Tech Show Paris 2025: Day 2 DevOps Keynote Insights

From FinOps-native engineering to AI-augmented delivery, Day 2 of DevOps Live Paris revealed how Europe’s teams are reshaping software operations for a faster, fairer, and more intelligent future.

– – – – – –

FinOps as Strategy: Aligning Architecture, Cost, and Value

Opening the day, ‘FinOps and DevOps: Cost-driven development in a cloud-economic world’ brought together Benoît Sibaud (LinuxFR), Emilie-Brenda Naoussi (Air France) and Fleur-Anne Blain (ENGIE Digital) to decode the economic pressures shaping modern engineering.

Naoussi explained how Air France treats FinOps as a transversal engineering function, not an after-the-fact budgeting tool. By embedding cost governance across architecture design, migration planning, and run, her team delivers forecasts that business leaders can actually act on. Blain emphasised the need for opinionated templates: reusable patterns that enforce both security and economic efficiency.

Sibaud highlighted the cultural challenge, helping teams see FinOps as acceleration rather than restraint. The shared message was unmistakable: FinOps is no longer a reporting exercise but a platform strategy.

For platform leads, this marks a shift from cost-cutting to cost-aware design, where carbon, performance, and resilience become first-class engineering inputs.

AI in ITSM: The First Wave of Operational Automation

In ‘How to transform your IT department into a value-creating machine with Atlassian AI? ‘, Christopher Thirion (CBTW) demonstrated how AI is becoming the first responder in enterprise support environments.

Thirion showed how conversational agents now guide users through troubleshooting, auto-suggest knowledge articles, and escalate only when needed. On the operations side, AI drafts incident summaries, identifies documentation gaps, and classifies requests with higher accuracy than static rules.

For large enterprises, the takeaway was clear: ITSM is the lowest-risk, highest-impact entry point for enterprise AI. By 2026, L1 work will largely be agent-assisted, allowing SRE and DevOps teams to focus on deeper operational challenges.

DevAIops: Agents Across the Delivery Chain

The panel ‘From DevOps to DevAIops: what role for intelligent agents in the transformation?’ united Calange Favreau (Bpifrance), Jean de Laulanié (Spectrum Groupe), Hind Cagnon (Sodexo) and Stéphane Bout (McKinsey & Company) to explore how AI is evolving from assistant to collaborator.

Teams described real deployments across the SDLC:

• Developer copilots integrated into secure environments

• Automated dependency upgrades

• AI-driven compliance and vulnerability assessments

• Incident summarisation and pattern-matching

• Architecture assistants proposing cloud designs aligned to internal standards

Bout noted that the largest productivity gains occur not at the IDE, but when AI is chained across planning, coding, testing, deployment, and incident response. Favreau shared how Bpifrance uses AI to accelerate post-incident reviews and reduce toil in run tasks.

The panel converged on a single idea: Engineers will increasingly orchestrate agents, not just services. 2026 will belong to teams that build evaluation frameworks, API guardrails, and LLM gateways to scale these agents safely.

Deep Agents & MCP: Architecting for Distributed Intelligence

In ‘Deep Agents MCP: The New Era of Intelligent Search’, Samir Akarioh (Scalingo) and Chrys Fé Marty NiongoLo (Eviden) presented a live demonstration of multi-agent systems built on MCP (Model Context Protocol).

Their setup included:

• A Ten-Agent orchestrator to plan and route tasks

• MCP servers for search, indexing, and metrics (SerpAPI, Algolia, OpenSearch)

• Mistral Chat for external control and execution

Agents worked in parallel, gathering information from multiple sources, indexing it, and producing structured analytical reports — complete with performance metrics.

For senior platform engineers, the implication was clear: DevOps platforms will soon need “agent-ready” APIs, allowing LLM agents to interact safely with observability, CI/CD, configuration, and documentation systems.

Diversity as Delivery: Expanding the DevOps Talent Pipeline

In ‘The pipeline of the problem: why diversity remains stuck in DevOps’, Mariatou Traore (Aneo), Lamiae Bernoussi (We4She), Dipty Chander (E-mma) and Gabriela Belaid  (SecNumCloud, CentraleSupélec & Olympe de Gouges) examined why DevOps roles remain among the least gender-balanced in technology.

Bernoussi emphasised that recruitment pipelines frequently filter out qualified women long before interview stages, often due to rigid role specifications or self-censorship. Traore discussed how structured hiring scorecards reduce bias, while Chander highlighted the need to recognise alternance and apprenticeship as real experience.

Belaid underscored the operational impact: diverse teams produce stronger systems, a wider perspective, and improved debugging and decision-making. The collective message: diversity is not an HR metric, it is a performance multiplier.

Invisible Inefficiencies: The Real Threat to Delivery

Next, ‘The hidden inefficiencies that are killing your delivery’ saw Matthieu Sénéchal (hones) break down the silent blockers that slow down organisations more than any tooling choice.

He described the patterns familiar to every engineering leader:

• Too much work-in-progress

• Scattered ownership

• Over-engineering

• “Zombie projects” contaminating roadmaps

• Meeting overload without outcome clarity

Sénéchal argued that delivery inefficiency is rarely a technical failure but rather a flow failure. His prescriptions were simple but powerful: limit WIP, enforce explicit ownership, align on a North Star metric, and eliminate work that does not advance measurable value.

LLMOps in Practice: Evaluating Conversational Agents at Scale

In ‘Implementation of the LLMOPS platform for the development of conversational agents in the Beauty Tech sector’, Ismail El Maarouf (L’Oréal) outlined how Beauty Tech teams industrialise conversational AI across diverse product lines and markets.

El Maarouf explained that evaluation is the hardest part of deploying LLM-powered agents: product information changes constantly, ethical constraints must be enforced, and models behave non-deterministically. To manage this, his team built a platform where datasets, prompts, models, and evaluation logic are versioned and executed automatically within CI/CD. Every pull request triggers tests against curated datasets, blocking releases if thresholds are not met.

To keep the process efficient, L’Oréal runs evaluations in parallel, caches LLM responses during tests, and separates development, QA and production datasets for controlled rollouts. The result: rapid iteration without sacrificing governance.

For engineering leaders, the lesson was clear: LLMOps is not just model deployment, it is continuous, automated verification of behaviour as systems evolve.

Key Takeaways for 2026 and Beyond

• FinOps = Platform Design: Cost, carbon, and performance must be engineered into standards and golden paths.

• AI as Operator: AI-driven deflection, triage, and summarisation are becoming baseline capabilities.

• From DevOps to DevAIops: Delivery teams will orchestrate agents across code, test, deploy, and run.

• Build Agent-Ready Platforms: Expose safe interfaces to observability, CI/CD, knowledge bases, and governance systems.

• Diversity Drives Reliability: Broaden talent pipelines to strengthen system thinking and engineering resilience.

• Fix Flow, Not Tools: WIP limits, ownership clarity, and meaningful metrics beat tooling sprawl.

Final Insight

Day 2 of the DevOps Live Paris revealed a discipline in transformation. As AI reshapes operations and FinOps becomes architectural, Europe’s engineering leaders are moving toward an era where intelligence, reliability, and human-centred delivery converge. Heading into 2026, the priority is clear: engineer for flow, design for value, and build platforms that think with you, not just for you.