Erik Dietrich

See the fastest LLM serving engine, 100% open source!!! If you’re running multi-round QA, RAG, or any long-context workload, you know the pain: slow time-to-first-token and expensive bills. You should take a look LMCache, a next-gen LLM serving engine extension designed to slash TTFT (time to first token) by 3–10x and boost throughput. How? LMCache stores KV caches from any reusable text (not just prefixes) across GPU, CPU DRAM, and disk, then shares them across serving instances. That means fewer recomputations, faster responses, and huge GPU savings. Key highlights: 🔥 Seamless integration with vLLM v1 ⚡ High-performance CPU KV offloading 🔗 P2P KVCache sharing across servers 💾 Storage on CPU, Disk, or NIXL ✅ Supported in vLLM production stack, llm-d, and KServe 📦 One-liner install: pip install lmcache If you’re scaling LLM inference, you should check it! 👉 GitHub link in the comments. #LLM #AI #Inference #OpenSource #RAG #AgenticAI

378

27 Comments

New video: When a datalayer push does not work, the cause isn’t always obvious. Learn how to fix the issues from simple typos in the code to more complex problems with how the data layer is configured on your website. https://lnkd.in/gbAbVRBN #gtm #googletagmanager

110

4 Comments

The fundamental coding agent abstraction is the CLI. It’s not a UI or form factor preference, it’s rooted in the fact that agents need access to the OS layer. Coding agents are, at their core, computer-use agents. They run programs, create new ones, install missing ones, scan the filesystem, read logs. More than text editors, they’re *automating your computer* at a low level. It’s more accurate to think of Claude Code as “AI for your operating system” than a continuation of the copilot autocompletion → IDE AI assistance paradigm. Another consequence is that it works for any computer. Your desktop computer, but crucially, also the ones running in the cloud. 𝚗𝚟𝚒𝚖 nerds like yours truly always liked that it’s an editor that runs on your Mac but also any machine you 𝚜𝚜𝚑 into. Imagine if you could launch a million Claude Codes concurrently to tackle a bug, find cybersecurity threats, process your issue backlog, build features based on user feedback, run QA… That’s why we’re building vercel.com/sandbox as the infinite compute layer for agents.

499

29 Comments

Innovation often hides behind wrappers. MCPs promise scale and speed, but let’s be honest: they’re just orchestrated prompts and APIs. Valuable? Yes. Foundational? No. The deeper question: are we funding primitives—or just better packaging? The history of technology is littered with companies that confused distribution hacks for paradigm shifts. Efficiency matters. But leaders must know the difference between an infrastructure play and a true leap in capability. So—what deserves our capital: elegant wrappers or new primitives? https://lnkd.in/g8yUQzDb #AI #InnovationStrategy #Technology

14

2 Comments

Remember when embedded development meant poring over register definitions and writing every line of code from scratch? Those days are fading fast. RTOSes, HALs, and auto-generated code now handle much of the heavy lifting, making development faster and more accessible. But there's a trade-off—as fewer engineers touch low-level code, we risk trading efficiency for convenience. Many engineers never touch the metal. They don’t need to. They grab a vendor library, visit GitHub, or ask AI. But what happens when performance bottlenecks creep in? When debugging goes beyond the abstraction layers? When an RTOS just isn’t precise enough? Bare-metal isn’t dead; it’s evolving. It’s becoming a specialized skill rather than a daily necessity. But when SDKs are too slow, power budgets are tight, or real-time constraints demand control, bare-metal expertise separates great engineers from good ones. So, where do you stand? Do you still have the skills to program at the register level, or are you riding the abstraction wave? Let’s discuss: has bare-metal programming truly disappeared, or has it just changed its role in modern embedded systems? Let me know what you think below. #EmbeddedSystems #RTOS #BareMetal #Firmware #SoftwareEngineering

167

51 Comments

Vertical software companies will say "we have embedded payments." I will ask "Do you have adoption?" 80% of the time, crickets. Often, "Embedded Payments" seems to mean payments simply sold 'with' software. An adjacent module, perhaps linked via SSO. That approach doesn't capture the real potential. True power isn't found in mere adjacency. It demands a deeper fusion. Like metals melted together, becoming inseparable. The payment capability needs to integrate until it's a core, invisible part of the user's primary workflow. Uber demonstrated this brilliantly. There’s no distinct ‘payment’ step. It vanishes into the background of the ride itself. Seamless. Automatic. Dare I say, cool? When integration reaches that depth, becoming an unnoticed, fundamental aspect of the experience, payments deliver that outsized value. The software doesn't just OFFER payments. It BECOMES the payment experience.

