Me trying to keep up with AI news while getting actual work done

I was missing key metadata info on Claude Code, and I found this little gem last week: https://cship.dev/ If you're already a Starship user like me, adding this extension is a no-brainer. It shows you the amount spent, context window usage, and the selected model!

"Can we add brand style and colors?" Old BI vendor: "I'm afraid that's not possible." Me: "Pick one."

You just burned $1k in tokens to rebuild the SaaS you refused to pay $29/month for. Great for the ego. Now maintain it for 3 years to break even. Yup, maintenance is a luxury these days.

Most CFPs have a ~6 month gap between submission and the actual talk. Given how much changed in the last 6 months... I suspect a lot of speakers will be quietly rewriting their slides the week before. So if the title doesn't quite match the content... keep an open mind. That's probably a good sign

GitHub has all the pieces for an agentic IDE: Copilot agent, Actions, Codespaces. But it's async—fire and forget. Cursor nails real-time pair programming. You and the agent coding together. The holy grail? Hybrid mode. GitHub could build it, unconstrained by VSCode legacy.

OK so I made a mini movie on how to survive in data engineering. No bias. Just ducks. https://youtu.be/f0DDhKaZDDc plz netflix contact for collab

I'm not sure you are all ready for this... See you tomorrow 🫡

2026 paradox: We question whether LLMs are intelligent enough. Instead of deciding whether we’ll stay intelligent while using them.

Two years ago I’d say niche down as a software engineer. Now I’m building far outside my niche (data engineering). It's not “Should I learn React?” but “Do I understand the problem well enough to direct AI?” Execution is commoditized; domain expertise still matters. Generalists win.

RIP Mehdio (2019) Time for a fresh new profile picture. Same person. Bolder. Less hair. Hopefully wiser. In 2026, I want to keep teaching curious students—both in person and online—and help the next generation learn not just data, but software engineering in this (post-?) AI wave.

Quick software engineers reality check: 2024: AI writes code quickly — quality is meh 2025: AI writes solid code and boosts productivity 2026: Not using AI effectively becomes a liability 2026 is the year coding with AI is no longer a nice-to-have, but a must-have.

A few observations from shipping with AI these past weeks : 1) TS quietly wins for AI apps (types catch hallucinations at lint time). 2) Juniors risk building human debt by skipping fundamentals. 3) AI + money works only with verification (SQL saved me). Constraints > vibes.

For all the parents out there, I’m launching something new: the Portable family newsletter After living in 4 countries over the past years with young kids, I’m sharing stories on how we did it. Just honestly covering what worked, and what didn't work.

This week's issue: thinking on how to not get more stupid with LLMs eating our brains, spinning up the new hype file format Vortex (it's gaining a lot of adoption), and other tech muse from the terminal of my mind. blog.mehdio.com/p/ctlrr-02-h...

Starting a weekly habit/newsletter. Each issue includes: - something I’ve been thinking about - something I’ve built, tried, or found useful - something I read or watched, with my take blog.mehdio.com/p/ctlrr-01-t...

We cover everything from interview prep and data quality struggles to the eternal "just a small schema change". Each question gets four different perspectives from @seattledataguy.bsky.social, @hachej.bsky.social, @mehdio.com, and me, with practical wisdom that goes beyond the hype.

2 things that my non tech friends asked me recently : WTH is - us-east-1 - astronomer I wonder what's next.

If you’ve ever found yourself in copy-paste hell during webdev — debugging logs, pasting API errors into an AI chat, or sending screenshots just to say “the button isn’t there" I wrote something for you. blog.mehdio.com/p/an-actuall...

I don't kill my side projects. I just quietly stop maintaining them while convincing myself I'll definitely get back to that one.

Using Docker Desktop or Podman on macOS? WWDC surprise: Apple's own containerization framework. The end of 3rd party containers on Mac? youtu.be/s06VHrb_Fss

There’s a big shift coming in dev tools: their main users will soon be AI agents, not humans. Neon already sees 4x more DBs spun up by AIs. The question now: How do you make your tool AI-consumable? MCP is one answer—but new platforms & integrations will matter too.

Having more fun building a metadata registry for my personal website with all my content. 3 goal for this : 1. Bring public visibility to all my content (written / video) in one place 2. Make it easier to search through all of it (WIP)

If you're wondering what all the fuss is about table formats—and this new one called DuckLake—this video is for you. If you don’t choose table formats but benchmark them, it’s for you too. If you don’t like table formats at all? Well, then share this video with your enemies.

