I often describe engineering leadership as choosing "the least flawed decision." There are no perfect trade-offs in software. Every choice has a cost. The job is to understand those costs, make them deliberately, and slow system entropy better than competitors.

"Software engineering is programming integrated over time" has never resonated more. Programming is getting close to free. The hard part is building systems that can live, change, and remain reliable over time.

A strong signal that your feature was designed well: it has clear boundaries. So clear that removing it feels like using a scalpel: delete a single directory, not start an excavation.

I'll be in the Bay Area for a few days Week 18. Ping me if you are around and wanna grab a coffee.

AI isn’t changing the job of engineering leadership. It’s amplifying it. Our job has always been to turn raw contributions into reliable systems. Before, that meant juniors & new hires. Now, it also means agents & AI-augmented engineers. What changed is the throughput.

Just because you can doesn’t mean you should.

Divide and conquer still wins. Separation of concerns still wins. Modularity still wins. Why? Because complexity is expensive. For humans (cognitive load, pace, quality) and for AI (reasoning depth, latency, token burn). Fighting complexity still pays off.

The fundamental shift with modern AI: the time it takes to execute is starting to match the speed at which ideas appear.

AI in software engineering is a megaphone: it makes both good and bad developers louder. The difference? One is music. The other is noise.

Compile-time flagging is a game-changer for large codebases. Just be sure to expose these variants only at the highest level of your build. If you don’t, you’ll end up with combinatorial complexity leaking everywhere in your code.

> We fixed a few minor bugs that were causing problems. Thanks for that detailed changelog. Super helpful 🤦‍♂️

Look who joined the meetup 😂

🙈

The @parisandroid.bsky.social is about to start. Let's begin with Compose magic 🤩

Thrilled to meet the Android community tomorrow and dive into video generation, dependency injection, and compiler plugins! Don’t miss out, register now: www.meetup.com/android-pari...

Join us with @ParisAndroidUG at @amoamoamo HQ on Jan 13, 2026! 🚀 On the agenda: generating videos off-screen and off-main thread from Composables, why & how we switched from Hilt to Metro DI, and how to build your own Kotlin compiler plugin.

Big news. On January 13, 2026, amo is serving up something special for the @parisandroid.bsky.social 🚀. Think Metro DI insights and video magic from Composables. Stay tuned after the holidays for all the details!🍹

Hidden perk of Metro DI over Dagger: no more wrestling with naming binding functions. We’ve migrated nearly our entire codebase to @Contributed* annotations cutting thousands of lines of code in the process.

My job isn't to design the perfect solution. It's to build the least flawed one possible.

Websites like apilevels.com are useful, but nothing beats analyzing your own data. At amo, we see a radically different Android versions distribution. It completely shifts our decision process regarding usage of AGSL, AVIF, HEIC, and more.

Tooling helps manage the complexity of large codebases. We built tools to track important metrics: ‘unused-modules’ to find deletable modules via DFS, ‘longest-path’ to calculate the longest dependency path - highlighting potential critical paths and more.

Defying g̶r̶a̶v̶i̶t̶y̶ complexity

The best code is no code

Watched 'The Whale' last night. It made me cry, and I still get goosebumps rewatching the ending scene. Strongly recommended!

I love enums/sealed classes/union types. I really do. They give exhaustivity at compile time ❤️. But they just don’t scale. I always end up making the painful choice of trading compile-time safety for scalability. It hurts, but it’s necessary.

Killing the game

They don’t teach you this enough in CS school: graph data structures and algorithms are everywhere. Maps, social networks, build systems, object graphs, page ranking, you name it. Master them, and you’ll unlock superpowers in problem-solving. 🚀

> During an investigation, your job is not to fix issues, it is to understand root causes. @p-y.wtf says (and does) it all…

Coupe d'Europe de la boulangerie. Pièce artistique de l'équipe française.

A vital rule for scalability: every compilation unit must build in <15s (our threshold). Though build time ≠ LOC, tracking it reveals bloat, enforces discipline, and keeps your project fast as it scales. Speed is non-negotiable.

In a highly modularized codebase, typing full Gradle task names can be tedious. Gradle supports task name abbreviation: just type the shortest unique prefix. If ambiguous, Gradle suggests candidates. ⚠️ Use locally for speed, but keep codebase references explicit for readability!

5 ways motion elevates your product: 1️⃣ Emphasize information hierarchy Guide attention effortlessly. 2️⃣ Add delight Make interactions feel magical. 3️⃣ Provide feedback & status Keep users informed in real time.

Jeudi 16 octobre à 12h, je serai en direct avec Mathias de The Product Crew (TPC pour les intimes) pour parler engineering et produit ! Hâte d’échanger avec vous et de répondre à vos questions. À très vite !

We built a custom animatable, collision-aware, collapsible annotation system on Google Maps. Huge UX improvement but we lost tilt support along the way. We finally cracked the math and extended our projection calculations adding a whole new dimension to the map. Just gorgeous.

In the age of AI agents, a monorepo isn’t just a dev setup. It’s a competitive edge. Everything in one place means agents can explore, analyze, and act across your entire codebase, seamlessly.

🚀 Introducing OpenZL, a new #opensource, format-aware compression framework! It uses a unique “graph model” for compression, enabling modular, high-performance codecs and a universal decoder - perfect for structured data at scale Link: https://loom.ly/z09xoNw#opensource #DataCompression

Software engineering is evolving fast thanks to AI. At amo, we built a team dedicated to leveraging AI across our stack. Our experience so far is pretty clear. AI isn’t (yet) great at building features to our craft standards. 1/2

Great animations should always add value. But if they also bring delight, that’s a win 🏆. Instead of a generic 'Search', explicitly specify what’s searchable and animate each character individually. It subtly guides users, catches the eye, and feels uniquely crafted.

One of the most underrated parts of my job is "chaos management". Turning disorder into order, and slowing down entropy before it speeds up. It’s the invisible work that keeps everything running smoothly.

Just received "Code Health Guardian" today. Can't wait to go through it 😋

We already had incremental compilation in the single-digit seconds range, but I’m still blown away by how much faster it is now that the entire codebase is fully on Metro. 🤯 github.com/ZacSweers/me...

Ever wish you could see what your friends are up to? Even when they’re off-screen? With sticky pins in Bump by amo, you can now spot friends typing in chat or sending super sends, no matter where they are on the map. ✨

🪦 Android Instant Apps 🪦 2016 - 2025 R.I.P.

There are two ways of constructing a software design. One way is to make it so simple that there are obviously no deficiencies and the other way is to make it so complicated that there are no obvious deficiencies. – Tony Hoare

I’ve spent my entire career in startups. Why? • I build things, not just org charts. • I move at 10x speed, no alignment meetings required. • I’m a generalist, not a cog in a machine. • My impact > my title.

Interviewing engineers is like watching kids play with blocks: • Some struggle with Duplos. • Some follow the manual. • Some build castles from thin air.

NADD - Naming Avoidance-Driven Development - for the win! Just refactored an API to avoid the impossible: naming things. Because if we can’t agree on a name, we’ll just make it disappear 😅.

Code consistency is the backbone of scalable tooling. As your codebase grows, inconsistency becomes technical debt that cripples automation, refactoring, and even AI-assisted workflows. Consistency enables tooling, and tooling enables scale. Ignore it at your peril.

This week in frontend engineering: • Wrote an iOS strings file parser • Reviewed a Liang-Barsky implementation • Built a URI matcher using a tree • Debated efficient geospatial collision detection Who said frontend is about tweaking padding and building UI like Lego blocks?