43

13 Comments

Two weeks ago, Steve Yegge released GasTown: a multi-agent orchestrator he describes as "an industrialized coding factory manned by superintelligent chimpanzees." A few days later, Dan Lorenc, inspired by GasTown, pushed multiclaude, built on what he calls the "Brownian Ratchet" principle: chaos is fine, as long as we ratchet forward. They landed on almost identical architecture: - tmux for observability - git worktrees for isolation - External state persistence - CI as the final arbiter While they have very different approaches (i like both! for different reasons!), when two experienced engineers solve the same problem and arrive at the same primitives, that's not coincidence. That's the problem space telling you what it actually requires. The "19-agent trap" - simulating an org chart with Analyst → PM → Architect → Dev → QA handoffs - is giving way to something more operational. Bounded roles. Infrastructure primitives. CI as coordination mechanism instead of real-time sync. The chimpanzee factory and the Brownian ratchet asked different questions but got the same answer. If you're building agent systems, these are probably your primitives too.

251

26 Comments

Claim Errors per Service Categories: They are widely different The chart below shows CMS error rates by service type from their latest CERT review - and some of these numbers are eye-opening. The error rate of lab tests is simply astonishing at 40%. I know from my experience that various lab tests are overused, but seeing that 4 out of 10 claims have issues is a refreshing reminder of how much work we have to do on appropriate utilization and documentation. Wheelchairs seem to be provided too easily as well - those medical necessity denials tell a story. Imaging shows similar patterns at 30%. By the way, this is based on claim volume, not dollar values. If you plot by dollar exposure, I believe the chart would look quite different - those expensive oncology and parenteral nutrition errors would probably dominate. What's your take? Are we seeing overutilization, documentation gaps, or both? #Healthcare #Medicare #Compliance #CERT

30

9 Comments

If you're looking for a new docs framework for your dev tool, I can't recommend Docusaurus enough. 1. You can change the appearance of any component—no matter how big or small—by ejecting its raw React code into your source tree. I've ejected almost all of its major components and restyled them with Tailwind. 2. It supports content written with MDX, which is like Markdown but better. MDX lets you write JSX tags so you can render React components (like badges and banners) directly inline your content. No more messy templating. 3. It bundles Prism for code snippet highlighting. So you if you want to change how each and every token of your language gets styled—which I do—you can override it with a simple centralized map—which I have. 4. You can swap out its on-site search engine with Algolia, now widely considered the best option for developer sites. It is, in other words, the perfect tool for people like me who are obsessed with every pixel.

118

16 Comments

Interoperability isn’t just a technical problem, it’s a human one. When we talk about interoperability in verifiable digital credential systems, too many conversations stop at technical standards, APIs, and data formats. Yes, these matter, but they’re only one layer of the stack. Real interoperability lives in the social and cultural systems that surround the technology: ✨The data models that determine what gets recognized and what gets left out. ✨The governance frameworks that shape who gets a seat at the table and who doesn’t. ✨The trust we build not just in the tools, but in the communities maintaining them. ✨The transparency to say when something isn’t interoperable, and the humility to fix it. Too often, vendors talk about interoperability as if it’s something you can solve with a spec sheet. Yet, if we’re serious about credentials that can move across systems, sectors, and societies we have to care as much about people, process, and power as we do about protocols. Interoperability isn’t just about systems talking to each other. It’s about humans agreeing on what’s worth saying. #DigitalCredentials #Interoperability #Trust #OpenStandards #Governance #DigitalIdentity #EthicalTech #PeopleFirst

31

2 Comments

These are some of my favorite lean phrases: 1. Wherever you ask the question is where the answer should be. 2. Use your head, not your wallet. 3. Fix what bugs you. 4. Leave everything better than you found it. 5. Attack the process, not the person. 6. Work should be easy. 7. Quality = Trust = Survival. 8. Money suffocates creativity. 9. Lean is the art of subtraction. 10. Your head is not a to-do list. 11. Make the process the expert, not the person. 12. Banish sloppiness, love precision. 13. Good thinking = good products. 14. Set the expectation, inspect the expectation, and celebrate the expectation. 15. “Costs don't exist to be calculated, costs exist to be reduced.” — Taiichi Ohno 16. 4 improvement criterias: safety, quality, simplicity, speed (in that order). 17. Work as if the customer is watching you. 18. If you don’t have time to do it right, you won’t have time to do it over. 19. Leaders improve people. People improve processes. Processes improve the product. 20. How you do anything is how you do everything. 21. Long-term consistency beats short-term intensity. 22. FOCUS: Follow One Course Until Successful 23. Real learning comes when you ask questions. 24. "Don’t look with your eyes. Look with your feet. Don’t think with your head. Think with your hands." — Taiichi Ohno 25. A company cannot develop unless its people are nurtured. 26. “Wealth is in your waste.” — Venu Srinivasan 27. Money is the shadow of doing the job right. 28. Borrowed knowledge is not equal to developed knowledge. 29. Sensei means teacher. Who are you learning from? And what are you learning from them? 30. If you're hurried and you don't have enough time then you're doing something wrong. Which one of these smacked you upside the head? 👇

