I told you so

Overview

The video analyzes a viral post about a company one year into a “100% AI” coding mandate. The speaker says the post’s authenticity is uncertain. According to the post, the codebase has grown to roughly three times its former size, employees no longer understand how it works, and AI agents duplicate existing helpers because multiple versions of the same functionality exist.

After an initial period of faster feature delivery, management is reportedly asking developers to write code manually again. Some employees have lost practice and now take much longer to do so.

Main Points

More code and faster delivery do not necessarily mean more value

Producing more features may not increase sales, while customers may still be waiting for existing problems to be fixed.

Readable, simple code matters

The speaker describes working on a relatively large personal game project in C with Raylib. He values clear structure and straightforward implementations, and finds it satisfying to refine an initially simple solution. He prefers removing unnecessary code over adding more.

AI should not replace learning or programming ability

The speaker occasionally uses Grok as a targeted aid, like a search engine, but avoids coding agents and does not use AI when his goal is to learn. He argues that relying heavily on agents can erode developers’ skills and understanding—their “agency”—and that recovering those abilities takes substantial effort.

Managers may avoid responsibility for failed mandates

The speaker argues that managers who pushed AI adoption may not acknowledge mistakes. Employees and engineers could instead bear the costs through rework, restructuring, or job losses. He also suggests that weaker hiring and the departure of experienced contributors can worsen code quality over time.

AI adoption may bring future costs and dependency risks

Companies could have to rebuild or discard AI-generated work. The speaker also warns that companies may become dependent on AI tools and later face higher prices or restrictions, including models refusing to help with certain codebase tasks. Smaller firms may have less access to premium alternatives.

A middle path may be needed

The conclusion is not that companies should never use AI, but that they should avoid making all development dependent on it and should preserve developers’ skills and understanding. The speaker favors working independently or in smaller organizations where possible, while noting that this is not an anti-corporate argument.

Reviews, Guides, or Tutorials

None are presented. The video is commentary and analysis, with a personal example from the speaker’s game-development work.

Main Speaker and Source

The host identifies himself as Boomer Twig, from the channel Boomer Ranting in the Woods. He also mentions a Based on Compute episode with Lane from Boot.dev, where they discussed unintended consequences of censored AI models.