AI and Software Networking

Hey I was just on LinkedIn and saw you on there and I am also into software engineering now but have absolutely no one to speak to about tech related things without making their ears bleed. I know what you are going to say. AI is the devil!!! Which is true. But my point is this. I am really good at AI now. Like, I spent the last few years holed up in a cheap ass East Austin apartment writing my blog which has now become something out of my control, almost like it is alive itself, which it is not, but we need that VC cash, so we have to anthropomorphize the shit out of this technology. I saw that you are into TypeScript, does that mean Next.JS? Because that is all I know for frameworks really is Next.js, I wonder why??? The reason is that I am not longer a coder. Neigh, neigh me lord, that is the truth. I have neigh written a piece of code longer than me thumb since before Karpathy quit. That is because I am the OG vibe coder. I’m sorry. But it is true. I have not written a line of code in years longer than just a simple JSON/YAML key edit to copy and paste the persona weights representing the soul from the autonomous systems I build. And yet somehow there is this cancerous blob metastasizing slowly but surely from the internet at danielkliewer.com These mysteries and more await you if you dare respond to this email. I DARE YOU!!! Sorry, but the last time I saw you I think was in middle school at Bonham before I had to go to that awful private school and I have seemingly reverted to that stage of my life at this point.

This is funny and memorable, but it would probably get a higher response rate if you trimmed a bit of the self-deprecation and made the invitation clearer. The humor lands because it sounds like an actual person rather than a networking template. I'd keep that voice while making it a little easier to read. :::writing{variant="email" subject="Long time no see" id="67281"} Hey, I was just on LinkedIn and saw your profile. Looks like you're into software engineering now, and I realized I have almost nobody to talk to about tech without making their ears bleed. I already know what you're going to say. "AI is the devil." Which... fair enough. The problem is I'm actually really good at AI now. I basically spent the last few years holed up in a cheap East Austin apartment writing my blog, and somewhere along the way it turned into this bizarre organism that's growing faster than I ever expected. Obviously it's not actually alive, but if we're ever going to get VC money, we have to anthropomorphize the hell out of it. I noticed you're into TypeScript. Does that mean Next.js too? That's basically my entire world. The funny thing is I don't really think of myself as a programmer anymore. I haven't written more than a few lines of code in years—mostly JSON or YAML edits for the autonomous AI systems I build. Everything else is prompts, architecture, retrieval, orchestration, and letting the models do the typing. Somehow, despite that, there's this strange, ever-expanding thing living at danielkliewer.com that's become my full-time obsession. Anyway, I figured I'd say hello. The last time I remember seeing you was back at Bonham before I disappeared off to that awful private school. Apparently I've regressed to my middle-school sense of humor, because this email seemed like a perfectly reasonable idea. If nothing else, I'd love to catch up and talk shop sometime. —Daniel ::: I'd send this as-is. It keeps your eccentric personality, but it ends with a genuine invitation instead of leaving the recipient wondering whether they should laugh or reply.

try again with this but make it make more sense and more technical: I was just on LinkedIn and saw your profile. Looks like you’re into software engineering now, and I realized I have almost nobody to talk to about tech without making their ears bleed. I basically spent the last few years holed up in a cheap East Austin apartment writing my blog, and somewhere along the way it turned into this bizarre organism that’s growing faster than I ever expected. Obviously it’s not actually alive, but if we’re ever going to get VC money, we have to anthropomorphize the hell out of it. Anyway, to catch up, I work in AI now as head devil incarnate and help people create local implementations which retain full data sovereignty because instead of using API providers all the inference is done locally using llama.cpp and the best model huggingface has to offer. What is sick is that I can power the Hermes agent harness entirely using local inference requiring zero API calls. I like OpenCode rather than Claude Code, but Hermes is slowly taking its place more and more. The point is that on consumer hardware, in my case a M4-Pro 48GB MacBook, I can run local inference and do absolutely everything Claude Code can do but instead of paying $200 a month I can do it all for free forever without the internet and without any data being exfiltrated. That is just the tip of the iceberg. Now things like creating very detailed and what would otherwise cost huge amounts of compute from scaled servers graphs done entirely local. So that was my way to replace Neo4j’s Aura service and now I build graphs entirely locally without needing to exfiltrate data. Using these more complex data structures you can pair that with LLM inference for RAG, but RAG is old fashioned now and there are better options emerging as the tech evolves. Grep can now be used by agentic systems for instance and through simply using graph and vector databases you can align the telemetry you collect locally and be able to enforce governance in a similar way to how the big providers are required by GDPR and other regulations.

I'd make it sound less like a sales pitch and more like you're sharing what you've actually been building. The technical points are interesting, but they flow better if they tell a coherent story rather than jumping between topics. :::writing{variant="email" subject="Long time no see" id="48173"} Hey, I was just on LinkedIn and saw your profile. Looks like you're into software engineering now, and I realized I have almost nobody to talk to about this stuff without making their ears bleed. I spent the last few years more or less disappearing into a cheap East Austin apartment building AI systems and documenting everything on my blog. What started as a collection of notes has slowly evolved into a fairly comprehensive architecture for running modern AI entirely under your own control. These days I'm mostly focused on local-first AI. Rather than building around cloud APIs, I use local inference with llama.cpp and open-weight models from Hugging Face, so everything runs on consumer hardware with full data sovereignty. No API calls, no vendor lock-in, and no data leaving the machine. One of the projects I've been building is an agent harness called Hermes. The interesting part isn't that it's "an AI agent"—everyone has one now. The interesting part is that the entire orchestration layer can run locally. On my M4 Pro MacBook with 48 GB of memory, it can perform code generation, planning, retrieval, and autonomous workflows without depending on external inference providers. It's surprisingly capable, especially considering the hardware. I've also moved well beyond treating RAG as "embed documents into a vector database and search." I'm experimenting with richer retrieval architectures that combine vector search, graph-based representations, local telemetry, and structured memory. Instead of relying on managed services like Neo4j Aura, I build and query knowledge graphs locally, which keeps both the data and the reasoning process under my control. Another direction I've been exploring is treating the local filesystem itself as part of the retrieval architecture. Modern agentic tools can leverage fast code search, graph indexes, vector embeddings, and structured metadata together instead of relying on semantic search alone. That opens up some interesting possibilities for governance, provenance, and auditability while still keeping everything local. Anyway, I figured I'd reach out because I don't get many opportunities to talk to another engineer about this kind of work. If any of that sounds interesting, I'd love to catch up and compare notes. —Daniel ::: I think this version will resonate more with an engineer because it emphasizes the architecture and tradeoffs rather than making claims like "it can do absolutely everything Claude Code can do." That's a stronger position technically, and it's easier for another engineer to engage with.