12

8 Comments

When it comes to selecting the right DXP, things can go off the rails. But don’t lose your caboose. Dan Drapeau has a two-track evaluation model that can keep you rolling – full steam ahead. In today's Guest Critic article, Dan proves why he’s a masterful conductor on the digital experience train. As he explains, the DXP competitive landscape has converged across architectural models, and evaluation approaches haven’t evolved at the same pace. Dan’s observation is that single-track processes are still chugging along – focusing on feature breadth and demo cohesion and favoring pre-composed suites. In his experience, what we really need is a more balanced approach that separates platform evaluation from ecosystem architecture. How do you leave the station? By running two structured tracks, which he outlines in detail. This helps vet not only current capabilities but also long-term durability relative to core considerations like vendor lock-in, CDP and DAM strategies, and the ever-evolving target of AI. In his post, Dan details the operational requirements, where the complexities exist, and how to synthesize results. He even digs into the critical topics of validation and stress testing, and distills it all into a handy checklist to help guide your DXP evaluation. As he says, composable and pre-composed DXPs represent architectural strategies along a spectrum, and these pointers can help keep you on track. “A structured two-track evaluation approach ensures that whichever model is selected, it is chosen for structural alignment and long-term outcomes rather than surface cohesion.” If you’re currently evaluating DXPs – or considering it in the future – this is a fantastic read from a sharp practitioner. Check it out on CMS Critic: 🔶 https://lnkd.in/e-XB3Pig #dxp #digitalexperience #digitalexperienceplatform #softwareevaluation #dam #digitalassetmanagement #cdp #customerdata #customerdataplatform #cmp #contentmarketingplatform #ai #artificialintelligence

15

1 Comment

Any best guesses on how long until Claude Code can: 1. Fire up the browser (without needing the MCP configuration) to evaluate changes 2. Fully leverage the Chrome Dev Tools 3. Use the Ruby debugger I’m finding Claude Code is creating fixes that would have been too tedious or painful. The gap right now is testing and debugging.

10

4 Comments

GitHub's recent announcement about starting to charge a $0.002 per-minute platform fee for self-hosted runners doesn't directly impact customers who are using GitHub Actions with Nx Agents. The Nx Agents model lets you kick off a job on a lightweight GH runner, then *delegate* the heavy lifting to Nx Agents. The good news is that GitHub's cutting prices on all runners, so even the main job will get cheaper. In other words, Nx–in particular Nx Agents–isn't self-hosted runners. Nx is a platform you delegate your CI workload to from within a lightweight GH runner to make your CI way faster and way more compute-efficient. GitHub announced yesterday that they were postponing the introduction of the platform fee for self-hosted runners, so it remains to be seen if/when the fee will kick in. So for now, it's only good news for all GitHub Actions users because prices are going down. If/when the platform fee kicks in, the biggest impact will be to users of services built specifically to run as self-hosted runners and undercut GitHub's pricing, since pricing will be more on par with what you'd pay to use a GitHub-hosted runner. See the link below for more info on setting up Nx with GitHub Actions. (P.S. not written by AI, I just like em dashes) https://lnkd.in/gMMek4KJ

35

Missing form labels, empty buttons, links that go nowhere... Let's stop calling this "accessibility debt." This is just incomplete development. The real issue isn't that accessible code costs more to retrofit. It's that we're calling basic HTML requirements "extra work." A button without text? That's not accessibility debt - that's unfinished code. A form input without a label? That's not technical debt - that's skipping steps. A link that doesn't go anywhere? That's not an edge case - that's broken. These things should be in your Definition of Done already. Not because of WCAG, but because they're fundamental to how HTML works. When we frame basic semantic markup as "accessibility work," we make it sound optional. It's not. It's just good development. Start there. Get the fundamentals right. The rest gets easier once your team stops thinking of working buttons and labeled forms as "nice to have." #WebDevelopment #Accessibility #TechDebt

109

15 Comments

A lot of people instinctively think of audit as something that slows an organization down. As someone who comes from the “control owner background”, I saw a decent amount of that sentiment amongst peers. Engineers in particular can feel like they’re negotiating for their time when partnering with audit counterparts. I was reminded of that tension recently at a conference, where a speaker used an analogy that stuck with me. Most people similarly think of brakes on a car as the thing that slow you down. Or worse: grind to a complete halt. But if you talk to an F1 driver, they’ll tell you the opposite. Brakes are what allow them to go as fast as they do. Without them, you can’t take turns at speed. You spin out. Audit works the same way. When audit is treated purely as a checkbox or blocking function, it feels like drag. But when it’s designed to surface real risk and create clarity, it becomes enabling. It provides true governance and allows the organization to move faster without losing control.

95

4 Comments