Software Engineering in the AI Era: Operating Model Shift
Op-ed by Olivier Rafal (Consulting Director Strategy, WeNvision — SFEIR group; former editor-in-chief of Le Monde Informatique) published on June 1, 2026 in CIO-Online, structured around a paradox: in the AI era, software engineering changes everything… and nothing changes. What changes = the operating model. Roles are redefined: the Product Owner shifts from backlog breakdown to generating context usable by AI; the developer shifts from writing code to framing, directing, and reviewing agent execution; QA gains the ability to define expected proof upfront.
By **Olivier Rafal**// Source cio-online.com ↗/Reading 2 min/.md// Auto-verified translation
In this op-ed from June 1, 2026 in CIO-Online, Olivier Rafal (Consulting Director Strategy at WeNvision, SFEIR group, former editor-in-chief of Le Monde Informatique) argues a paradox: in the AI era, software engineering changes everything… and nothing changes.
What changes is the operating model. Roles are being redefined: the Product Owner shifts from backlog breakdown to generating context usable by AI; the developer moves away from writing code to framing, directing, and reviewing agent execution; QA can define expected proof upfront. Structures are evolving from "double pizza teams" (eight-person hand-off chains) to "sandwich teams": a tight pairing of a business expert and a tech lead, both AI-augmented, with other skills in support. At Sfeir, "this pairing now drives roughly 80% of the production chain", with the remaining 20% (architecture, data governance, security) centralized. The formula sums it up: "the issue is not a tooling issue, but an operating-model issue".
What doesn't change is the discipline of the cycle. The SDLC phases — define, build, verify, deploy, maintain — remain identical; AI abolishes none of them, it intensifies them. "All the slack that human-paced work absorbed, one way or another, becomes, at AI speed, industrial-grade defects" — like sport, amateur or professional. Hence three inviolable gates (specification, planning, delivery review), validation by proof, and systematic capitalization that cuts correction iterations by 30% after about ten cycles. Guiding principle: "the faster the execution, the stricter the framework must be". The harness frames the agents; vibe-coding is deemed untenable in the enterprise.
Third pillar: governance, FinOps, and value-driven steering. AI costs are variable and recurring (~€10/hour per augmented seat); as with cloud in the 2010s, the shift is from flat-rate to usage-based. FinOps does not aim to cut costs but to "optimize tool efficiency", weighing cost against value (time-to-market, features, performance, eco-design).
Conclusion: acceleration makes the fundamentals non-negotiable; the challenge is first and foremost organizational and cultural, not technological. Without a healthy business relationship or collective discipline, an AI-boosted SDLC merely amplifies problems.
Key takeaways
Date / source.June 1, 2026, CIO-Online (Op-ed). Author: Olivier Rafal, Consulting Director Strategy WeNvision (SFEIR group), former editor-in-chief of Le Monde Informatique.
Paradox thesis."everything changes… and nothing changes" — the operating model reinvents itself, the SDLC cycle stays invariant but becomes non-negotiable. ### What CHANGES (operating model)
Roles. PO → generates context usable by AI; dev → frames/directs/reviews agent execution; QA → defines expected proof upfront.
Teams.double pizza teams (≈8, hand-offs) → sandwich teams (pairing of business expert + tech lead, AI-augmented, rest in support).
Sfeir. this pairing drives "roughly 80% of the production chain"; ~20% (architecture, data governance, security) centralized.
Quote: "the issue is not a tooling issue, but an operating-model issue." ### What DOES NOT CHANGE (discipline of the cycle)
Validation by proof. (not AI's own assertions) + capitalization → −30% correction iterations after ~10 cycles.
*"The faster the execution, the stricter the framework must be.". * Concepts: harness, vibe-coding untenable in the enterprise. ### 3rd pillar: governance / FinOps / value
AI costs variable & recurring ≈ €10/hour per augmented seat; flat-rate → usage-based (2010s cloud parallel).
FinOps ≠ cost-cutting."optimize tool efficiency" (cost/value), not just token consumption.
Align business metrics upfront: time-to-market, features, performance, eco-design. ### To draw on for engagements / presentations
Ready-made transformation framework. (diptych + 3 pillars + 3 gates) — particularly relevant on the SFEIR/consulting side (internal proof-points).
Connects: systems around the model (Dropbox/Okumura), harness engineering (Osmani), agentic Salesforce, token FinOps (his own Tokenomics Foundation op-ed), agent manager doctrine (BFM/Girard).
Reusable formula: "the faster the execution, the stricter the framework must be" = antidote to vibe-coding and to "hitting the wall faster."
Key figures
the business + tech lead pair drives ~80% of the production chain at Sfeir
the challenge is the operating model, not the tool
— Olivier Rafal
AI does not eliminate any SDLC phase, it intensifies them
— Olivier Rafal
"the faster the execution, the stricter the framework must be"
— Olivier Rafal
"all the slack that the human pace absorbed, more or less, becomes, at AI speed, industrial defects"
— Olivier Rafal
vibe coding is untenable in the enterprise
— Olivier Rafal
The knowledge graph extracted from this fiche — 13 entities, 15 relations.
In this graph :Olivier Rafal · WeNvision · SFEIR · L'ingénierie logicielle à l'ère de l'IA : tout change... et rien ne change · modèle opérationnel agentique · sandwich teams · double pizza teams · SDLC · les 3 gates · capitalisation · harnais · FinOps appliqué à l'IA · vibe-coding