By Gologic with the collaboration of Alexandre Couëdelo.
Everyone has to agree on one thing: 2025 marks the advent of AI. Gone are the days of AI demos that barely work in everyday tasks. Gone are the days of AI tools that just glorified autocomplete. We have clearly moved beyond the simple “wow effect.” In 12 months, the progress has been staggering: a rapid succession of increasingly powerful development tools and ingenious developers exploring this technology from every angle—context management, diagnostic tools, integrations with all development tools, etc.
At Gologic, we have been following this phenomenon very very very closely to redefine what DevOps will become in the age of artificial intelligence. Here are our findings and our roadmap for 2026.
The Pitfalls of AI in 2025
The adoption of AI is massive: 90% of IT teams use it every day, with an average daily interaction of over 2 hours.
We have observed up to “an estimated 30 minutes of time saved per day per developer with the use of Github Copilot“, but writing code is only a fraction of the total effort required to deliver a reliable, production-ready feature.
However, the AI-powered software supply chain is currently facing two major paradoxes.
A Productivity Paradox
Numerous studies, including those from DORA, Faros, and Atlassian, converge and reveal that individuals’ productivity gain does not translate into faster software delivery:
- Explosion in review workload: Pull Request (PR) review times have increased by 91% for teams that are heavy AI users. [3]
- Code verbosity: The average size of PRs jumped by 154%. AI generates verbose blocks of logic that don’t reuse existing functions, violating the DRY (Don’t Repeat Yourself) principle. [3]
This result was predictable: AI can lighten the load of implementation tasks and certain mechanical aspects of code writing. But the critical phases, planning, architectural design, validation, still require considerable time. These essential phases are performed at the pace of human cognition.
“That 80% of the work (planning, architectural design, validation) is the hardest part, and that’s where you really need the developer… Once you have that, you can say, ‘Do the same thing for this case.'” — Justine Gehring, Head of AI Business Line. Gologic Podcast
The conclusion is simple: there’s no point in coding at top speed if you don’t also focus on the bottleneck, both upstream and downstream of the code, in the delivery chain.
A Quality Paradox
Increased code generation leads to a second paradox that affects software quality. More code written quickly leads to increased complexity in software solutions, with more lines of code to maintain, of course, but even more seriously, which often comes at the expense of reliability, for example:
- Increased security risks: research reveals that half of the code generated by AI contains security flaws, introducing known vulnerabilities. AI tools are trained on code examples that are neither verified nor necessarily up to date. [5]
- “Almost right” code: 45% of developers responding to Stack Overflow’s annual survey cite “almost right, but not quite” solutions as a major frustration, leading them to fix bugs that can be difficult to detect in generated code. [7]
Added to this is the fatigue associated with code review: analyzing the number of lines of code is not easy. In addition, developers become less critical when it is easier to ask AI to redo something than to do it correctly from the outset. The result: as many, if not more, bugs in production related to the use of AI.
“Misuse will lead to technical debt. You know, it’s like giving someone a Ferrari; they can go faster, but the consequences could be much more difficult compared to a little go-kart.” — Spencer Handfield, AI DevOps Specialist and Trainer. Gologic Podcast
The conclusion is clear: quality has many facets, and validating a software product requires rigour. Without new capabilities to combine speed and rigour, the Ferrari will crash into a wall.
Our Vision for 2026: Towards Industrialized and Agent-Based AI
Faced with these challenges and paradoxes of AI, blind faith in tools is not a strategic solution. The answer lies in process engineering. The challenge is to move from AI tools to complete systems integrated into the software delivery chain. This takes shape in two areas of development:
- The Plateform Engineering and Developer Experience (DevEx): To navigate tomorrow’s AI ecosystem, we need a platform that acts as a layer for distributing AI tools and governance. It offers “paved paths” that make AI safe and easy to use for developers. The watchword in generative AI is “control before you trust.” Hence, the need for a robust internal platform. This involves concrete actions such as:
- An agent hosting platform and a centralized agent library.
- Better MCP (Model Context Protocol) governance to secure the connection between AI and your internal tools (Jira, logs).
- Standardization of AI assistant use via a corporate prompt registry.
- Optimization through Value Stream Management (VSM): VSM is a strategic tool that identifies levers and bottlenecks in a process. VSM is used to measure the success of concrete projects such as:
- AI agents for migrating from one version to another to eliminate legacy code.
- Agentic contributions to documentation and unit testing.
- Validation agents that ensure security standards and CI/CD standardization.
In 2026, the goal is no longer to move quickly for fear of missing out on the AI revolution, but to build sustainable and controlled velocity.
By Gologic with the collaboration of Alexandre Couëdelo.
