my brain isn’t working before my morning coffee!”
January 22, 2026 at 1:42 PM
“Me explaining why my brain isn’t working before my morning coffee!”
"Prompt Engineering" is starting to feel a lot like just... engineering.
It’s less about finding magic words and more about managing state, memory, and flow control.
We’re back to building state machines, just with fuzzier logic.
It’s less about finding magic words and more about managing state, memory, and flow control.
We’re back to building state machines, just with fuzzier logic.
January 21, 2026 at 8:59 PM
"Prompt Engineering" is starting to feel a lot like just... engineering.
It’s less about finding magic words and more about managing state, memory, and flow control.
We’re back to building state machines, just with fuzzier logic.
It’s less about finding magic words and more about managing state, memory, and flow control.
We’re back to building state machines, just with fuzzier logic.
We all want fully autonomous agents, but let's be real: we aren't there yet.
For the next year (or more?), "Human-in-the-Loop" is the only safe way to deploy this stuff. Design for failure. When the agent gets confused (and it will), make sure it hands the wheel to a human instead of crashing.
For the next year (or more?), "Human-in-the-Loop" is the only safe way to deploy this stuff. Design for failure. When the agent gets confused (and it will), make sure it hands the wheel to a human instead of crashing.
January 21, 2026 at 7:59 PM
We all want fully autonomous agents, but let's be real: we aren't there yet.
For the next year (or more?), "Human-in-the-Loop" is the only safe way to deploy this stuff. Design for failure. When the agent gets confused (and it will), make sure it hands the wheel to a human instead of crashing.
For the next year (or more?), "Human-in-the-Loop" is the only safe way to deploy this stuff. Design for failure. When the agent gets confused (and it will), make sure it hands the wheel to a human instead of crashing.
Satya Nadella at the World Economic Forum shared AI insights: AI acts as a "cognitive booster" freeing professionals from tedious tasks.
He sees AI output as new "tokens," essential for economic growth.
Firms must control their data to stay competitive.
#AI #Innovation
He sees AI output as new "tokens," essential for economic growth.
Firms must control their data to stay competitive.
#AI #Innovation
Microsoft CEO Satya Nadella at WEF session
A conversation with Microsoft CEO Satya Nadella at a World Economic Forum session moderated by BlackRock CEO Larry Fink.
www.youtube.com
January 20, 2026 at 9:35 PM
Satya Nadella at the World Economic Forum shared AI insights: AI acts as a "cognitive booster" freeing professionals from tedious tasks.
He sees AI output as new "tokens," essential for economic growth.
Firms must control their data to stay competitive.
#AI #Innovation
He sees AI output as new "tokens," essential for economic growth.
Firms must control their data to stay competitive.
#AI #Innovation
keep your Agent on the target.
January 20, 2026 at 6:18 PM
keep your Agent on the target.
tools can help you inject a bit of determinism into a non deterministic tool.
January 20, 2026 at 5:18 PM
tools can help you inject a bit of determinism into a non deterministic tool.
Moving past AI as just a coding assistant. 🚀
Automating whole workflows with agentic tools opens a completely different set of possibilities.
Have you made the jump to agentic workflows yet? What tools are you using?
#AI #DevTools #AgenticAI
Automating whole workflows with agentic tools opens a completely different set of possibilities.
Have you made the jump to agentic workflows yet? What tools are you using?
#AI #DevTools #AgenticAI
January 19, 2026 at 5:01 PM
Moving past AI as just a coding assistant. 🚀
Automating whole workflows with agentic tools opens a completely different set of possibilities.
Have you made the jump to agentic workflows yet? What tools are you using?
#AI #DevTools #AgenticAI
Automating whole workflows with agentic tools opens a completely different set of possibilities.
Have you made the jump to agentic workflows yet? What tools are you using?
#AI #DevTools #AgenticAI
If it wasn't clear I meant leading the AI by example
With AI, leading by example matters more than ever.
