Ned Batchelder

JSON-structure is a modern practical alternative or replacement for JSON-Schema by Clemens Vasters - it's a draft IETF standard and the SDK now includes a Perl implementation after some not so gentle nudging. Clemens has been enthusiastically responsive to feedback and knows his stuff, if you're working with JSON and Perl (or other languages) this is well worth taking for a spin and sending any bug reports or suggestions.. code is at https://lnkd.in/e5fZvw_F and I'm looking forward to seeing the perl library on cpan shortly. #JSON #perl #javascript

8

This is the kind of story we need to be telling about LLMs. Not hype. Not job replacement. Not techbros. Instead, a friend and colleague finding their life improved by a new technology. Bridges bring people together. Thanks for your honesty and vulnerability, Shawn Zamechek. This was the story I needed to hear today.

8

The most succinct way to put how AI is extracting value from creatives and humans in general without giving anything back .

1

Visual Studio Code 1.109 positions itself as "the home for multi-agent development," so I put that claim to the test with a hands-on proof of concept outside the usual coding scenario. Instead of running one large, monolithic AI task, I broke an editorial formatting workflow into multiple subagents, each with a single responsibility, and added a final QA pass. The result was slower execution, but noticeably higher reliability on long, code-heavy documents. What stood out most was inspectability. Clear handoffs, visible intermediate results, and a timestamped report made it much easier to verify what happened and catch issues near the end of the file. This experiment reinforced that multi-agent orchestration is less about flash and more about quality control — a pattern that translates directly to production coding workflows where missed steps and silent failures are costly.

3

Another development in the seemingly never-ending cacophony of conflicting messages for software engineers over the last three years. Engineers are "more important than ever" at Anthropic despite the fact Claude is doing a lot of the legwork with coding these days. Yet CEO Dario Amodei thinks AI will be doing "most, maybe all, of what software engineers do end-to-end" within six to 12 months. The script will likely flip next month.

7

1 Comment

��� Woot, woot, very cool to see: Folks giving #Hardwood a try for parsing their Parquet files. Performance hasn't been a big focus yet, but it seems good enough when processing entire data sets (thus not requiring predicate push down). 👉 https://lnkd.in/eaty-Q6g

23

In the world of Retrieval-Augmented Generation (RAG), pure vector retrieval often suffers from reduced recall when dealing with sparse queries or out-of-distribution terms. That’s where hybrid search comes in. By combining dense vector retrieval with lexical methods like BM25 (TF-IDF + document length normalization), we can significantly improve recall and overall answer accuracy — especially in production-grade agent pipelines. Don’t just chase semantic similarity — anchor it with lexical precision. #HybridSearch #RAG #BM25 #InformationRetrieval #LLM #VectorSearch #AIEngineering #MIR #SearchSystems

14

"My Claude Code performance has tanked and I'm not sure why" This is one of the most common posts in Reddit's Claude Code community. In this post, I will share why it happens and what the fix is. Read on! The first question everyone asks: Did Anthropic change something on the server? Curious, I searched around. Turns out, users have been asking this all year long. --- I had the same experience. When I started using Claude Code for a new project, it was magical. As my project grew bigger, Claude Code started: - Violating best practices I gave it - Hallucinating code that contradicted my instructions - Claiming completion when tasks weren't done I wrote about my journey here: https://lnkd.in/gXjGiPZS My hypothesis? Context collapse. --- Anthropic just released a best practice guide that confirms this. https://lnkd.in/gJW_iSgc Here's what I found from research: - Performance starts to degrade at 40-50% of context window - The more best practices you give, the faster it fills up - When context is filling up, Claude ignores your instructions So if you give Claude Code lots of best practices to follow, and your codebase keeps growing... It means nothing if Claude Code didn't follow them. --- The good news? There's a fix. (1) Break best practices into individual skills. Skills allow Claude to load best practices on demand. (2) Agents + Skills. Skills alone isn't enough. When I tested, Claude loaded 5 relevant skills at once. Context filled up quickly. The real solution: One agent. One skill. (3) Apply at every stage: during planning, plan review, and implementation. (4) Create master skills for planning, plan review, and implementation. These skills repeat reviewers until all best practices pass. --- The irony? To use Claude Code effectively, you need to understand how Claude Code works. Vibe coding is not "just vibe." It requires engineering discipline. We are just shifting from software engineering to context and workflow engineering. --- Are you experiencing Claude Code performance issues? What's your workaround? #ClaudeCode #VibeCoding #AI #SoftwareEngineering #DeveloperExperience

