Samergo: a day working with seef

What does Muriël use seef for?

Most often for e-mail. She built an agent that helps bring structure to her e-mail and drafts it for her. An example from last week: a first message to a project partner she had never been in touch with, with a background document alongside it. She reads the result through and usually sends it with barely a change.

Then there is the thinking work. On a complicated facilitation project, seef helped her pin down where it was stuck by asking pointed questions. That let her put it to the project leads with a solid case behind it. The same for a course with an unclear assignment: she put her preparation and the relevant documents into a conversation and asked how to approach it. Structure came out of it.

And for meetings. She explains the goal and talks through possible formats. Otherwise she would reach for a book, or search Samergo's own database.

The pattern is always the same: she supplies the context, seef supplies a first version or a structure, and she can use that as the basis for working it out.

“What I like about it is that seef asks questions, and that it gets you thinking again. A report comes out of that, which is your basis, and you work on from there.”

Which agents does the team use?

An agent is an assistant with fixed instructions for one task, so you do not have to explain what you want every time. Muriël names three. Her own e-mail improver, the one she uses most. A copywriting agent her colleague Joris built. And an agent for the service report, which comes round twice a year and saves a lot of time, she says.

So two of the three are not hers, and that is exactly the point. Once someone has worked out which instructions produce a good text or a good report, they capture that in an agent. You can then share it with the whole team, so the whole team gets something out of it. The knowledge stops sitting with one colleague and becomes part of how Samergo works, without anyone writing a manual for it.

How much time does it save?

These are her own estimates, not measurements. Improving an e-mail is a small job, so perhaps five minutes saved per message. Preparing a meeting saves an hour or two, she reckons. The biggest gain is in analysing data such as interviews.

“I had analysed quite a lot of interviews within a day. Normally that is a week at least, I think. It really depends on the job you bring AI in for.”

She is careful about quality, and that goes for AI in general, whether it is Copilot, ChatGPT or seef. It is not a matter of putting something in and being done; AI makes mistakes too, so you keep checking. Stay critical of the output and the quality matches what a person delivers, she says. What seef adds on top: connections you might not spot yourself, speed, a different angle, and sharpening her own eye for what she is analysing.

That is the time seef gives her back: from five minutes on an e-mail to days on an analysis.

Copilot, ChatGPT or seef: what did she use before, and what does she still use?

For work she used to have Microsoft Copilot, and ChatGPT privately, deliberately kept apart. Copilot sat inside her documents: improving a sentence as you type was quick. Now she copies the text into seef and pastes the result back. It took getting used to, she says, but it is not something she misses. She hardly opens Copilot any more. The last time she did, the difference stood out: that text reads like AI, while seef by now writes more the way she does. That is no accident. seef remembers, for each user, the tone, length and style they want their texts in, and carries that into every following conversation. She never noticed that with Copilot. For her work, seef has replaced it.

For her own company she uses ChatGPT, and lately she gets stuck with it: the same thing every time, no longer original, not doing what she asks. She has not run into that with seef.

What is not finished yet?

The shared knowledge base. A colleague set up the document library, and Muriël does not know exactly what is in it, or why those documents were chosen. Sometimes seef pulls something out of it that she did not want in there. For conversations about a single project she turns the library off and works only with the files she uploads herself. Her verdict: we have not set up the shared knowledge base properly ourselves yet. Once we do, she says, it will be very good. That is how seef is meant to work: the company decides which documents go in the shared library, and for each conversation you decide whether seef may search it.

Large files. Sometimes a file is too big to upload. She compresses it first and then it goes through; some colleagues reach for another AI for that kind of job.

What has been fixed, after her feedback: at first seef used an uploaded file for the next answer only and then went back to searching the library. Now seef keeps searching the file she shared in the conversation.

“I thought: I should just keep using this, and share everything I run into with you. The next time, I did not run into it any more.”

Who does this not work for?

Colleagues who are less digitally skilled need more guidance before they get good use out of it.

And people without patience, she says. You have to get used to AI in general, and to seef as well. When an answer is wrong, you talk to the model until it understands where the mistake is. And you keep checking for yourself. Anyone doing it purely for convenience and speed is better off not using AI at all, she reckons. Those who do have the patience find many tasks get lighter, and sometimes gain more depth.