During the California Gold Rush, hundreds of thousands of people showed up chasing gold. Most of them left with less money than they arrived with.

The people who actually got rich weren’t panning in the river. They were named Levi Strauss and Sam Brannan, and they were selling pickaxes, tents, and durable pants to the guys who were panning in the river.

Brannan bought every shovel in San Francisco before he told anyone gold had been found, then sold them at a massive markup once the rush started.

Strauss just made pants that didn’t fall apart in a river.

Neither of them needed to find an ounce of gold. They just needed everyone else to keep digging.

I keep thinking about that story because it’s basically what’s happening right now with AI.

Everyone’s Panning for Gold

Every founder I know is deep in the river right now - myself included. A new AI model, new agentic framework, new AI coding tools, and AI assisted legal documents. We’re all convinced the next nugget is one prompt away.

But the tools that have actually made the biggest difference in how I build over the last six months aren’t AI by themselves.

They’re PostHog and Supabase. Neither one is AI by itself. They’re the shovels, tools, and the pants that were sold in the California gold rush. They’re marketing or infrastructure.

And that’s exactly why they matter more than most of the AI tools I’ve tried this year.

Supabase: The Backend I Stopped Thinking About

Supabase is Postgres with the boring parts already solved. Auth, storage, a real relational database, an API layer, all set up in the time it used to take me to configure a single WordPress plugin.

I’ve spent 20 years around WordPress, but I’ve spent just as long building web software before anyone called it SaaS. Back then we called it Web 2.0, which tells you exactly how long ago that was.

I remember hand coding user sign up forms from scratch. Writing my own authentication from scratch. Building spam protection from scratch, back before reCAPTCHA existed, which meant rolling your own janky logic and still getting flooded with bot signups anyway.

None of that was hard exactly. It was just tedious, and it was the same tedious every single time, on every single project. You’d rebuild the same login flow for the tenth time and think, someone should really solve this once so the rest of us stop solving it badly, over and over, forever.

I know what “boring parts already solved” is worth, because I’ve spent two decades building the boring parts myself.

With both PressMeGPT and InstantLadyBirdDeed, Supabase is the layer underneath everything the AI is doing. When someone generates a theme, iterates on it, exports it, that all has to be stored, versioned, and tied to a real user account somewhere. AI models are great at generating content. They are not a database. They forget everything the moment the context window closes.

Supabase is the critical layer every application needs whether you notice it or not. Database, authentication, security, email. The stuff that doesn’t show up in a demo but breaks your entire product the moment it’s missing.

The AI features get the demo. Supabase is why the demo still works the next morning.

It has become a critical layer of getting a SaaS application or website built quickly, while also allowing for some scalability.

PostHog: Watching What People Actually Do

PostHog is product analytics, session replay, and feature flags in one place, and it’s the tool that finally answered a question I used to guess at for years: what are people actually doing on the site, not what do I assume they’re doing.

It also showed up at exactly the right moment. Right around when I found PostHog, I realized Google had quietly turned Google Analytics 4 into a tool that can’t be trusted for anything specific anymore — especially once you factor in how much AI traffic it’s misreporting entirely. I’d been leaning on GA4 out of habit more than anything. PostHog wasn’t just a nicer dashboard. It was a replacement for a tool that had stopped doing its job right when I needed the data most.

At my previous company, we’d ship a redesign based on gut instinct and client opinions, then never really know if would work for sure. Sometimes we’d get it right, but often times we’d get it wrong.

You’d get a vague sense from support tickets and feature requests and hope the bounce rate looked better in Analytics next month.

PostHog replaced the guessing with a replay. I can watch a real session of someone using PressMeGPT, see exactly where they paused, where they got confused, where they gave up and left. That’s not a metric. That’s a person, in real time, hitting a wall I put in front of them without meaning to.

PostHog also has AI built into the product itself now, and it’s actually useful. It can look at your data in aggregate and surface patterns or recommendations you’d never think to go dig for on your own.

That combination is the powerful part. Useful metrics plus an AI layer that can reason over them, the same way you’d hand a stack of numbers to ChatGPT or Claude and ask what it notices. PostHog basically built that in.

Neither One Is AI by Itself

Both of these products have added AI features. PostHog has AI-assisted insights. Supabase has vector support and plays nicely with AI-generated backends. That’s not really the point though.

The point is that they utilize AI to solve a practical problem in an elegant way. AI tools are mostly about generation, making something out of nothing faster than before.

But if you rely on AI the first time without pushing back, it is just as good at making mistakes or giving ill advice. I’ve discovered this first hand.

PostHog and Supabase are about everything that happens after the thing exists. Where does it live? Who’s using it? Is it actually working?

You can have the best AI website builder, the best AI coding agent, the best AI writing tool in the world, and none of it matters if you don’t know whether real people are getting value from what it made, and if you have nowhere durable to store the result.

That’s the gap the shovel sellers filled during the gold rush too. Finding gold was flashy. Surviving long enough in the field to actually dig it up required the boring stuff nobody bragged about at the saloon.

It is incredible on how quickly we can iterate products because of these two tools.

The Boring Stuff Is the Business

Six months ago I would’ve told you my job was building an AI-Powered WordPress builder. Now I’d tell you my job is helping developers save massive amounts of time with AI, and the tools that made that possible weren’t the AI chats at all.

Every founder chasing the next model release should ask themselves a version of Brannan’s question. Are you panning for gold, or are you the one selling the shovels?

The more durable business may be the one solving the unglamorous problem everyone else is stepping over on their way to the river.