How fast must your local AI be? The obvious answer is: my local AI must be as fast and as smart as the one offered by OpenAI (or other commercial vendor). "I want the best quality, and I want it now!" Ok, I get that, but let’s focus on the business reality for a sec. Probably, you won’t be able to match the performance of commercial AI providers. And that is totally fine.

Not always the latest, greatest, most hyped solutions deliver the best business outcomes. Often, it is hard to measure the actual impact of the “new shiny thing” on the business. Everyone writes about the MTP (Multi-Token Prediction) and how it improves the LLM performance. I wanted to leverage it to boost my local AI development team. My business case was the following: I wanted to switch from Qwen3-Coder-Next-UD-Q4_K_XL to Qwen3.6-27B-MTP-UD-Q4_K_XL for local agentic coding.

That is the common belief I see on the internet. I do not agree with that and have a process to prove otherwise. Currently, open-source models that can run on affordable hardware are not as capable as commercial AI models running in huge data centers. Everyone knows it, and I do not argue with that. My point is that it is perfectly possible to use open-source models running on local hardware to generate enterprise-class solutions that directly impact business operations and provide tangible business results.

I use local AI agents to build enterprise-grade software, not fancy demos. After testing all major models that can run on affordable local hardware, I am convinced that the Qwen3-Coder-Next is the best option. It might not be the “most fancy” model, it might not design “most outstanding” websites, but (typically) it does deliver enterprise-grade outcomes. Yes, I am aware Qwen3-Coder-Next is not as good as II from OpenAI or Anthropic.

I interviewed Hans Stam about practical lessons from building hardware in China versus Europe and why “hardware is hard” but becomes manageable when you understand the process. Hans explains how Chinese factory ecosystems move from prototype to mass production. The conversation covers pitfalls in Europe’s decision-by-committee culture, critical-path project planning, choosing certification partners, selecting Chinese factories, CapEx vs. OpEx trade-offs, component shortages, and many other topics. 👉 Let me know if those topics are relevant to your business.

Intro Vanilla coding agents are optimized for shipping web apps fast, and they are great at building SaaS products. But if you’re connecting sensors, managing edge deployments, or orchestrating hardware fleets, the default agents get completely lost. Connected infrastructure has constraints that generic AI has never encountered during training. Missing training data Hardware-specific protocols MQTT, Modbus, BACnet, LoRaWAN, Zigbee — these aren’t in the training data at scale. An agent that’s great at React components will hallucinate when building solutions using those technologies.