Scott Arbeit

To MCP or not to MCP?? https://lnkd.in/gSt-k4FK When we're building automation tools in 2025, I see two main approaches: * New school: Agent + MCP * Old school: APIs + LLM sprinkles Despite MCP being the new hotness, I keep writing the old school way. Why? * Control * Accuracy * Cost / Token waste 💵 ^^ Read more details in the blog post and let me know what you think!

87

16 Comments

New blog post: A Fork in the Road: Deciding Kafka's Diskless Future. Kafka is getting serious about S3 and finds itself at an architectural crossroads that will shape its next decade. Several new KIPs (1150, 1176, 1183) aim to reduce replication costs across cloud availability zones, but the implications go far beyond networking cost. It’s a mistake to think of S3 as simply a cheaper disk or a networking cheat. Building on object storage opens the door to operational benefits such as elastic, stateless compute — something many modern analytics systems already exploit. In my latest post, I outline two competing future paths: 🔹 Evolutionary path: Reuse large parts of existing Kafka components to reduce code changes and long-term maintenance. 🔹 Revolutionary path: Separate stateless and stateful layers to realize the full operational benefits of disaggregated storage. The post examines the trade-offs of each path, how the current KIPs map to these two paths and poses a broader question: what should Kafka become? And what will keep it relevant in the decade ahead? https://lnkd.in/d4E2BAe4

149

12 Comments

We just introduced Docker Sandboxes as an early experimental preview. They give coding agents a safe, isolated workspace where they can run code, install packages, and modify files without affecting your host system. This initial version works today with agents like Claude Code and the Gemini CLI, with more agent support on the way. Coming next, we’re aiming for stronger isolation through micro-VMs, granular network controls, better secret handling, and the ability to safely run multiple agents in parallel within their own sandboxes. If you’re experimenting with AI agents, give it a try in Docker Desktop 4.50+. Feedback welcome as we continue to shape this. cc: Eric Jia and Kevin Wittek who're leading this work. https://lnkd.in/g89w_UUH

229

16 Comments

Pro tip for open source: There are a few OSS projects I've leveraged in the enterprise that were maintained by a single person. Lodash is one. Mountebank another. They always made me nervous but the way they communicated with users and the care they took about docs (popularity isn't enough for me) gave me enough confidence to use them. I was recently starting to evaluate another open source project that's been getting traction recently. I asked Claude to give some feedback on it. - Alpha release with an unstable API. Fair enough, something to test out. It's very popular, so that will improve - Single maintainer Not awesome, but not a deal killer - Docs and implementation do not match Hmmmm - 300+ open issues Hmmm Then I talked to someone who had ask the maintainer to do some consulting - Self aggrandizing - Hard to tell if he's really smart or a confident BSer (Elon Musk Syndrome) - Makes wild claims about accomplishments and people stealing ideas Yeah, I'm out. Open source is a mixed bag. You need to understand the risk profile of your supply chain. I'm looking for a core capability for an enterprise, not a toy, and this project is a pass. No, I won't say which one in public for obvious legal reasons.

25

11 Comments