The best part of being active online is turning connections into real-life moments. Just hosted a sold-out DATA+AI meetup in Zurich at Google!

I just had an epiphany, what Iceberg, Delta Lake, and Duck Lake have in common? Their logos are all blue. Like a lake... You’re welcome.

What if medium-sized software companies are slowly dying? I’ve seen more and more solo-preneurs and small teams shipping fast, staying lean — and still landing enterprise customers. So what if scaling headcount… wasn’t actually necessary? blog.mehdio.com/p/the-slow-d...

Remember: AI won't replace software engineers—but it will replace those who don't use AI. I wrote about my latest AI productivity hacks using Cursor. blog.mehdio.com/p/making-cur...

Today we're launching DuckLake, an integrated data lake and catalog format powered by SQL. DuckLake unlocks next-generation data warehousing where compute is local, consistency central, and storage scales till infinity. ⁠ducklake is an open standard and we implemented it in the "ducklake" extension.

Since I teach #dataengineering, I figured my personal website should reflect that—it’s all about pipelines. I’ve automated most of the publishing workflow so the site acts as a central metadata hub, no matter where I post content online.

There is a lot of noise about MCP - but are they actually useful for data engineering ? @mehdio.com dives into a pragmatic example of how MCP, and specifically @duckdb MCP, can speed up data pipeline development Full video 📺 www.youtube.com/watch?v=yG1m...

Quack... Quack... and code! @mehdio.com and @medriscoll.com from @rilldata.com are diving into how GenAI is reshaping BI-as-code — from idea to implementation. This one’s for data folks who want to see beyond the hype. Register : lu.ma/w4ncmttn

🇨🇭 DATA + AI ZURICH — LET’S GO! We’re hosting a no-pitch, insight-packed meetup at Google Zurich with the Applied GenAI community 🧠✨ ⚡ 10 lightning talks 🧠 Real-world data + AI 📍 My first time in Zurich! 📅 Join us: meetup.com/applied-gena...

Real productivity starts before using AI blog.mehdio.com/p/macos-esse...

Plane’s boarding. Me: One quick "ollama run" before takeoff… ✈️💻

No LLM will save you from this: Regularly updating packages and runtimes is an underrated part of software engineering. Ignore it, and you’re stacking tech debt. Wait too long, and breaking changes + dropped runtime support will catch you. Make maintenance a habit 🧹

The team delivered an epic feature, so I had to deliver an epic video 🤷‍♂️

Data engineering feedback loops too slow? 🐌 MCP connects AI copilots (Cursor, etc.) DIRECTLY to tools like DuckDB! Imagine AI inspecting schemas & testing queries before generating code. 🤯 @mehdio.com shows how w/ DuckDB/dbt 👇 motherduck.com/blog/faster-...

Local LLMs running IN your browser? 🤯 WebGPU is making it happen faster than you think. I dove deep into why this tech is key for client-side AI + built a real-time speech-to-text app. Practical guide + code examples inside! 👇 blog.mehdio.com/p/local-llms...

Good technical diagram is super valuable - especially in an era where AI is generating more code than ever. I made a walkthrough video showcasing a few pragmatic ways to generate solid first-draft tech diagrams using AI. www.youtube.com/watch?v=od9x...

Today is a good day to be on Google Cloud.

Be careful what features you wish for. Before you know it, you’re stuck in an over-engineered data project. Been there, done that—because hey, it’s fun.

My goal has always been simple: to educate. To lower the technical barrier. To make data more inclusive. Even for ducks. 🦆

While there’s all the rage about new OpenAI image generation, I’ve put everything I’ve learned so far about generating technical diagrams with AI into some practical workflows. mehdio.substack.com/p/how-to-use... I promise this blog doesn’t contain any Ghibli-style images.

You’re not a real data engineer until you’ve broken production

TIL that NoSQL doesn’t mean “No SQL”—it actually stands for “Not Only SQL.” SQL never really left. it was just laying low, waiting for its grand comeback.

I hereby declare today the official Data Engineer Day. Say thank you to this unsung hero of your data team Working in the shadows, always ready.

Irony: build a tool to curate weekly data content online, and only to have your own content surfaced 🤦 Latest data subTLDR is live : www.subtldr.com/newsletters/...

If you missed the memo on the new @duckdb.org UI, I just just dropped a walkthrough video on @motherduck.com channel, check it out! 👇 www.youtube.com/watch?v=vCbO...

Me: “Hey AI, can you code this feature?” AI: “Go learn it yourself, you lazy human.” Turns out AI isn’t here to destroy humanity—it’s trying to save it.