.png)
The chatbot has popularized LLMs (large language models), but it also reveals their limitations.
The “chat” format assumes that users know what to ask, how to phrase it, and are willing to do so every time.
It’s a one-off interaction — disembodied and disconnected from the actual workflow.
“Today, most people in the workplace are staring at a big chatbot bar thinking:
What am I even supposed to ask it?”
— Pierre de La Grand’Rive, CEO of Delos
This hesitation is telling: the interface is perceived as powerful… but unclear.
It impresses, but it intimidates.
The result? Low adoption, shallow use, and massive untapped potential.
A prompt is an instruction you write in a chatbot to trigger an AI action.
Example: “Translate this text into English in a professional tone, while respecting cultural nuances.”
The more precise the prompt, the more reliable the output — but writing a good prompt takes time, structure… and sometimes skills most users don’t have.
.png)
The logic behind chatbots starts from a good idea: create a universal interface that adapts to every need.
But in practice, it doesn't fit into everyday workflows.
Take a simple example: translation.
Large models are excellent at translating — when given the right guidance.
But asking users to retype detailed instructions every time is counterproductive.
It complicates something that, in a traditional tool, would be fast and seamless.
At Delos, the approach is different.
Instead of offering a generic chatbot, the platform is made up of dedicated applications — each designed for a specific use case: writing, translating, summarizing, searching, interacting with documents.
The goal: build interfaces people use 50 times a day — not twice a month.
“Doing translation in a chatbot is frustrating.
What you want is an interface you can use 50 times a day.”
— Pierre de La Grand’Rive
Take Trad, for example: translations appear live as you type, sentence by sentence — in an interface inspired by DeepL, but powered by newer LLMs.
Users can refine, tweak, and compare different versions — all without ever writing a prompt.
AI Act marks a crucial step in the regulation of artificial intelligence, but it is only the first framework.
Businesses must now adopt a proactive approach to anticipate regulatory and ethical changes. Between innovation and control, the future of AI depends on the ability of economic and political stakeholders to establish clear, fair, and fundamental rights-respecting rules.
This shift is strategic: it’s no longer about assistance — it’s about tooling.
The chatbot remains a “one-shot” tool, where the user bears the cognitive load:
figuring out the instruction, checking the output, redoing it if needed.
By contrast, a well-designed AI application includes:
The result: AI becomes a true work tool — not a black box.
The choice of interface is not trivial — it directly impacts usage rates, and therefore the actual value delivered.
A great model poorly integrated is just a poorly invested budget.
In this context, the chatbot isn’t neutral: it steers AI projects toward exploratory logic, often misaligned with goals of efficiency and transformation.
By contrast, a clear, contextual, task-oriented interface reduces friction, structures usage, and enables wide-scale adoption.
The chatbot marked a turning point.
It made AI visible, tangible, accessible.
But now it’s time to move to the next phase: integration.
Useful AI is AI that fits into tools, habits, and processes.
And for that, we need to move beyond the conversational paradigm.
What Delos demonstrates is that the alternative already exists:
AI integrated by use case, designed from the ground up for adoption.
So the question is no longer “Which model should we use?”
but rather: “Which interface for which use case?”
-min.png)
