<- Back to links

// Video / Founder story / AI workflows

From CEO to vibe coder

I learned AI-assisted software development backwards: algorithmic trading first, simple apps later.

10-minute story Post-exit build Engineering, not trading advice
Then

Two remote teams and roughly $70,000.

Start

A duct-tape proof of concept.

Now

AI compresses the build cycle.

Bottleneck

Live evidence, reliability, and time.

The reverse journey

Most people begin vibe coding with a budget app, a fitness tracker, or another small project. I started with an algorithmic trading system before the models were ready.

The earlier versions of the idea had already consumed two distributed teams, months of coordination, and about $70,000. Rebuilding it with AI meant starting at maximum difficulty.

Weak models forced discipline

  • Break the system into microscopic tasks.
  • Test continuously instead of trusting a plausible answer.
  • Reconstruct failures until the system is understandable again.
  • Keep the architecture small enough to reason about.

The bottleneck moved

As the models improved, the workflow accelerated. Features that once required prolonged coordination could be built and tested quickly. Eventually the coder stopped being the limiting factor.

The new constraint is reality: canary tests, odd market conditions, reliability, security, and enough elapsed time to know whether a system actually works.

The practical lesson

AI can compress the build. It cannot compress evidence.

This is an engineering and learning story. It is not a claim of trading performance or financial advice.