Julien Letessier

The project I mentioned in my last post works great, but it definitely took longer than I expected to get everything tuned. So for my next build — a new controller for a spin coater — I tried a different approach. If you haven’t seen one, a spin coater is basically a machine that spins a substrate at controlled speeds so you can lay down a thin, even film. Commercial ones are expensive, so I built my own with a drone motor, 3D‑printed parts, and an ESC. Up until now I was driving it with an RC transmitter, but I needed proper ramp‑up/ramp‑down curves and accurate RPM control. This time I used Google Gemini, and it took me about eight hours to go from idea to a fully working system: firmware, UI, PID tuning, ESC calibration, and support for pro‑level spin profiles. I shared both the prompt and the generated code in this repo. 👉 https://lnkd.in/gWu98ejp But the real story here isn’t the project — it’s the workflow that’s starting to emerge: My LLM‑Driven Engineering Workflow: • I start by having the LLM help define the module’s specs. It’s like a design review partner that never gets tired. • Once the specs look good, I ask the model to write the prompt that will generate the code. • Then I take that prompt to a second LLM and let it critique and improve it. (It usually says “this is already solid, but…” and then gives genuinely useful tweaks.) • I run the refined prompt in my code‑gen model and save that prompt as part of the project. • And yes — the model always misses something. That’s normal. I issue follow‑up prompts and keep them in a “prompt log.” • The magic happens later: when a new LLM version comes out, I can re‑run the same prompts and instantly get a better implementation without rewriting anything. And this is where things get interesting. LLMs are about to change code reuse in a big way I’m realizing that the old idea of “reusing code” is going to shift. Sometimes it’s not worth pulling in a third‑party library anymore. If the model already knows how to implement a PID controller, for example, it can generate a clean, purpose‑built version that fits your architecture better than a generic library. That’s a big deal. Because it means the thing we reuse isn’t the code — it’s the instructions that generate the code. We’re moving from: • reusing libraries → to reusing prompts • sharing code → to sharing the recipe that produces the code • maintaining a codebase → to regenerating it when the models improve The prompt becomes the durable artifact. The code becomes a snapshot — something you can always recreate, refine, or upgrade later. This feels like the beginning of a pretty big shift in how we build software.

6

1 commentaire

If you’ve ever struggled with keeping database changes in sync across dev, staging, and prod - Flyway might be the tool you didn’t know you needed. I've written a 2-part blog series tailored for developers: Part 1 : Why use Flyway? Understand the why behind Flyway, versioned migrations, idempotency, and what it brings to the table for modern dev teams. -> https://lnkd.in/dZtPAK5N Part 2 : Hands-on with MySQL A step-by-step walkthrough: setting up multi-env DBs, running migrations, seeding data, lifecycle hooks, CI/CD, and more! -> https://lnkd.in/dbGbQNc3 Built for developers by a developer🤓

7

2 commentaires

I've been running Claude Code to build my new project — but until recently, I couldn't tell what a feature costs. Full TDD cycle, GitHub issue creation, PR creation, all handled autonomously. One thing I'm missing: how much does each feature actually cost? Claude Code emits OpenTelemetry metrics (tokens, USD cost), which is great. But they're tagged with session UUIDs, not anything meaningful like a feature name or ticket ID. You can add RESOURCES, but that's static information — you can obviously update it every time you start a new feature, but that's not a great solution. So I built a workaround: OTel collector → Prometheus → Grafana, with textfile metrics mapping session IDs to feature IDs (via NodeExporter). It works — until you run concurrent sessions. One more issue: the OTel Collector expires metric series after 5 minutes of inactivity (by default) — so if Claude Code goes idle mid-session, the next conversation starts a fresh accumulation from zero. The workaround is to increase expiration time but this is not ideal solution. Here's a real example from a recent PR: Cost: $1.49 Tokens: 3,755 input / 30,300 output / 3M cache-read / 140K cache-write That's a full feature — tests, implementation, PR. The fix is straightforward: when you set a custom session title in Claude Code, emit it as an OTLP resource attribute. One label. The Grafana query becomes: sum by (feature_id) (claude_code_cost_usage_USD_total * on(session_id) group_left(feature_id) claude_session_feature) Filed a feature request with Anthropic: https://lnkd.in/eW_4phzy It's sitting alongside 5,000+ open issues, so I'm not holding my breath :) If you're using AI coding agents in any serious capacity, per-task cost visibility is a necessity. How are you approaching this? There may be simpler solutions I'm missing.  #ClaudeCode #OpenTelemetry #AIEngineering #DevTools #EngineeringLeadership #Observability

