My AI Stack — April 2026: What I'm Using, What I Dropped, What's Next

My primary AI assistant for everything — writing, coding, analysis, brainstorming, document creation. Claude handles probably 60% of my daily AI interactions. The long context window is what keeps me here. I can paste entire codebases, full documents, or multi-page spreadsheets and get coherent responses. I've tried switching to ChatGPT multiple times and keep coming back.

The backbone of my entire automation infrastructure. Self-hosted on my Mac mini M4, running 40,000-60,000 executions per month across all my projects. Handles lead gen pipelines, content distribution, AI agent orchestration, monitoring, and notifications. The AI Agent Tool Node in n8n 2.0 has been a game-changer — I'm building workflows where AI agents chain tool calls and make decisions autonomously.

Local image generation running on my Mac mini. I use it primarily for my Night Caps YouTube channel — generating ambient scene images that feed into the FFmpeg video pipeline. Also use it for blog thumbnails and social media graphics. SDXL at 1344x768 runs fast enough on the M4's GPU that I can batch-generate 20-30 images while doing other work.

The unsung hero. FFmpeg handles all my video and audio processing — the Ken Burns effect on ComfyUI images, 3-layer audio mixing (ambient + nature + noise), and final render to 8-hour YouTube videos. It's free, it's fast, and it does things that would cost hundreds per month on cloud rendering platforms. Not glamorous, but irreplaceable.

Local LLM inference. I run smaller models through Ollama for tasks that don't need Claude's full power — quick text classification, data extraction, template filling. Keeps my API costs down and my data local. On the Mac mini M4 with 32GB RAM, I can run 7B-13B parameter models at usable speeds.

The machine everything runs on. n8n, Ollama, ComfyUI, OpenClaw — all on a $599 base machine sitting on my desk. It runs 24/7, uses about 15 watts at idle, and handles everything I throw at it. Best tech purchase I've made in years. If you're building any kind of AI automation stack, this is the hardware to start with.

My AI agent framework. Molty is the agent personality running on OpenClaw with a custom SOUL.md and AGENTS.md configuration that enforces execute-first behavior. Primary model is GPT-5.3 Codex with a fallback chain through DeepSeek, Claude Sonnet, Gemini Flash, and local Ollama. The emergency Telegram reset trigger has saved me from stuck loops more than once.

Analytics across all my sites. I run custom named click events on every affiliate link so I can track exactly which content drives conversions. Not exciting, but you can't optimize what you don't measure.

Lead enrichment and prospecting for outbound campaigns. The data quality has been solid for B2B contacts. I pair it with Instantly for cold email sequences. Together they've been the engine behind my lead gen pipeline.

My notification and control layer. n8n sends me Telegram alerts for workflow failures, daily revenue summaries, and content performance updates. I also use a Telegram command to trigger Molty's emergency reset when things go sideways. It's the simplest, most reliable notification system I've found.

I still think Make is great — and I recommend it to anyone who doesn't want to self-host. But I fully migrated my last remaining Make scenarios to n8n this month. The tipping point was AI agent capabilities. n8n's native LangChain integration and Tool Nodes let me build agent workflows that Make simply can't match yet. No hard feelings. If n8n's self-hosting overhead ever gets to me, Make is where I'd go back.

I was paying $30/month for Midjourney and using it maybe twice a week for blog thumbnails. Once I dialed in my ComfyUI workflows with SDXL, I realized I could generate comparable quality locally for free — and with more control over the output. The Discord-based interface was also increasingly annoying for production use. Cancelled.

AI-native code editor that's been getting a lot of attention. Two weeks in, I'm impressed by the codebase-aware autocomplete and multi-file editing. The background agent feature — where it works on a task autonomously while you code something else — is promising but inconsistent. Jury's still out on whether it's worth $20/month over Claude Code, which I get through my existing Claude subscription. Will report back next month.

Newsletter platform for The Operator Stack Weekly. Just set it up this month so it's early, but the editor is clean, the analytics are solid, and the free tier is generous enough to validate before spending money. Testing the lead magnet → email sequence → paid tier pipeline. If growth hits 1,000 subscribers, I'll upgrade.