AI adoption: nemrut-core for engineering teams

Tailoring an AI foundation around real work gives teams a clear reason to use it. For an engineering company, Elia Studio proposed nemrut-core and implemented a proposal-preparation assistant. The founder reports that this approach substantially accelerated AI adoption. The workflow produces editable documents while keeping human review and approval in place, rather than replacing professional judgement.

Start with a task people recognise

Opening a chat interface does not, by itself, turn an adoption ambition into everyday use. The initial tool should address a recognisable task and produce something professionals can examine.

In this engagement, the starting point was commercial proposal preparation. Inputs included customer requests, specifications and email exchanges. The task was to structure that material into a first proposal, not replace engineers in their decisions.

Working definition: business adoption means incorporating a tool into real work. It is not equivalent to account creation or the size of an organisation.

The client remains anonymous. This account is based on Alexandre Gon’s project summary and clarifications, not an independent statistical study of adoption.

Why propose nemrut-core instead of starting again?

Elia Studio proposed its nemrut-core foundation, then tailored it around the selected first use case. The value lies in concentrating delivery on the workflow, documents and rules specific to the business.

Think of an equipped workshop: the tools are available, but the object being made remains specific. A document template, review step or writing rule does not become universal because a component can be reused.

Our Nemrut introduction distinguishes a platform’s role from a delivery engagement. Here, the evidence concerns the customisation performed, not a promise that everything is ready for every company.

A usable document, not merely convincing prose

The requested output needed to follow the company’s Word template and remain editable. That requirement is part of the business task: people need to review, correct and finish a proposal in a usable format.

The implemented workflow generated a proposal in that template and allowed its content to be revised in an editor. Iterations addressed fidelity to the brief, wording, editing and document rendering.

There is a meaningful difference between producing text in a window and preparing a document within a process. Fluent output is not sufficient if someone has to reconstruct the proposal elsewhere.

AI prepares; professionals retain the decision

Verification, signing and sending remained human responsibilities. That boundary is not a weakness: it makes the assistant’s role in the work explicit.

The French data protection authority, CNIL, explains the limitations of generated outputs and the need to verify them. It also recommends making users familiar with authorised uses and the information they may submit. CNIL FAQ, published 18 July 2024, reviewed 4 November 2026, in French.

Human approval for AI agents should therefore represent a genuine review. People need to understand what has been proposed and what remains to check, rather than treat every draft as ready to send.

What was observed?

Business reviews confirmed several parts of the workflow: request structuring, proposal preparation, Word-template generation and editing. Further corrections and improvements remained identified.

The founder reports substantially faster adoption following this customisation. That is a qualitative founder-confirmed outcome, not an independently measured usage rate or quantified time saving.

This separates a tangible implementation from potential deployment reach. It also gives prospective buyers a clearer picture of what they can examine than an unsupported productivity percentage would.

A cloud path prepared alongside the workflow

The engagement also covered cloud requirements, comparison of offers and a phased recommendation. The aim was to reconcile data control, reversibility and budget without committing prematurely to disproportionate permanent capacity.

A recommendation is not proof that the target environment has been contracted or deployed. Our guide to sizing sovereign AI for mid-sized organisations addresses that architecture decision separately. This case focuses on adoption of the first business tool.

A transferable method, not an identical recipe

For another company, we propose starting with a document or outcome that users know how to evaluate. Identify authorised inputs, workflow stages and decisions that must remain human.

Acceptance questions can be concrete: does the draft reflect the request? Are missing details visible? Is the document editable? What changes are needed before approval?

These are proposed criteria for future scoping, not a claim that every check was exhaustively executed in this engagement. Our pre-launch evaluation guide helps structure that work.

A first implementation should also reveal what is missing: examples, templates, business rules or an unresolved operating responsibility. Those findings can guide the next iteration rather than be concealed behind a polished demonstration.

Move from intention to a working first agent

Elia Studio builds tailored AI solutions so business teams can focus on their work. The initial sprint from €5,000 includes development and delivery of a working agent MVP within an agreed scope.

That entry price is not presented as the price of this engagement. It does not guarantee every integration or company-wide deployment. The proposal specifies what will be developed, delivered and evaluated.

Explore our tailored AI services or contact Elia Studio to identify the first use case worth equipping in your organisation.