Jean-Baptiste Quéru

We are excited to announce our new open source project App Platform! App Platform is a lightweight application framework for state and memory management suitable for Kotlin Multiplatform projects. We use App Platform to share code between Android, iOS and Linux machines for faster development cycles and more maintainable, platform-independent code. The framework is a culmination of solutions we developed, deployed and battle-tested in our own apps over the past 2.5 years. App Platform offers a powerful and efficient foundation for your Kotlin Multiplatform projects and we believe it's useful outside of Amazon, too. We look forward to your feedback and contributions. https://lnkd.in/gaGdKcqJ

302

11 σχόλια

The software engineering feedback loop has collapsed. We used to identify problems for someone else to solve; now, the fix is often finished by the time the report is written. The migration from informing -> executing is the new standard. #GeminiCLI #AgenticAI

47

5 σχόλια

Today we're sharing Signals: our closed-loop system for recursive self-improvement in software development agents. Droid learns from interactions with hundreds of thousands of users - detecting improvement opportunities and autonomously implementing them in its own codebase. Self-evolving agents have been a research goal for years. Signals is our implementation: it uses a privacy-preserving aggregation technique to extract failure trajectories from our agents, then transforms these into actionable work items in our project management tools. Today, humans review and merge these changes. Droid's success rate is high enough that we will likely remove the need for human review soon. We attribute a meaningful portion of our recent quality gains to this closed-loop approach. Read more below!

132

5 σχόλια

At Cardano Foundation we have started collaborating with Antithesis to run deterministic simulation testing over a Cardano network. We strongly believe this tool could nicely supplement the sophisticated testing infrastructure already in place for the cardano-node, especially as we are approaching the point where there will be more than one implementation of a Cardano node able to validate the chain, diffuse transactions and blocks, and forge new blocks. Here is a short report on how it went so far, what we found, some of the challenges we had to overcome, and what we are planning to work on in the near future. https://lnkd.in/e3WKkxBy If you are building some infrastructure component or system on #Cardano, we would like to hear from you as our goal is to make this tool a community asset.

61

14 σχόλια

🎉 Exciting news, we've just released A2A Java SDK 0.3.0.Alpha1! This release brings: 🚀 Support for gRPC ✨ Significant client-side enhancements ✅ Alignment with the latest, more stable A2A protocol v0.3.0 specification Ready to start building polyglot multi-agent systems? Check out our blog post for all the details on how to get started: https://lnkd.in/eshtkAju #A2A #Agent2Agent #Java #AI #Quarkus #MultiAgentSystems #OpenSource

60

🚀 The first email of my newsletter is live! I’ve just published the kickoff article for a new series diving into the internals of the Go compiler: 👉 The Go Lexer: https://lnkd.in/dKcFadQE This post explores how Go’s lexer works — the first step in transforming source code into something the compiler can understand. In upcoming posts, I’ll be covering the parser, IR, SSA, and binary generation — taking a look under the hood of how Go code becomes an executable. If you enjoy understanding how things really work behind the scenes, you can read more (Go and not Go related) on my blog or subscribe to the newsletter here: 🔗 https://lnkd.in/dRZpv-As

76

8 σχόλια

A new team member at LittleHorse told me that we have not a workflow engine but rather a new and distributed programming paradigm. In a way, he's right: - We have a fluent DSL (code) that gets compiled into a WfSpec Protobuf, which is like "assembly instructions" for littlehorse - WfSpec's in LittleHorse are like programs - A WfSpec has threads, conditionals, interrupts, exception handling, and function calls (TaskRuns in LittleHorse) The only difference is that in LittleHorse, the runtime is *Distributed*. A TaskRun can run anywhere. The WfSpec gives you a very reliable and simple way to orchestrate complex processes over multiple systems.

21

6 σχόλια

Ever wonder why background processes are called "daemons"? It’s not because they’re evil, though anyone who’s fought with systemd at 3AM might disagree. The roots go way deeper than computing. In ancient Greece, a daimon was a guiding spirit, neither good nor evil, just... present. A quiet influence. A force working behind the scenes. Socrates even claimed to have one whispering advice in his ear. Fast forward a couple millennia, and Unix engineers borrow the idea. Plus a little boost from physics: “Maxwell’s demon,” a thought experiment involving an invisible being organizing energy without being observed. Put those together, and you get the perfect metaphor for background processes: silent, tireless, slightly judgmental. So they called them "daemons", invisible janitors of your system, running quietly in the background, moving files, scheduling tasks, and watching you paste that Stack Overflow snippet... again. You don’t notice them until something breaks. And when it does, the daemon doesn’t scream. It logs. Silently. Mockingly. So next time your cron job fails without warning, don't get mad. Get spiritual. You’ve merely angered the ancient spirits of Unix. #WhyWeSayThat #DevHumor #TechLore #UnixHistory #SystemDesign #Daemon

479

32 σχόλια

"Hardwood--A minimal dependency #Parquet parser" Spent some time over the holidays building a Parquet parser in #Java, avoiding any dependencies besides compression algorithms. Far from complete, but the basics are there, e.g. it can parse the NY taxi ride data set. The goal is two-fold: a) I actually want such a library, i.e. Parquet support without pulling in Hadoop dependencies. b) I want to explore how far I can get building this mostly via AI. I think it's a perfect case for LLMs: a concise and well-defined problem, authoritative spec, lots of training material out there. Note the idea is *not* to vibe code this, but rather to use AI as a tool for building something which makes sense, is maintainable, extensible, and performant. It's still early, but any feedback will be welcome. I'm planning to keep iterating on this (e.g. right now there's no statistics or predicate push-down support yet) and publish a release to Maven Central eventually, if there's interest. Happy 2026! 👉 https://lnkd.in/evusse_H

147

17 σχόλια

New blog post, "Symmetric ACP": https://lnkd.in/dDU7FDbq This post describes **SymmACP** -- a proposed extension to Zed Industries' Agent Client Protocol that lets you build AI tools like Unix pipes or browser extensions. Want a better TUI? Found some cool slash commands on GitHub? Prefer a different backend? With SymmACP, you can mix and match these pieces and have them all work together without knowing about each other.

18

On one of the medium complexity projects, we successfully achieved 70% correctness in migrating from C# to C++ using Product Migrator from Product X. In other words, GenAI agents autonomously handle 70% of the work first, and humans complete the remaining 30% second. If done right, that would slice the cost of the effort to only 30%. There is a lot of talk about agentic development. But when it comes to the actual implementation and delivering results, reality is a bit different. It is like having a good truck but still walking 3 hrs one way to buy good groceries instead of a 20-minute drive because you don't know how to drive it right. The ability to use the technology with enough confidence and understanding so you can commit to results and expected efficiencies - this is what distinguishes product marketing from software engineering. In an agentic age, even more so.

58

5 σχόλια

I'm seeing too many job openings/requests for people who have experience in implementing Agentic Systems/ GenAI. Dear Companies, It's great your hiring for the implementations of such systems, but : - Are you ready for the uncertainty? - Are you ready for evals? - Do you know that the technology is still in it's infancy? If you've already considered these, great. If not, please do before you pour money into it. AI isn't magic. I'm sorry if some people have sold you the idea that it is. AI is a tool like any other. It works well some places but needs more work is some.

70

5 σχόλια