Alan Mosca

Please kill the bloody TOON posts. While I value innovative open-source projects, of which this is at least the latter, LinkedIn must surely have an algorithm with something more exciting! The last thing we need is yet another data representation; we’ve had XML, ASN.1, JSON, and many more, and now TOON. Meanwhile, most of the world is still using CSV. The claim is efficiency with AI. LLMs are trained on human text and existing formats like JSON and CSVs. LLMs do not work well with large structured data so changing the format isn’t going to help much. IT might save a few fractions of a cent over a week or mean you can fit a few more lines of data but CVS is more efficient anyway. Today’s LLM challenge is tool-calling, the core of agentic AI. Those tools are typically defined in XML or JSON (mostly the latter for open source LLMs), the majority are the tool and parameter descriptions, TOON is no help here. If someone wants to re-train or fine-tune an LLM to work with TOON defined tools then great, it might save us a few percent of tokens. Until then, I’ve had my fill of TOON posts, thank you!

90

50 Comments

We process 14M messages/sec with sub-100μs latency. When rewriting our feed handler, Rust seemed like the obvious choice - we use it successfully across our stack... We chose C++ instead. I'm re-sharing our blog post on the technical choices behind that decision.

281

13 Comments

Traceloop today launched its observability tool for LLM-based applications into general availability and also announced a $6.1 million seed funding The post Traceloop Launches an Observability Platform for LLM-Based Applications appeared first on The New Stack .

17

1 Comment

Just released secxbrl, a python package to parse SEC xbrl. Turns out the SEC XBRL endpoint has some limitations: 1. Sometimes takes hours to update after a submission has been filed. 2. The endpoint has a lot of missing data. secxbrl fixes this using a fast and lightweight approach, suitable to running on small instances such as AWS t4g.nanos. GitHub, MIT License. https://lnkd.in/gtSiBbAT

34

3 Comments

Claude just built a C compiler. A real one -- parsing, IR, LLVM-style code generation. Compilers are no joke. Chris Lattner, who spent his life so far building a few compilers, broke down what this tells us about where things are going. A few quotes that resonated with me: "As implementation becomes cheaper, the role of engineers shifts upward. The scarce skills become choosing the right abstractions, defining meaningful problems, and designing systems that humans and AI can evolve together." "AI amplifies both good and bad structure, so we can expect to see poorly managed code scale into incomprehensible nightmares." That second one is the one nobody wants to talk about. We are all excited about shipping faster. But a sloppy architecture that used to slow you down now ships to production in a weekend. The mess scales too. There is also something subtle in the compiler itself. CCC passes tests brilliantly -- but it optimizes toward measurable criteria. As Lattner points out, "there is no test suite for an idea that does not exist." The moment a problem is open-ended, AI has nothing to optimize against. Or as he puts it: "Writing code has never been the goal. Building meaningful software is." Knowing what to build, having the taste to recognize good structure, and the discipline to keep it clean when adding more costs almost nothing -- that is where it gets hard. And that is where software engineering is here to stay. Full article well worth the read: https://lnkd.in/dN4M9gJX

14

4 Comments

Two major events today: 1️⃣ Launch of the yearly SPS in Germany 🏭 2️⃣ Release of Claude Code Opus 4.5 ✨ This morning I am keeping half an eye on my LinkedIn feed showing posts from SPS, while I test the capabilities of Opus 4.5. Such a crazy time to live in, where every x months coding models become y % better. And there are no signs that things are slowing down. But industrial automation is stuck. We can't use the modern coding workflows with the 'legacy' automation IDE's. Yes during SPS multiple vendors are presenting some kind of solutions. But here are the problems: ❌ Coding agents have no direct access to the code, everything runs through API's/MCP/... ❌ No git for tracking the agents. (Or worse: broken git) No granular commits and reviews for what the agents are doing. No diffs. No SDLC. ❌ No or limited BYOM (Bring your own Model). ❌ Separate licensing for the coding agents WITHOUT the model. ❌ ... ? The root causes are: 1️⃣ Legacy IDE's that shield the code from direct access 2️⃣ No CLI commands for things like code compiling (coding agents need feedback loops) 2️⃣ Fragmented IEC-61131 dialects (yes most automation engineers stayed in a single vendor ecosystem, but agents are general, and will be massively confused by this) But at least there's good news: next-gen IDE's are coming! This is what has already been announced/released: - Simatic AX - Beckhoff PLC++ - Codesys Go! - B&R Automation Studio Code ... and more coming that are not announced yet. My recommendation? Stop putting effort in legacy IDE's. We'll need to transform our coding workflows. It will take effort. Using coding agents with legacy IDE's will be almost impossible to control. Spend the effort by taking the opportunity to switch to a next-gen IDE. Talk to your vendor. Don't take 'not ready' for an answer. (Send me a DM if you don't know which person(s) to talk to) And if you're one of the automation vendors? Well we have been asking to fix the IDE's and toolchains since many years. At least some of you realized the need and took action a while ago. But there is no way to hold this back. We are losing competitiveness, speed and talent. Things still feel like they are moving relatively slowly. Yet soon it will be suddenly. How about the DCS system? Well ... you all know that I live and breathe process automation. There also is a viable path forward there. But no vendor seems to see the path through the trees yet ... 🤷‍♂️. I am not at SPS, so add a comment if I missed something ... 🫵 👇

28

2 Comments

Just reviewed the source code of @ethena_labs USDe on TonViewer.com; here's an initial analysis of the contract code: 1/ The code is essentially the same as USDT (@Tether_to) on #TON. It still allows for contract upgrades, admin changes, and the assignment of a new admin role. 2/ The main difference is that USDe includes a feature to designate the "receiver" of the USDe. This is a minor update and optimization compared to USDT, similar to the OFT contract. My perspective -- overall, it seems like an extension based on USDT's code safety capabilities; successful cases could lead to more adoption at the blockchain level, attracting more clients to choose us! Source Code: https://lnkd.in/ga4E7tb5 https://lnkd.in/gs25DGCq *Note: Assets like USDe still have "Status" management, which means even if your funds are compromised, @ethena_labs can freeze your balance. Avoid engaging in malicious activities! Though progress is slow, we are improving. Let's enhance @TON_Blockchain! https://lnkd.in/gth8M7aH

8

What are recursive language models (RLMs)? And why have they become the focus of so much attention in the past few days? An explainer on RLMs for the WHERE MACHINES THINK Substack: https://lnkd.in/dxBhQUSH

24

1 Comment

Business managers have a misperception that open source engineers are business people, too. So, a business manager may expect that an open source maintainer understands the imperative to answer the questions they have about a maintainer's software bill of materials. I spoke with Christopher Robinson of the Open Source Software Security Foundation at the Open Source Summit here in Amsterdam about this gap and the ways it extends into the inability for funding to flow to the open source maintainer community. So, why projects like Curl face a lack of support from manufacturers, there is also no easy way to get the money to maintainers who are not business people. They just want to create software, not run a business.

15

2 Comments