Jack Rudenko on why most AI rollouts are religion, not engineering

The episode

Jack Rudenko has not written a line of code himself in nine months. In that time his team shipped six products that are live and used by millions of people each week, and he rebuilt on his own, in two weeks, a system that had taken three engineers three months. He runs an AI-native engineering practice called Magus with 500 skills in it, and he has spent 25 years in engineering, including contributing to the Linux kernel.

He also thinks most of what the industry believes about AI engineering has never been measured. He splits the market into two camps: religion, where you add a tool, believe it works and keep going, and engineering, where you wait and you measure. He and James get into the project his team would have turned down before AI, why working with agents exhausts people rather than freeing them, why he treats every API key he owns as already compromised, and what still separates an engineer from an artist or a scientist when the code itself is no longer the hard part.

What we get into

• 01The bottleneck in AI engineering is human, not technical. Company processes were built around human unreliability, and Jack argues you can run 50 agents where you would have hired three people.
• 02Skills nobody measured: a major vendor rolled skills out across its own company and found agents loaded them roughly 40 percent of the time. The industry had used skills for eighteen months before anyone checked.
• 03The project his team called impossible: three disconnected systems, hundreds of people copying data between them, 500 edge cases. Without AI they would have declined it. They shipped in three months, and the second version took two weeks.
• 04From 10X to 50X. Jack was running a 10X approach before AI, six-week lockdowns delivering what teams of ten could not in eight months. He puts the change since at 50X, and explains why he cannot prove it cleanly.
• 05Why no project is secure once an API key exists, why he no longer keeps keys on the file system, and what changed when smarter models started finding vulnerabilities humans had missed for twenty years.
• 06Artist, scientist, engineer. Artists have no borders, scientists have no final product, and engineers work inside constraints and carry the responsibility when the thing breaks. AI changed what is possible. It did not change who is accountable.

Lines worth keeping

Tap to copy, then drop it into a post.

Why it matters

Australian teams are buying AI tools faster than they are measuring them, and Jack is one of the few people here who can show what the measured version looks like.