Recruiting Robot Co-Founders

The very first commit for this site wasn’t HTML—it was me whispering to a chatbot at 1 a.m., “Hey, can you help me remember how Astro routing works?” Welcome to my origin story of building with AI coding assistants, where caffeine is optional but clever prompting is not.

Chapter 0: Meeting the Cast

• Claude is that empathetic friend who will rewrite your paragraph and your self-esteem.
• GPT is a trivia champion with an encyclopedic memory of every framework, ever.
• GitHub Copilot is the roommate who finishes your sentences—occasionally in Latin.

Together they form the world’s most chaotic but brilliant pair programming squad. My job quickly shifted from “typing” to “conducting a choir of robot interns.”

Chapter 1: Planning With Prompts

Before touching the repo, I mapped the build in a shared doc titled “Okay, robots, hear me out.” The prompt outline looked like this:

Goal: Launch norseson.site/blog
Context: Astro theme, minimal styling tweaks, writing-heavy
Deliverables: hero copy, color tokens, first article

Turns out giving AIs context is like giving a toddler a juice box—you avoid 80% of the chaos. With that one snippet, GPT sketched the file tree, Claude suggested tone guidelines, and Copilot started auto-completing components like it had rent due.

Chapter 2: Building the Skeleton

I asked GPT to “explain Astro content collections like I’m speed-running,” and it replied with a plan that slotted perfectly into content/blogs/. Copilot took those hints and spat out TypeScript definitions before I could overthink imports. My role? Guardrails:

1. Confirm commands before pasting.
2. Interrogate anything that smelled like copy/paste from 2017.
3. Keep the tone guide taped next to the keyboard.

Result: a functional blog scaffold and an existential crisis about whether I still qualify as “hands-on.”

Chapter 3: Writing With a Laugh Track

Drafting this very article felt like co-hosting a late-night show. Claude tossed in quips (“please stop calling Copilot a golden retriever”), GPT fact-checked release dates, and I stitched it into something that sounds like a voice memo to a friend. The trick was alternating between:

• Chat-first for brainstorming lists, analogies, and section titles.
• Human-first for the connective tissue, because sarcasm lands better when it comes from someone who has actually stared down a failing build.

Chapter 4: Lessons Learned Before Coffee

What I Tried	Result
“Write the whole blog post for me.”	Word salad with zero personality. Hard pass.
“Here’s my outline, punch it up.”	Chef’s kiss; jokes plus clarity.
“Refactor this Astro component?”	Copilot + GPT tag team = fewer typos than my solo runs.

The golden rule: AI loves constraints. Give it intent, stakes, even a fake deadline (“pretend this ships in 2 hours”) and it suddenly acts like a senior engineer.

Chapter 5: Human Responsibilities (Still a Thing)

• Curate the vibe. Robots don’t know if neon gradients are “tastefully retro” or “early Myspace.”
• Own the decisions. AI can suggest three routing strategies; you pick the one you debug at 3 a.m.
• Document the journey so future-you remembers why slug and title don’t match (hi, that’s this post).

What’s Next

From here the plan is simple: keep the AI assistants on payroll, but make them earn it. Next up is automating component tests, experimenting with Claude for UX copy, and maybe training a mini model on my writing quirks so it stops suggesting semicolons in blog prose.

If you’re kicking off your own build, invite a few AI copilots early. Treat them like overeager interns—give direction, ask follow-up questions, and always run the final spellcheck yourself. The payoff is absurd: more momentum, fewer blank pages, and enough extra time to actually enjoy the launch espresso.