18/08/2026
🤔 Code is getting cheaper to produce. Bad decisions are getting more expensive.
Coding agents can already analyze repositories, modify multiple files, run tests, fix errors, and prepare code for review.
That shifts the bottleneck.
As implementation gets faster, more weight moves to:
▪️defining the right problem
▪️making sound architectural decisions
▪️giving agents the right context
▪️testing and validating the output
▪️knowing what is actually good enough to ship
AI can speed up implementation, but ut it can also help teams implement a bad specification or poor architecture faster.
Faster implementation doesn’t remove engineering constraints. It moves them upstream: into decisions, architecture and validation.
We see this in client projects at Directio as well. The biggest gains come when AI is introduced into a well-designed engineering process, with clear requirements, the right architecture and strong validation.
That’s where faster coding actually translates into faster delivery.