THE AI INTERVIEW
“If AI-generated code causes a production outage or a security vulnerability, that is ultimately a failure of human governance”
Shayan Mohanty Chief Data & AI Officer Thoughtworks
However, Shayan stresses that classic engineering principles – clean code, modularity, testing, observability, peer review – arguably matter more than ever before.
“ Now we also need automated guardrails wrapped directly around AI coding systems,” he says.“ If the AI generates poor or insecure code, the system itself should immediately detect, reject and force correction before a human even gets involved.”
Shayan goes on to highlight that, while junior roles are not disappearing,“ the starting line has shifted dramatically”.
He states:“ AI can eliminate a huge amount of rote work almost instantly, which means juniors are progressing into reviewer and editor-style responsibilities much earlier in their careers.
“ What’ s interesting is that the biggest disruption may actually happen in the middle of organisations rather than at the bottom. Mid-level engineers and managers increasingly need to become systems architects and AI supervisors.”
AI accountability Asked where accountability sits when AI-generated code causes problems, Shayan is unequivocal:“ With humans, always, full stop.”
He continues:“ If AI-generated code causes a production outage or a security vulnerability, that is ultimately a failure of human governance and system design.” Blaming the technology itself is, in Shayan’ s view, a“ dangerous cop-out” since AI has no legal responsibility or contextual understanding of consequences.
“ The organisations getting this right treat AI as an untrusted participant inside a carefully controlled engineering process, not as an autonomous decision-maker,” he says.
“ Instead of just rolling out thousands of Copilot licences and hoping for the best, they invest heavily in internal developer platforms, governance layers and operational guardrails.”
Governing agent fleets Interestingly, Shayan contends that most organisations are still underestimating just how profoundly AI will change dayto-day operations over the coming years. He points to the emergence of“ permissionhungry agents” that are deeply integrated into company infrastructure such as Slack, databases and CRM tools.
26 August 2026