10

2 Comments

This week, Anthropic's Claude 4 seems to be again on a roller coaster: after pushing Amazon's new Kiro IDE (which was briefly available for download before an avalanche of demand slammed it shut), his coding performance plummeted noticeably. Tool usage has dropped. The reason? Degraded. And as usual at Anthropic, no one knows why. We just get silence and a fresh dose of usage limits. Meanwhile, Microsoft's Copilot is getting a makeover literally. The new "appearance" feature could soon allow it to simulate traits like age or personality. Whether it's charming or dystopian depends on your tolerance for digital companions with synthetic charm. But one thing is clear: AI is no longer just smart, it's present. Speaking of IAG, Meta appointed Shengjia Zhao to lead its new Superintelligence Lab. If that name doesn't ring a bell, it soon will: Meta is delving into fundamental models with AI ambitions, indicating that the "big three" (Meta, OpenAI, Google) are gearing up for the final showdown. But it's not all polished demos and ambitious projects. A troubling moment at Sketch.dev exposed the dark side of AI programming assistants: a subtle AI-written database optimization caused a crash in production under load. Reminder: just because your AI can write code doesn't mean it should. Test, test, and test again. Mistral's Codestral also launched, quietly stealing the show. With 22 billion parameters and support for over 80 languages, it's open-source, meaning it has no barriers. Elsewhere, Google's NotebookLM received a major update powered by Gemini. It now summarizes YouTube videos and drafts of content like your personal caffeine research companion. Think of it as a fusion of CliffsNotes and ChatGPT: optimized, hyper-efficient, and incredibly good at synthesizing ideas. In the worst-case scenario for cybersecurity, researchers have revealed AI-powered malware that thinks for itself. It interprets its environment, adapts, and executes evasive strategies with chilling precision. It's not HAL yet, but it's time we reconsider the true meaning of "autonomous threat." Finally, some news that caught my attention: A new concept unlike the transformers and tokens we know today, the Hierarchical Reasoning Model (HRM) concept promises to make the large models we know more efficient. In its first milestone, a small model of just 27M was able to beat o3-mini on pre-trained tasks. for more information follow me here: https://lnkd.in/gNVYeZST

4

Goodbye GIL. You will not be missed. 🍷🐍 Python 3.14 is removing the Global Interpreter Lock (GIL)… after 30 years of acting like that one coworker who must hold the whiteboard marker at all times. For decades, Python could technically run on multiple cores, but the GIL said: “No. Only one thread at a time. Everyone else sit down.” But now… • Real multi-core parallelism • Threads that actually thread • Async that doesn’t break down crying • AI pipelines that don’t require spiritual forgiveness and message queues For Data & ML folks, this means: • Faster data processing • More scalable inference + agents • CPU-heavy workflows without “fine, I’ll rewrite it in Rust” energy This is not just a performance upgrade. This is Python getting on steroid 🏋️ (Note: Python 3.14 ships with both the standard build and the new free-threaded build — so adoption will roll in gradually.) #MLEngineering #Pyhton #AI #DataEngineering