3

What's new at Private Packagist, August update: dependency usage tracking across your packages, automatic GitLab token rotation, and Conductor improvements with custom labels and smarter PR handling. Read the full details on our blog 👇  #php #composer #composerphp #phpc https://lnkd.in/efu-xDrB

10

1 commentaire

Coding is no longer about writing lines — it’s about directing intent. 🎼💻 Claude Code Update 2 reveals how agentic systems reshape developer workflows, safety, and scale. A must-read for serious engineers in 2026. Full story 👇 https://smpl.is/agc69 #SoftwareArchitecture #AgenticAI #Engineering

1

1 commentaire

Absolutely blown away by all the positive feedback on the #Hardwood 1.0.0.Alpha1 release 🥳! Thanks a lot to all the comments, discussions on issues in the tracker, pull requests, stars on the repo! So cool to see that this parser for #ApacheParquet, optimized for minimal dependencies and great performance, seems to itch a scratch for folks. I got invited to present Hardwood on the Parquet community call where we'll explore opportunities for collaboration, and I'll hop on a few podcasts to talk about the project, too. For the 1.0 release, the key items we're planning to implement initial support for predicate push-down, reading files from S3, and a compatibility layer with parquet-java. And, of course, further performance improvements: just today, Rion Williams opened a PR which substantially cuts down allocation rates, thus reducing GC pressure and improving overall latency. Onwards 🚀!

63

1 commentaire

As always the Kroxylicious team are very pleased to announce the release the 0.19.0 release (https://lnkd.in/e5sRC4EU). Kroxylicious will now bite down on Kafka 4.2 RPC calls, snack on idle connections and other quality of life improvements.

19

2 commentaires

This case study on "Clopus-Watcher" by Denislav Gavrilov is a brilliant piece of technical research. It's a fascinating proof-of-concept, but it also inadvertently convinced me that we are still years away from safely handing AI the pager. In one scenario, the agent "successfully" fixed a missing health check file by recreating it inside the pod. The alert cleared. The dashboard turned green. From an engineering standpoint, this is terrifying. The AI fixed the state (the missing file) but ignored the cause (buggy code deleting it). It didn’t fix the engine—it just kept covering the warning light. In production, this creates zombie services: healthy metrics masking broken logic. Even worse, the agent applies these "hotfixes" by directly modifying the running container. This violates the core principle of immutable infrastructure. You end up with production code that exists nowhere in Git—and bypasses your entire CI/CD safety net. In the experiment, the platform fought back—entering a restart loop to enforce consistency. Honestly? That was the lucky outcome. The alternative is Schrödinger’s Deployment: ghost code running in production that matches nothing in your repo, making debugging impossible. But even in this "lucky" scenario, you might wake up happy you weren’t paged, only to discover your cloud bill looks like you ran a crypto miner all night, fighting your own orchestration platform. Disclaimer: I’ve spent the last several years working on backend data transformation pipelines. In my world, even a catastrophic infrastructure failure is often just a "P2 data refresh delay" that can wait until Monday. I haven’t carried a pager in over six years. So take this perspective with a grain of salt. But today, giving AI unsupervised terminal access feels less like a runbook replacement—and more like a chaos monkey with admin privileges.

3

3 commentaires

📖 Great blog post by Samantha Coyle from our engineering team, breaking down the magic behind the Workflow Composer app — a tool that transforms any diagram into fully functional code in seconds, with built-in durable execution and observability. Read here: https://lnkd.in/g4nEzWbJ

24

1 commentaire

AI has brought back a feeling I hadn’t had in years. The tight feedback loop I used to associate with good TDD. Not the dogma. Not “tests catch bugs”. The fast cycle of iterate → validate → adjust that tells you whether you’re actually building the right thing. For a long time, AI was too slow or too noisy to fit into that loop. Recent improvements have changed that. AI now sits inside the workflow: it writes quickly, I validate, I simplify, I decide. I’m still driving — just with much faster feedback. I wrote more about this here 👇 https://lnkd.in/ecmWvKTy #AI #SoftwareEngineering #TDD

