How I Code

For better or worse, I’ve been tinkering on products for years. Some come to life, most fail, but always rattle in my head.

Cliche to say but AI has allowed much of what I wanted to build become something I could build. Interestingly, my son has been able to do the same thing with his 3D printer (especially with tools like Meshy to be able to simply “chat out” what you want to make).

This guide is my ever-evolving resource to how I code in the age of AI tools. The reality is that there are many ways to code (just go to YouTube for proof) - this just happens to be my organic, iterated and iterative way.

The Minimally Viable Tools

Beyond the tools above, products have their own special idiosyncratic requirements necessitating some other vendor. But the tools above are the essentials in my mind.

Building Hacky First Drafts

A hacky first draft could be a single page website or a full blown web application with login and all.

Whatever it is, below is a step-by-step for how to get from an idea to booted-up thing.

Create a notion page for “thing” (I call it a “thing” because it could be a project, product, contraption or a whole fleshed out business idea with a hypothesized business model)

Write a “business idea/initial product description” sub-page with some inkling of how the thing would work and how it could potentially make money

Run the business idea aspect through Grok to get their not-so-humble opinion (which is good because I shouldn’t be wasting time on shitty ideas no matter how fast they can be built)

Polish the initial product description aspect through Grok to turn into a more functional document

Create a GitHub project on github.com and then clone the repository on my machine which kicks off a fresh new folder

Open the folder in Visual Studio Code and then save it as a workspace

Install Cline into Visual Studio Code (if I don’t have it already) - configure it with an Anthropic API Key as well as some custom instructions

Custom instructions templates

Configure Cline with several MCP servers so that Cline can access resources on demand it otherwise wouldn’t be able to (GitHub MCP, Render MCP, Neon MCP, PostgreSQL MCP)

Create a .clinerules folder inside the project main directory and drop a few key files in (initial-product-specifications.md, development-resources.md, stack.md)

Create a devnotes folder and put in tasks-history.md, general-deployment-and-application-instructions.md (these will automatically be populated through the dev process to keep track of changes)

Populate the files in .clinerules with the appropriate material (put the initial product specs in, if there are 3rd party APIs that may be needed, drop into dev resources, put tech stack in stack.md (postgresql via neon, node.js for backend with express, postmark for email, etc etc);

Open Cline in Plan mode and feed in the initial product product specs and then “work with” Cline to flesh out the dev plan. Because we’ve seeded things with the .clinerules directory, this keeps things on-rail from the start and minimizes the fielding questions Clines tends to ask.

Move to Act mode in Cline once things look good-to-go in Plan mode. By default Cline will “step through” most of the actions it takes (which can be valuable to review it’s changes before it goes too deep) but at some point you can turn on cruise-control by auto-approving some actions.

The first build typically requires some further setup (database connection strings, moving environment files around, etc)

First build live!

MCPs

MCPs are cool because they add next-level automation to projects.

I can instruct Cline to commit the code to the GitHub repository which could kick off a deployment process to Render. Or I can turn off auto-deploy in Render and I can then instruct Cline to commit the code to GitHub, run some tests and only if the tests pass, then deploy the service to Render.

From a database perspective, by “connecting” Cline to the database service on Neon, AI has “eyes” to the database structure/schema which is important as tables need to be modified as the app evolves and to debug strange bugs that get introduced as the application gets more complex.

Other tools

Illustration.app - good for AI-generated vector art (good for infographics, website graphics, etc)

Sequel.sh - good for “ask your database anything” type queries - ideally this would be baked into neon.sh (but likely could be done via an MCP)

v0.dev - good for quick landing page/app prototyping before going full-blown Cline/VS Code dev

Retool - good for reporting/admin dashboard stuff (need to research how to connect it quickly to a db in a way that I’m not wasting time building UIs)

Skyvern - good for automated browser testing especially with Retool workflows

Visualping - good for uptime monitoring from a top-down approach

n8n.io - move over Zapier

VAPI - good for voice agent stuff (mash together text-to-speech, speech-to-text with LLM brain between the two)