It Has To Be Human Auditable

We can't find the internet

Attempting to reconnect

Something went wrong!

Hang in there while we get back on track

I have been AI-pilled for over a year at this point. It's pathetic, I rarely touch-code any more. Love it or hate it agentic coding is here to stay. It's going to get more expensive and less impressive over time but there's not going to be some great apocolyptic moment where we all roll back to the before-times. Accepting that fate, one thing that really concerns me is handing the keys entirely over. Ultimately people need to be the last line of defense when it comes to the code that we put into production. If something goes wrong a person needs to be able to determine what and why. I'm not saying that agentic tools cannot be used for this, but I'm talking about where the buck should stop. If all else fails, if the AI isn't solving the problem, the code and infrastructure that the AI is building must be human auditable.

This almost feels like it should be a legal requirement built into the models. Some sort of regulatory self-imposed limit that we won't accept AI producing systems that are obfuscated against human intervention. For example, perhaps you can imagine a point where a AI-specific programming language or communication protocol or even just a compact syntax (i.e. brainfuck) ends up being a significant token saver. Should we accept the risk that a person could not audit that code for the sake of performance or cost reduction? I don't think we should.

Thankfully I don't believe this is an issue yet, but I can see things going in this direction. Especially when the labs start using self-reinforced learning. If the AI is training itself that feels like a situation where these optimizations can come into play.

Newsletter

Stay in the Know

Get the latest news and insights on Elixir, Phoenix, machine learning, product strategy, and more—delivered straight to<br>your inbox.

Your Email

Subscribe

DockYard Home

li]:inline-block">

DockYard on YouTube

DockYard on Twitter

DockYard on Bluesky

DockYard on LinkedIn

DockYard on Dribbble

DockYard on GitHub

Copyright 2026. DockYard Inc. All Rights Reserved

Terms of Service and Privacy Policy