NEXTHINK
“ For coding, it ' s straightforward,” he says.“ But software is more than coding. There is ideation, testing, deployment, maintenance. How can we automate those tasks? We have people looking at all of those things and asking how they can leverage AI and agentic systems to automate those tasks and make their lives easier.”
The shift from traditional engineering to AI-first brings significant challenges, however. Moe frames the core issue as a fundamental change in the nature of software itself.
“ When you are doing normal software development, it ' s a deterministic model,” he explains.“ If‘ A’, you’ ll have‘ B’.
Validating the outcome is easy and the processes around that are mature.
“ With AI, you move to a probabilistic model: if it ' s‘ A’, it ' s probably‘ B’. You have to put a lot of new disciplines in place to validate and verify that the product is doing what it needs to do.”
Those new disciplines include evaluations, benchmarks, guardrails and more iterative experimentations. Teams must also hire for different profiles or upskill existing staff so they understand probabilistic systems, and can design experiments and interpret model behaviour. It is, as Moe puts it, a substantial transformation of mindset, process and culture.
8 May 2026