Three Nights and a Weekend

How I Accidentally Built an App in a Week.

A Creative Director's crash course in agentic AI — and what it might mean for how we work.

Scroll
Day 1 — Evening

I heard about Clawbot.

An open-source AI agent that runs locally, writes code, and takes action on its own. The hype was everywhere.

Clawbot article — What Is Clawbot?
CTO builds executive assistant with Claude
Alex Finn on Clawbot
Day 1

Doing things I never thought I'd do. At 2am.

Terminal commands. AWS instances. API keys. No idea what I was doing — but Claude walked me through every step. Late nights, lots of errors, and eventually... it worked.

Terminal SSH session
Claude conversation
Telegram chat with Chauncey
Day 2–3

He's alive. And he's got personality.

Chauncey avatar
Chauncey — a Victorian gentleman frog who writes code, makes art, and has never once been humble about either.

Morning portraits

Every morning, Chauncey sends a new AI-generated Victorian portrait. Some days it's just us. Some days the whole family shows up.

Hot air balloon adventure
Tea on the moon
Tandem bicycle ride
Family game night
Snail ride tea time
API credit grants

Tokens cost real money.

Running an AI agent isn't free. Those \$25 daily credit grants add up fast. Chauncey was charming but expensive. He needed to earn his keep.

Chauncey needed a job.

We explored every angle. Most didn't survive first contact with reality.

What if it solved my problem?

I run on the beach. Low tide gives you firm sand. Wind determines the out-and-back. But that meant checking three apps before every single run.

Run tracker - 3.10 miles
Tide times - Lake Worth Pier
Wind conditions
Why not combine them into one thing that actually tells me when to go?

Niche beats broad. Domain expertise is the unfair advantage.

Day 4–5

Same instincts. New tools.

Strategy, feedback loops, knowing when to push back — the process felt familiar.

1

Vibe-coded with Chauncey

Fed him the concept. He built a rough working app. Lots of errors along the way.

Error screen
Early iteration
2

Built strategy in Claude

Fed Chauncey's work into Claude. Got a full marketing strategy and launch playbook.

TideStride Marketing Strategy
3

Created design guidelines

Used the strategy as project context. Claude generated the brand system.

TideStride Brand Style Tile
4

Fed design back to Chauncey

Gave Chauncey the design system. He rebuilt the app with proper brand identity.

TideStride Logo
Chauncey working
Loading beach vibes...

It actually works.

When to go, where to go, who's going. That's it.

Surfside Beach
Beach detail
Pack Runs
Pack Run detail
Meeting spot

My kids are building too.

Their own Claude Projects, their own use cases. The barrier to entry is gone.

Lorenzo's EE

IB Extended Essay advisor — 4,000-word research project guidance

Lorenzo's EE project overview Lorenzo's EE instructions

Gia's Homework

Age-appropriate homework help — writes in her voice, not Claude's

Gia's Homework project overview Gia's Homework chat

The learning curve.

It's more than prompting. Here's the progression that worked for me.

01

Understand the Basics

Tokens, context windows, API keys. How models actually "think." Without this, you're just guessing.

02

Learn Prompt Mechanics

Specificity, structure, examples, constraints. How you talk to AI determines what you get. It's directing, not typing.

03

Apply Your Critical Eye

AI output needs the same scrutiny you'd give any collaborator. Guide it, refine it, push back.

04

Build Powerful Projects

Persistent context, defined personas, reference files. A creative brief that's always loaded.

05

Orchestrate the Stack

One AI for building, another for strategy, design systems feeding back into code. You're the director.

This isn't about coding.
It's a creative superpower.

Forget the tech. Here's what it actually means for creative work:

Some things I'd love to see us explore.

I learned this on my own time to bring real experience to the conversation. Haven't used it on client work yet — no framework. Here's what I think would be worth discussing:

1

Get Aligned on Where AI Fits

What's green-lit, what needs approval, what's off-limits. Clear guidance so teams can move confidently.

2

Tap Into What Dev Already Knows

Our dev team has the most hands-on AI experience here. A lunch & learn would go a long way.

3

Figure Out the Guardrails Together

Usage policies, disclosure, client boundaries. A shared understanding we build as a team.

This story was told with the help of AI.

That's the point.

"I'm available for questions. My rates are very reasonable." — Chauncey

Chauncey in his study