7 1/2, 11, 4.5, 2.5 = Why Context Matters
By Hal Henningsmoen, hal@sage-tech.ai
With AI coding assistants context, especially accurate domain knowledge, is vital. In his Substack post “Spinal Tap Meets John Malkovich: What’s So Funny About the Need for World Models?”, cognitive scientist Gary Marcus discusses the perils of getting the world model wrong in AI-generated content. He starts with the movie world where it’s fine for a character to work on floor 7 1/2, and we all love Spinal Tap’s amplifiers that go to 11. The writers chose these silly inaccuracies purposefully to produce an effect.
In AI-generated content, however, inaccuracies can spell the difference between success and failure. Marcus looks at AI-generated images of Derek Smalls (of Spinal Tap fame) riding a unicycle on the beach. In the first image, the unicycle lacks spokes and pedals. A second attempt — created by a Grok user and shared in Marcus’s article — looks better at first glance.
But a closer look at the bass in these images, cropped for review from the original displayed in Marcus’ post, reveals 4.5 strings and 2.5 tuning pegs.
Still a pretty impressive image and maybe these, and other inaccuracies, don’t matter much.
These quirks may not matter in a novelty image. But in code, even tiny inaccuracies can have serious consequences. The accumulation of such errors can quickly lead to an unmanageable, unreliable codebase that is expensive to maintain and risky to deploy. To turn a phrase: “There is a fine line between clever and… disastrous.”
A carefully derived and curated world model for the software system is a critical part of the context for AI coding assistants and a key element of context engineering. Context engineering makes the difference between a successful and unsuccessful transition to AI-assisted coding — between cost savings and ballooning costs.
In the end, context engineering isn’t just technical hygiene — it’s what separates innovation from chaos when adopting AI coding assistants.
“But hey,” as Marty Di Bergi said in Spinal Tap, “Enough of my yakkin’! Whaddaya say? Let’s boogie!”