I keep a Notion page full of ideas. For years, I’d open it, read through the list, and close it with a quiet sadness — not because the ideas were bad, but because building them would have needed a team, weeks of runway, and more hours than I had. Some of those ideas would’ve taken 2–3 focused developers months to build properly.
Then, in the span of about two months — December 2025 to February 2026 — everything changed.
Agents Were Niche. Now They’re Essential.
A year ago, AI coding agents were a curiosity. By late 2025, tools like Claude and Codex had matured significantly, but they were still mostly assistants — you’d ask, they’d answer, you’d integrate the output yourself.
Then something shifted. Better reasoning, autonomous tool use, and orchestration frameworks turned “assistants” into genuine agents — systems that could plan, execute, debug, and iterate on their own. What was niche became essential almost overnight. The developer conversation changed from “is this useful?” to “how is anyone building without this?”
My Ideas Used to Collect Dust
I’m a developer with too many ideas and not enough bandwidth. Some lived in Notion for years, slowly going stale, waiting for a version of me with more hours in the day. The backlog was demoralising.
That changed when I went all-in on agents.
My Setup: Self-Hosted, Private, Always On
I didn’t want to depend entirely on cloud APIs. I wanted something always available, private, and customisable. So I built a self-hosted AI stack on my Raspberry Pi:
- OpenClaw — self-hosted interface to manage everything
- Codex 5.3 — primary model for development tasks
- MiniMax — secondary model for creative/generative work
- Tailscale — secure access from anywhere on my network (Running on a Raspberry Pi — affiliate link, I get a small kickback if you buy one at no extra cost to you)
- Linear — project and task management, connected to AI workflows
- Notion — where ideas live, now feeding directly into active projects
Ideas go into Notion, tasks land in Linear, agents do the heavy lifting. I can access the whole thing from my phone or laptop from anywhere.
What I’m Building Now
My current project: take a set of lyrics and generate a full song — music, arrangement, everything — using Google’s music generation models. A year ago this would have felt like a stretch goal for a team. Now I’m building it solo, evenings and weekends, and shipping.
I’ll be posting the architecture, the open source stack, and lessons learned as I go — here and on Twitter. Stay connected.
The Real Unlock: Technical → Creative
AI agents didn’t just make developers faster. They shifted where the constraint lives.
Before: the bottleneck was technical. Could you build it? Did you have the time, the skills, the team?
Now: the bottleneck is creative. Do you have a good idea? Can you describe it clearly?
That’s a fundamentally different game. It opens the door to people who’ve always had ideas but hit a wall at implementation — not just engineers, but designers, writers, researchers, domain experts. The gate has been lifted.
That’s why everything I build is going open source. Bring your own models, run it on your own hardware, customise it for your needs. Creativity shouldn’t be locked behind a subscription or a proprietary API.
On AI and Human Creativity
There’s a reasonable worry: does AI-generated music rob from real creators?
I don’t think so — I think it expands who gets to create.
Every time a tool lowered the barrier to music — electric guitar, DAWs, bedroom production — people worried it would cheapen the art. Instead it brought in more voices, more genres, more experimentation. The same will happen here. The goal isn’t to replace musicians. It’s to let a developer who hears music in their head but can’t play an instrument actually make something.
The best human creators will still stand out — great ideas and great taste don’t get automated away. But the field gets wider, and that’s a good thing. I’m going to prove it by shipping.
What’s Next
Expect posts on:
- My full self-hosted AI architecture (Raspberry Pi + OpenClaw + Tailscale)
- The lyrics-to-music project: architecture, demos, open source release
- Ongoing project updates
If you’ve got a backlog of ideas that’s been making you sad — now’s the time. The gate is open.