This tracks with a broader pattern in technology adoption. The most productive people with any new tool are never the ones who try to use every feature on day one. They're the ones who pick one workflow, master it, and expand from there. Kit learned this the hard way, and his honesty about it is worth more than a hundred setup tutorials.

The Specific vs. Autonomous Problem

Kit draws a sharp line between two modes of using AI agents. When he gives specific, detailed instructions — read my Twitter DMs, find the sponsorship message, create an invoice on Infact, send it via email — the results are excellent. When he lets agents run autonomously on heartbeats and cron jobs, things go sideways fast.

His email archiving bot is the perfect cautionary tale. He gave it strict rules about what to archive. It still archived an email from a potential investor because it "seemed like cold outreach." Even if that mistake only happens once in 100 emails, it's the kind of error that makes you shut the whole thing off. And he did.

This is the fundamental tension of AI agents in 2026. They're powerful enough to do real work but not reliable enough to do it unsupervised. The gap between "can do" and "can be trusted to do" is where most people's frustration lives. Kit's solution is pragmatic: use agents as force multipliers for specific tasks, not as autonomous employees. At least for now.

The 39 Use Cases — Highlights and Commentary

Kit rattled through dozens of use cases, from practical to whimsical. A few stand out. His Pi-Hole network setup — blocking ads across every device in his home — took 30 minutes of back-and-forth with his agent. Not magic, but faster than doing it manually with ten years of procrastination behind him. His dental record visualization pulled data from bank transactions and vet emails to build a 3D mouth map showing every procedure on every tooth. Absurd? Maybe. But try explaining your dental history in a foreign country without it [1].

The custom podcast feed is genuinely clever. Kit downloads podcast episodes, transcribes them, strips out segments he doesn't care about — sports talk, political tangents, ads — and restitches the audio into a clean personal feed. It's the kind of thing that sounds like a gimmick until you realize how much podcast listening time is spent skipping through content you didn't ask for.

His Android phone integration is where things get serious. By switching from iOS to Android, Kit gave his agents full control — notifications, screen brightness, do-not-disturb, lock screen wallpapers, even deploying custom-built native apps directly to his phone. He vibe-coded an entire app store for himself. New apps get built on his Mac Studio and installed on his phone via ADB automation. No App Store approval needed [1].

The Safety Question Nobody Wants to Talk About

Kit identifies three prompt injection vectors: email, messages, and browsing. His approach is layered defense — a jury of five independent LLMs that judge every incoming email for safety before his agent can access it, a locked-down messaging setup where only trusted contacts can reach him, and a dedicated agent-only browser (he's calling it "Clauser") that scans websites before allowing interaction.

His strongest recommendation is simple: use smart models. Opus caught prompt injection attempts and laughed at them. Cheaper models like Gemini Flash might fall for the same attacks. This is a point worth emphasizing — in a world where your AI agent has access to your email, your bank, and your home devices, the cost difference between Opus and a budget model is the cheapest insurance policy you'll ever buy. Kit spends $600/month on LLM credits. For what he's automating, that's a rounding error.

The Real Takeaway

The most important moment in the conversation isn't any single use case. It's when Kit says: "It doesn't matter who owns it at this point. It opened our eyes to how the future could be." He's right. OpenClaw — or whatever replaces it — has shown that personal AI agents aren't science fiction. They're messy, frustrating, sometimes broken, and occasionally magical. The people who will benefit most aren't the ones with the flashiest setups. They're the ones willing to do the boring work of writing good instructions, building one skill at a time, and accepting that the honeymoon phase ends.

Kit's 39 use cases aren't a flex. They're a field report from someone living six months in the future. Some of it works beautifully. Some of it is held together with duct tape and optimism. All of it is pointing in a direction that every knowledge worker, business owner, and curious tinkerer should be paying attention to. The question isn't whether AI agents will run your life. It's whether you'll be ready when they can.

It's like someone is giving you a god computer that can do anything. And your first thought is, I don't know what to use this for. That's mind-blowing to me.

— Kit, OpenClaw power user