Didier Girard

Didier Girard is CTO and Managing Director of SFEIR, a French ESN of 1,000 people. On BFM Business on 5 May 2026 he put his position plainly: «écrire du code est devenu un anti-pattern». AI already produces better code than most engineers, he argued, while the decision stays human.

One method runs through his writing: accept the symptom, refuse the number. He contests «la méthode de l'étude MIT (95 % des pilotes sans ROI)» and keeps its signal, «le symptôme est vrai même quand le chiffre est faux». Reading Mistral's Shieldstral 1.0 3B on 7 August 2026, he validated the architecture (the policy is read at inference rather than baked into the weights), then refused the benchmark scores: constructor figures, constructor test sets, no third-party evaluation as of 6 August 2026.

«AI4IT vs AI4Business : le renversement (article SFEIR)», published 24 June 2026, inverts the 2024 consensus. Measurable value comes from AI applied to the information system, because code verifies itself through compilation, tests and CI while a business process does not. His 2027 budget advice follows: fund AI4IT first, steer on cost per outcome rather than token price, count humans and agents together.

What stays open is governance. Searching «ACP» in his own watch base returned twelve results, all on the commerce protocol, none on Zed's client-to-agent one, hence the rule he adopted: never index an acronym alone. With Shieldstral, responsibility lands on the deployer while auditability, robustness and availability are missing.