Let's be direct: in eighteen months, the question won't be «Have you experimented with AI in your IT projects?» but rather, «By how much does your delivery capacity exceed that of your competitors thanks to agentics?»
This is not foresight, but what we observe on the ground, project by project, since we industrialized our own enhanced development practices at Silamir.
The real problem with IT projects hasn't changed in twenty years.
Delays in schedules, budget overruns – the well-known over-ground projects that no one has truly resolved. And 7 out of 10 projects that do not reach their initial objectives (Standish Group data, 2024).
Yet, this is not a problem of talent, as the teams are competent. It's a structural problem: unpredictable productivity from one developer to another, documentation lagging behind code, quality and security rules applied inconsistently, turnover of tech profiles oscillating between 13 and 21 % depending on the sectors…
Agency is not a magic bullet for all of this. But it's the first approach we've seen capable of simultaneously tackling these chronic irritants and ongoing projects.
What agent-based systems concretely change in the execution of an IT project
When we talk about agentics for IT, we're not talking about Copilot open in an IDE, but rather an agentic chain orchestrated across the entire project lifecycle, from framing to production.
Specifically, what we have implemented and observed:
- The inter-phase latency disappears. Between writing a user story and the moment a developer can actually start coding, there are usually back-and-forth exchanges, ambiguities, and incomplete specifications. Specialized agents (user story agent, specification agent, contextualization agent) produce ready-to-dev user stories from the very first iteration. The latency between specification and development 80 drop %
- Code quality becomes an invariant, not a variable.. When code agents work from enriched context (document RAG, project rules, technical constraints), and when test agents simultaneously generate unit tests, the residual bug rate drops by 70 %. This isn't optimism, it's simply what we're measuring.
- The documentary debt falls to zero. Each project step produces readable, version-controlled documentation that the relevant stakeholder can challenge. Gone is the chronic misalignment between what is coded and what is documented.
- The productivity of senior profiles is multiplied. Our experienced experts, freed from repetitive transition tasks, They say they are 2 to 3 times more productive. This is far from anecdotal: it changes the profitability of a project and team satisfaction.
Why is the differentiation window today, not in two years
The AI innovation cycle is about six times faster than the average IT decision-making cycle. That's where the real risk lies.
Companies that wait for the market to stabilize, for tools to mature, or for use cases to be «proven» before launching, are actually taking a considerable risk. Because while they are analyzing, others are accumulating.
Agent-based development is not a tool you buy and plug in. It's a capability that is built: libraries of skills by technology, contexts enriched by project, teams trained in a new way of working. Each project carried out with an agent-based approach strengthens this capability. Each month spent without integrating it creates a gap that will be difficult to close.
By 2030, approximately 70 % of repetitive development tasks will be automated. The question isn’t “if,” it’s “when and by whom.”.
The four objections we hear most often (and what they really reveal)
«We don't have the in-house expertise.» That's precisely the problem. Agentic systems require rare profiles, context architects, and engineers capable of composing reliable agent chains. The solution isn't to wait for recruitment, but to combine external expertise with internal upskilling right now.
«We're afraid the agents might do anything.» An industrial agentic chain cannot function without a human in the loop. The Tech Lead validates the rules and security. The Developer supervises and adjusts. QA validates test results. Agentics accelerate and enhance production reliability; they do not replace human judgment, but rather free humans from low-value tasks.
«Our IT system is too complex / too legacy.» Complexity is precisely one of the arguments for agentics, not against. An agent that has integrated your data model, your business rules, and your regulatory constraints is more reliable than a junior developer discovering the project. Contextualization takes time, all the more reason to start.
«We're waiting for the tools to stabilize.» Frameworks evolve quickly, that's true. But the fundamental principles of context engineering, document RAG, and multi-agent orchestration are sufficiently mature today to be industrialized. Silamir has been doing this for several months on real client projects.
What we've learned from industrializing
At Silamir, we didn't wait for the offering to be perfect before applying it to our own projects. Our first client for the agency factory is Silamir itself.
We have learned that agency cannot be improvised in a production environment. The reliability of the chain – quality of injected context, robustness of agents, integration with existing IS, access rights management – requires true engineering. The era of AI POCs is over; CIOs want industrial solutions.
Context engineering is the real cornerstone. It's not the agent that makes the difference, but the quality of the context in which it operates. Precise functional specifications, up-to-date documentation, formalized project rules... Everything we tended to neglect suddenly becomes critical.
And finally, change management remains the main limiting factor. The teams that best adopt agentics are not those with the best developers, but those with a culture of discipline and documentation. The human element remains central.
The question we ask the CIOs we meet is: «In eighteen months, if your competitors have integrated an agentic approach into their delivery chain and you haven't, what does that mean for your ability to innovate and meet your commitments to the business?»
The answer is always the same. And that's where the conversation gets interesting!
→ Contact us to learn more about Silamir's IT4IT Agentic Factory approach
