

Bias by Design
Smart engineers make predictably irrational decisions. Every day. Not because they lack skill or information, but because the human brain runs on two systems: a fast, intuitive autopilot (System 1) and a slow, deliberate override (System 2), and System 1 wins most of the time.
This talk is a tour through 14 cognitive biases that quietly shape software development: the Overconfidence Bias that turns 2-day estimates into 3-week sagas, the Anchoring Bias that locks sprint planning to the first number on the table, the Sunk Cost Fallacy that keeps legacy monoliths alive well past their expiry date, and the Confirmation Bias that turns code reviews into self-validation exercises.
For each bias, we'll look at how it manifests in the code, the team, and the product, and then at something less obvious: how that same bias is baked into AI systems. LLMs that hallucinate with confidence. RAG pipelines anchored to the first retrieved document. RLHF feedback loops that build echo chambers at scale. AI doesn't eliminate human bias. It amplifies it.
The talk closes with structural countermeasures, not platitudes, but concrete practices. Awareness is not the solution. But it is the prerequisite.
"Le persone perfette non esistono". But aware people make better mistakes.