Agentic platforms like n8n, Make or OpenAI Agent Builder accelerate the creation of AI agents and simple automation, but remain limited to peripheral and non-critical uses
We see the proliferation of tools such as n8n, Make or OpenAI Agent Builder.
The promise? “Connect” AI to the information system, create “agents” and automate without having to do too much coding.
In our opinion, these platforms are to AI what No-Code is to the development of business applications: a formidable accelerator for simple, even superficial, use cases, but rarely the basis of critical systems.
1. Core vs. Peripheral
All studies show it: the adoption of No-Code is exploding for non-critical internal tools or prototypes. But “Core Business” (Core banking, Supply Chain, Mass Billing) continues to be treated with “industrial” approaches.
AI platforms are replaying this scenario. They are excellent for:
· Gain individual productivity or equip small teams
· Develop PoCs and acculturate quickly
· Address lightweight processes requiring little integration.
2. The double technical wall
When it comes to complex agents integrated into the heart of the information system, two major limitations appear.
Integration with “Legacy” constitutes a major challenge. If connecting Gmail to Slack is trivial, getting an agent to communicate with an aging IS, on-premise databases and secure protocols is a whole other level of complexity and requirement.
The “Wall of Complexity” is insurmountable. In code, complex logic is contained in a few lines. In visual representation, the management of special cases (“edge cases”) transforms the screen into a sprawling and unreadable tree. Maintenance using a platform-type tool then becomes more expensive than maintaining the code would have been!
3. A target that does not exist
In large organizations, these tools do not always meet their audience.
Developers find them too rigid and restrictive (“Black box”).
Businesses still find them too technical (as soon as they have to manage a loop, an API error or a webhook, the average “citizen developer” is helpless!).
4. A natural polarization of performance and the risk of creating new technical debt
The “Vibe Coding” trend consists of coding instinctively, as you go, while being accompanied on the journey by a caring AI that will fill in your gaps and sweep up behind you. In reality, AI acts as an amplifier. It makes experts excellent, because they know how to architect their applications, structure their code and reread it to debug it, improve its maintainability or optimize its execution performance. But AI also allows novices to produce a lot, very quickly, without real mastery.
With the proliferation of no-code/low-code agentic platforms, the risk is to see the phenomenon of “Vibe Coding” become widespread: generating “feeling” workflows, without architecture. As studies on code quality (e.g. GitClear) suggest, the risk is to generate massive and invisible technical debt, due to lack of testing, observability and robust CI/CD (i.e. without a structured code integration process covering its entire development cycle).
If these platforms are excellent accelerators for ideation and peripherals, they are not (yet?) the hardened steel necessary to erect critical agents at the heart of the company’s information system.