21

5 commentaires

To all my founder & non-developer friends - Just do these two things this weekend and come back stronger on the other side next week! 1. Get started with Claude Code. It's dead simple. Sign up with Claude → Install the CLI on your local machine → Done. Just 2 steps. Bonus: They're giving away 1 month of Pro subscription free! So no excuses. :) https://lnkd.in/gvunaqhQ 2. Use Meta Prompting. Ask AI to teach you how to communicate with it better. It's the holy grail of learning to learn! Example: "Teach me how to write better prompts for building a world-class landing page. What should I specify? What details matter? Then help me write the perfect prompt to create an Apple-style product page with all the animations, effects, typography and brand ethos." By following just these two, you'll be surprised how far you can go and the range of things you can achieve. All the backlogs collecting dust. The bandwidth issues you face with the engineering team. The ideas that strike you overnight — you can now bring them to life before filtering them out. Your engineering, product, design, and marketing teams will love you. You'll start co-creating with your team instead of standing by while they execute. You'll feed them sharper inputs and let them do what they do best. A few examples to get started:  → Create a landing page like Apple's world-class website  → Generate a detailed report on your website's performance last week  → Prototype a new feature for your  website  → Analyze your customer support queries for insights  → Clone your writing style for brand consistency→ Extract key information from emails and attachments Basically, you're becoming a 1–5× developer, while your engineering, product, and design teams aim to become 10–20×!

57

7 commentaires

Sick of repeating yourself to your coding agent? At ApprovIQ I was sick of repeating our “house style”. So I automated it! When the agent is about to present a plan to a human, our harness stops it and gives it extra advice, things like: - Write end-to-end tests first. When you’re done they will pass - Does a simpler fix exist? With less duplication? - Prefer removing code over adding code - Reuse existing code where it makes sense. That causes it to loop back on itself before presenting the plan, which means now I only see plans that are super refined and high quality. Here’s the docs: https://lnkd.in/gY_Sd5XQ Stop repeating yourself. Create a hook that loops it back on itself and says what you say every time. Think of it like a linter except for concepts, not code. For advice or ideas I’m always happy to chat!

2

One of the coolest features of C# 14 is extension members (releasing with .NET 10 in November). You can finally add extensions to static classes and static extensions in general. This is fantastic for logical composition and discovery. https://lnkd.in/gjnNBMBw One cool use case for this is more complete polyfill when you're multi-targeting for .NET and .NET Framework (yep, this works on both!). I've been fiddling with this in one of my libraries. One example is adding throw extensions so you can do things like `ArgumentNullException.ThrowIfNull` when multi-targeting. https://lnkd.in/gQN-2uGp I'm hoping we can eventually provide this sort of assembly directly from the .NET team in the future. I'm poking on it internally because I believe it has a lot of value to long-term .NET users that still need to build for .NET Framework. If this would be valuable to you, please let me know. :)

10

3 commentaires

It's sometimes not exactly obvious why certain things are done the way they are. One of those not-so-obvious settings is PHP's memory_limit. Matthias Van Woensel explains in greater details why you should care, without having to be a devops engineer. Also, you can clearly see the lived experience in his words. 😅

2

🛑 Stop flying blind with your CD pipeline! Here's the reality: Most DevOps teams are stuck in reactive mode. Something breaks → scramble to fix → repeat. But what if you had a dedicated expert who: ✅ Thinks 3 steps ahead of your deployment challenges ✅ Draws from real-world patterns across hundreds of organizations ✅ Helps you avoid the pitfalls that tank CD initiatives That's exactly what Octopus Deploy's Technical Account Managers deliver. This isn't just another support tier. It's proactive guidance from CD experts who've seen it all - from startups scaling fast to enterprises managing compliance across thousands of deployments. Your TAM becomes part of your team, helping you: ✔️ Optimize your Octopus, Codefresh, and Argo CD setups ✔️ Navigate complex architectural decisions ✔️ Scale deployments without breaking things ✔️ Stay focused on delivering value, not fighting tools The difference? Support fixes what's broken. A TAM helps you avoid breaking things in the first place. Ready to level up your CD game? 👀 Learn more here: octopus.com/tam #DevOps #ContinuousDelivery #OctopusDeploy #CICD #GitOps

16

🚧 If scaling your dev team feels like a struggle — it’s time to rethink the model. 💡 This is the smarter way forward. #InnoverTEK #SmartScaling #UKTech #TheFutureOfDevelopment #TechHiring

4