We stopped “Doing AI”, and it changed everything

ai-starter-pack-itd-clickonsite-landscape-proportions

Over the past few weeks, a particular image has been making the rounds on LinkedIn. It shows a sheep following a sign labeled “AI”… and the rest of the flock trailing behind. It’s a telling illustration of the hype — and herd mentality — surrounding artificial intelligence.

Will it last? Hopefully not. We’ve seen this kind of buzz before, and it often distracts from what matters.

So, what’s AI’s role at ITD?

At IT-Development, we’re strong believers that when mastered, AI can amplify knowledge and boost performance. But it’s not a magic solution, and above all does not replace real expertise.

That’s why we stopped “doing AI” for the sake of it. We treat it like any other technology component, something we integrate purposefully. The goal is simple: solve a real customer need.

Practical, purposeful use cases

AI is already part of our daily operations across multiple teams:

In communications, it helps us edit text and generate visuals.
In project teams, it supports the structure and drafting of complex documents like procedures.
Our developers mostly use it to debug code rather than generate it.
The testing team uses it to create test cases.

As for ClickOnSite, our telecom site management software, we’ve explored several promising directions for years now:

Counting antennas on towers,
Extracting data from technical diagrams,
Automatically classifying administrative documents (invoices, legal notices, etc.).

These prototypes are often impressive. But when it comes to production, the results tell a different story: performance hovering around 60%, with high development costs. That makes it hard to deliver real, lasting value to our clients.

Getting back to what matters: the AI actual value

So we made a strategic decision: stop innovating for the sake of innovation.

Frankly, we don’t need to anymore. Instead, we’ve chosen to integrate AI where it truly adds value. Today, we use technical components like Vertex AI just as we would any other tool.

We’ve returned to the basics: start with a customer need, then bring together all the tools that can help: LLMs, RAG, basic algorithms, OCR, BPM, our new mobile app, and so on. And most of the time, AI alone isn’t enough to meet the need.

We also adjusted our functional ambitions. We’re no longer building AI-centered features. Instead, we’re embedding AI into existing modules. It’s still important, but it’s no longer the star of the show.

Our latest use case says it all: we’re developing a simple feature that extracts data from a photo and injects it into our database. It’s basic, but incredibly effective. And most importantly, it’s useful.