January 18, 2026 at 6:20 AM
If it wasn't clear I meant leading the AI by example
With AI, leading by example matters more than ever.
January 17, 2026 at 2:21 PM
With AI, leading by example matters more than ever.
The first picture is the original, and the second is the result of the following prompt to Nano Banana:
"Can you apply the stickers shown in the picture to the real lego on the right?"
"Can you apply the stickers shown in the picture to the real lego on the right?"
January 17, 2026 at 10:47 AM
The first picture is the original, and the second is the result of the following prompt to Nano Banana:
"Can you apply the stickers shown in the picture to the real lego on the right?"
"Can you apply the stickers shown in the picture to the real lego on the right?"
Halfway through the Agentic AI course on DeepLearning.AI. 🚀
Thank you Andrew Ng for the course 🙏
#KeepLearning
Thank you Andrew Ng for the course 🙏
#KeepLearning
January 17, 2026 at 10:23 AM
Halfway through the Agentic AI course on DeepLearning.AI. 🚀
Thank you Andrew Ng for the course 🙏
#KeepLearning
Thank you Andrew Ng for the course 🙏
#KeepLearning
Last week: Building agents with Google ADK/LangChain and testing community workarounds to access Copilot.
Yesterday: @github drops the official Copilot SDK. 🤯
Talk about impeccable timing. Time to refactor! 🚀
Yesterday: @github drops the official Copilot SDK. 🤯
Talk about impeccable timing. Time to refactor! 🚀
January 15, 2026 at 7:32 PM
Last week: Building agents with Google ADK/LangChain and testing community workarounds to access Copilot.
Yesterday: @github drops the official Copilot SDK. 🤯
Talk about impeccable timing. Time to refactor! 🚀
Yesterday: @github drops the official Copilot SDK. 🤯
Talk about impeccable timing. Time to refactor! 🚀
We are ok with the productivity gap from tool mastery because it is linear.
AI proficiency is driving a non-linear divergence. The output gap between AI-native workflows and traditional ones is becoming too significant to treat as a mere margin of error.
AI proficiency is driving a non-linear divergence. The output gap between AI-native workflows and traditional ones is becoming too significant to treat as a mere margin of error.
January 12, 2026 at 5:03 PM
We are ok with the productivity gap from tool mastery because it is linear.
AI proficiency is driving a non-linear divergence. The output gap between AI-native workflows and traditional ones is becoming too significant to treat as a mere margin of error.
AI proficiency is driving a non-linear divergence. The output gap between AI-native workflows and traditional ones is becoming too significant to treat as a mere margin of error.
TIL: You can define handoffs in your custom Copilot Agents.
This adds a button to the chat that allows you to switch between agents during specific flows. It is particularly useful for BMAD when following various workflows.
This will make the experience a lot smoother for your developers.
This adds a button to the chat that allows you to switch between agents during specific flows. It is particularly useful for BMAD when following various workflows.
This will make the experience a lot smoother for your developers.
January 11, 2026 at 11:16 AM
TIL: You can define handoffs in your custom Copilot Agents.
This adds a button to the chat that allows you to switch between agents during specific flows. It is particularly useful for BMAD when following various workflows.
This will make the experience a lot smoother for your developers.
This adds a button to the chat that allows you to switch between agents during specific flows. It is particularly useful for BMAD when following various workflows.
This will make the experience a lot smoother for your developers.
I asked Gemini to describe the difference between Spec-Driven Development and Vibe Coding, and then I asked it to generate an image.
Do you agree with the result?
Do you agree with the result?
January 10, 2026 at 3:25 PM
I asked Gemini to describe the difference between Spec-Driven Development and Vibe Coding, and then I asked it to generate an image.
Do you agree with the result?
Do you agree with the result?
How are you people prompting Nano Banana? Any examples?
Curious to see how everyone is approaching it.
Curious to see how everyone is approaching it.
January 9, 2026 at 8:29 PM
How are you people prompting Nano Banana? Any examples?
Curious to see how everyone is approaching it.
Curious to see how everyone is approaching it.