Building a side project with AI agents: what's fast, what's slow

People ask: "So you built the site with an AI agent in just a few days?"

Honest answer: the core, yes — a few days. The product I want to live with takes longer — not because AI is slow, but because polish and ownership don't outsource well.

This is a condensed diary from mylittleworld.dev — personal corner, not a tutorial.

Where it's actually fast

Scaffolding and boilerplate

Monorepo, Next.js, Hono API, Drizzle, i18n, Docker — solo from zero could be a week. With an agent I had a running skeleton in 1–2 days.

Agents are good at:

• files that match existing patterns
• fixing repetitive import/type/config errors
• boring CRUD from a schema

Midnight bugs

"TeVo's second chat message returns empty", "Studio generate 500" — describe symptom, paste logs, get a patch. Shorter debug loops.

Small UI passes

Daily brief overlapping chat, Bug Slayer Start clipped on mobile — screenshot, expected behavior, Tailwind tweak. No Figma for every pixel.

Bilingual content

Blog seeds, changelog, TeVo copy — agent drafts, I tune voice. No copywriter for every line.

Where it's still slow

Product decisions

TeVo chat or not? Daily brief? Bug Slayer in Studio — who decides? Me. The agent doesn't know "enough" vs "too much".

UX polish

Motion, spacing, mobile feel, collapsible chat chips — 10–30 minutes each, compounding into weeks. Agents suggest; taste stays human.

Production and ownership

SSL, nginx, missing DB tables, rate limits, prod env — agents write scripts, but first real deploy is still me on SSH, reading logs, understanding the system. Responsibility doesn't shortcut.

Review and trust

Merging agent code unread = tech debt. I still read diffs, run lint, manually test critical flows. Faster than solo, not faster than thinking.

How I work with an agent (no drama)

I treat it like a part-time dev friend:

• Me: owner — scope, priority, ship / don't ship
• Agent: implement, draft, grep the codebase
• Rules: one task at a time; lint when done; checklist for prod

Same vibe as my corner of the internet: I own it, the agent is a trusted helper.

If you're building a side project this way

1. Expect fast phase 1, long phase 2 — phase 2 is polish + production.
2. Keep a tiny smoke suite even solo (ten tests).
3. Write a changelog — future you will forget why TeVo exists.
4. Don't compare to Twitter demos — they show day one, not week six fixing the footer.

Closing

An AI agent doesn't turn a side project into one prompt. It turns one person into a smaller team — still tired, less stuck on boilerplate.

I build because I want a corner that's mine. The agent gets it online sooner. Quality and story — still written bit by bit, late nights included.

---

My Little World still gets tweaked daily. If you're building too — be fast where it counts, patient where it matters.