30

New: I spoke with the head of Claude Code, Boris Cherny, about how the viral AI coding tool is reshaping Anthropic. The buzz around Claude Code has reached a fever pitch recently, but some developers say AI coding products in general are reaching an inflection point. A few years ago, AI coding tools basically amounted to autocomplete, suggesting lines of code after developers started typing. Now, the space looks completely different. Developers can spin up AI agents from their phone that will spend hours coding up a feature. Just how good are these tools? Well, engineers inside of Anthropic seems to be using Claude Code for almost everything, and its business is growing like wildfire. - Claude Code made up ~12% of Anthropic's ARR by the end of last year, WIRED has learned. The product soared past $1 billion in ARR, by at least $100 million, in December, and could play a significant role in the company's revenue growth moving forward. - Cherny says nearly everyone at Anthropic, even its sales team, is using the agent for almost everything. That said, competition in AI coding is fierce. Cursor is at least as large as Claude Code, in terms of revenue. OpenAI and Google are hot on their tails. Definitely a space to watch in 2026. Read more in my newsletter, Model Behavior. https://lnkd.in/gCUzcjd2

213

9 Comments

 I spent the last few months building a server framework, (think Rails with gRPC and a message bus) in Zig to understand how sub-millisecond systems actually work. It's a pet project. Not for production, just for learning. Why Zig?  - C++: Didn't want to deal with CRTP (started before C++26 reflection)  - Rust: Couldn't prototype freely while fighting the borrow checker  - Zig: Why not? What I learned: Architecture matters way more than micro-optimizations. Switching from blocking I/O to epoll gave a 1,139x improvement. Everything else I did afterward combined for maybe 25%. Zig's inline for over comptime tuples is wild. You write generic code, and it compiles down to direct function calls. No vtables, no hashmaps. Got it to 13,667 req/s with 0.30 µs publish latency and zero allocations per request. I'm currently writing a blog series walking through the whole thing, comptime dispatch, lock-free message bus, binary protocol parsing, UDP multicast. It's open source. If you work on real-time or low-latency systems, I'd genuinely appreciate feedback on what I'm missing or what could be better. As a disclaimer, as I wish more people would do this, I am definitely having AI assist in my writing, as I often cant translate my thoughts to cohesive text. Source Code: https://lnkd.in/eY-9D8De Blog Post: https://lnkd.in/ePY4xpKX  #Zig #SystemsProgramming #LowLatency

24

New blog post on progress in directed graph hashing. Here I discuss the new concept of directed graph language, which is very similar to DFAs/NFAs: https://lnkd.in/eazxiQUN

13

1 Comment

Imagine if this was code flying a fighter jet loaded with munitions (Helsing). Trust, but verify. AI/LLMs are amazing, it’s this kind of gap that we need to close next if Claude (et. al.) is to survive. I’m sure this isn’t news to anyone, but cautionary tales like this are worthy of repeating.

2

It’s been quite sometime since I started working on the topic of E-ID.. It’s a completely different beast a typical SAAS that I used to build before! Along the way, I’ve gathered many resources, courses, books, etc. on the subject trying to gather all the threads to build a good mental model of this domain. We compiled these resources in the article shared by Linus below!

10

“the properties of a sound test suite * Predict success in production * Fast to get feedback * Support ongoing code design change * Low total cost of ownership” Great read.

1

Just finished reading Meta’s engineering post on their transition back to upstream FFmpeg. It brought back some memories of the specific challenges we face in the media processing world. A couple of things really hit home: ➡️ The "Fork" Debt: Maintaining an internal version of an open-source tool like FFmpeg is a massive undertaking. I’ve spent far too many hours in the past manually merging code and fixing broken unit tests because our internal interfaces drifted too far from upstream. The risk of regressions is constant, and it’s the kind of "invisible" work that can really drain a team’s velocity over time. ➡️ API vs. CLI: It’s great to see the FFmpeg CLI getting a major architectural overhaul for better parallelism. However, for the production systems I've built, I still find myself leaning toward direct API integration. When you need deep observability, custom event callbacks, or a specific integration with your monitoring stack, the CLI can sometimes feel like a "black box" that’s hard to peer into. It’s a solid move by Meta to push these changes back to the community instead of keeping them behind closed doors. It makes life easier for everyone else in the industry. #FFmpeg #MediaProcessing #OpenSource #Meta https://lnkd.in/eeJtj9yy

8