AI at work: Turning an answer into completed work

AI can save time on writing. Whether it also makes collaboration easier within a business, however, depends on what happens after the answer: Where is the result saved? Who reviews it? And can the next person find the context without first having to piece together documents and chat histories?

For Swiss SMEs, this is an important distinction. A chat window helps with thinking and drafting. A connected work system can also draw on business records and save results where they will be used next. Both have their place. What matters is the work that actually needs doing.

The chat window provides answers; the connection preserves context

A quotation is open, with the project documents alongside it. An employee copies information into an AI chat window, asks it to draft some text and transfers the result back into the document. She then adds missing details and saves the version. The same process starts again with the next assignment.

AI helps with the writing. The employee still maintains the connection between the customer, the project and the result. In the chat window considered here, which has no connection to business systems, documents and information must be supplied manually. The person doing the work must ensure that the current template and the latest project status have been used.

Even a saved chat does not replace a record of work within the business. It does not, by itself, show in the customer file or accounting system who created and subsequently used which result.

Hegyon is one example of a connected work system. Here, the same class of model works with the business's existing records: documents, customers, projects and accounting. Results are saved where the work continues, for example as an article in the newsroom, a note in the customer file or an invoice in the accounting system.

The difference, then, is not necessarily a better answer. It is whether people have to keep moving context and results between applications.

Preparing a company article

A business wants to create an article from existing project documents. In a standalone chat, someone gathers the relevant documents, copies in excerpts and explains the audience and purpose. The draft is followed by questions, corrections and transfer to the newsroom.

For a single text, this is manageable. The direction can change during writing, and different openings can be tried out. This openness is one of the chat window's strengths.

But when the task recurs, so do the manual steps. Gathering sources, writing the brief, checking claims and saving the result remain individual work. What worked for the last article does not become a reliable process simply because it is still in the chat history.

In a connected system, this work can be organised as a routine. It starts on a schedule, researches, writes, checks and revises. It then presents the result to a person. Approval gates between steps can limit how far the process proceeds.

For the business, the benefit is not just the finished text. Recurring preparation and transfer also become part of a defined process. Responsibility for the content remains with the people in charge.

Recording a customer meeting

After a customer meeting, notes need to become a useful record in the customer file. A chat window can help organise statements, identify open questions and formulate next steps. Someone then has to review the text, assign it to the right customer and transfer it to the file.

The demanding part is not just summarising. An option discussed is not yet an order. An open question is not a commitment. The person who led the meeting must check these distinctions in the draft.

In a connected system, the note can be created directly in the customer file. The existing customer and project context does not have to be explained anew each time. The result is available where the next person will look for it.

The benefit becomes clear at handover: a colleague does not need to know which chat produced a particular wording. She needs the reviewed note in its working context. Saving it there does not, however, prove that every statement is correct.

Boundaries around sources are equally important. Not all internal information should be available for this kind of processing. Content in confidential communication spaces and whistleblowing channels must remain protected. Complete context is not a goal that overrides every boundary.

Preparing an invoice

For an invoice, the service details, customer assignment and project reference must match. A chat window can turn pasted information into a clear description of services. But that does not create the invoice in the accounting system. Transfer and cross-checking remain separate steps.

This makes the distinction between text quality and completed work especially clear. A convincingly worded service description says nothing about whether the underlying service has been recorded correctly. Accounting needs accurate information, not just a readable paragraph.

In a connected system, the benefit lies in drawing on the business's own project and accounting records and writing the result back to the accounting system as an invoice. Processing does not end with an answer in the chat.

This does not automatically make the content factually correct. Connecting incomplete records does not make them complete. Before approval, the person responsible must check that the content and assignment are correct.

There is also the question of permissions: to write to a destination, the machine needs the same rights as the person who would make the entry manually. Automation must not create a special route around these requirements. A controlled process therefore depends not just on data sources, but also on write permissions and responsibilities.

Writing and checking are different tasks

A routine needs different capabilities. A model that writes clearly is not necessarily suited to every checking step. A suitable model can therefore be used for each step, rather than one model for everything.

The distinction between reviewing language and checking against reality is especially important. A text can be coherent yet factually wrong. Checking claims about current matters requires a search for appropriate sources. Simply rereading the same draft provides no independent basis.

Search does not replace judgement either. The sources found must be relevant to the claim. Nor does switching models replace human approval. Dividing the work improves the process only when it is clear what each step should check and when a person must decide.

The routine takes care of recurring processing. It does not take responsibility for publication, customer communication or accounting.

Which way of working suits your business?

A chat window is suitable when the task is open-ended: refining an idea, finding the right wording or starting research. If you do not yet know which question matters most, direct exchange is useful. A fixed process would add work before the actual task has been understood.

A connected system is suitable when the same work recurs, business context is needed and the result will be used in a specific place. Other indicators include fixed review steps, defined write permissions and required approvals.

The dividing line is not between simple and demanding work. It is between open exploration and repeatable processing with a defined requirement for where the result is saved. A demanding one-off question may be well suited to a chat window. Even a short customer note may need a controlled connection.

That connection has prerequisites: sources, responsibilities, write destinations and approvals must be clear. Without this structure, automation does not reduce the ambiguity. A routine is not worthwhile simply because a task can technically be turned into a chain of steps.

The first step: make an actual process visible

This week, take a recurring task and follow one actual run-through. Record which documents are opened, which details are copied, which claims are checked and where results are saved. Also note who is allowed to write and who is allowed to approve.

This provides a concrete basis for a decision. If the main benefit lies in thinking and drafting, that points towards a chat window. If the recurring work lies in gathering, transferring, checking and saving, the connection becomes decisive.

Start with that process, not with the choice of model.