An agent handles one job that keeps coming back, and one of your people approves the result before it goes out.
Your rollout partner for 
Microsoft Partner · AWS Partner
Your estimator decides
RequestsA request arrives as a PDF, and someone types the items into the system by hand.
QuotesFor every quote, your estimator pulls prices together from lists and old quotes.
Partner documentsThey sit in the mailbox, and your QA lead checks page by page whether they match the specification.
Board packThe board pack gets put together at the weekend, from folders only a few people know.
When that person is on holiday, the work waits.
If you run the firm and sit on its board as well, both questions land on your desk: what AI costs, and what it has done so far.
Someone got the task, with no method and no place to start. Often that someone is you.
After a session or two, people went back to the old way of working.
Often on personal accounts, so company data can go in with no contract and no log.
In the back office, the work is still done by hand. That is where we start, with one job.
Three recent projects for larger firms. We name a client only once they agree.
A European software company at €100 million ARRThe AI assessment ran end to end, with eleven discovery sessions and thirty-six opportunities scored. With their IT and Security team, we then built the governance for the Claude Enterprise rollout.
A European software vendorWe rebuilt its go-to-market with AI: strategy, marketing, partner programme, sales and operations.
A global consulting firmAn AI outbound engine on ChatGPT Enterprise, connected to Salesforce and external B2B databases, which gives every sales rep a call list each day.
Our engineers have also built data and AI systems in manufacturing, energy, transport, finance, healthcare, the public sector, real estate and security.
An agent takes data from several of your systems, applies your rules and hands the result to the person in charge to check.
Before
Chat
One person
Please summarise this supplier email.
The three main points: the price goes up in March, …
Now write a reply.
You copy every answer into your systems by hand.
After
Agent
The agent
skillsplugins
Your purchasing team checks it
The work runs between your systems, by your rules.
The AI inside your ERP knows your ERP. The agent works between your ERP, mailbox, files and CRM, and your rules come with it as skills and plugins: your price list, your approval steps, your wording.
Before the first run we set up who may access what, the guardrails and a log. Where it matters, a person decides.
An agent follows the same steps every time: it reads the request, collects what it needs from your systems, checks it, writes a draft and stops for someone on your team. Click through three examples.
An inquiry comes in
“Please quote 40 brackets as in drawing B-12, powder-coated, delivered by the end of March.”
The agent knows this job: a quote in your format, checked by your estimator.
Pick your industry or the part of the business where the work waits longest. Each page names the drafts an agent writes there.
Machine and plant buildersService requests, spare parts and manuals
Contract and metal manufacturingRequests with drawings, and costing
Windows, doors and façadesTenders and quotes
Kitchens and interiorsQuotes with many suppliers involved
Technical wholesaleOrders from mail and PDF
Building servicesQuotes and site reports
ConstructionTender lists, measurements and site reports
Food producersOrders and labels
Other industriesRequests, quotes and confirmations
Quotes and ordersQuotes, orders and confirmations
Purchasing and supplyPurchase orders and supplier confirmations
Service and after-salesService requests and spare parts
Documents and qualityRecords and specifications
Partners and networkChecking your partners' documents
Finance back officeFrom site report to invoice
Reports and knowledgeThe board pack and your numbers
Planning and operationsDates, capacity and open items
Personal assistantAn assistant for each person who needs one
Unless you choose otherwise, your agents run in your own environment and we operate them. You choose the model for each job.
The job: drafting quotes
Your Microsoft 365
In Copilot and Teams, where your team already works.
For public and internal files
Your own server
Custom and open-source agents, on a server you control.
For confidential files
Private AI
A model that runs under your firm's name.
For secret files
On 12 November 2026, Outlook, Teams and Copilot in Microsoft 365 went down for more than a day. An agent that runs in one cloud only stops when that cloud stops. So we keep the job separate from the platform and the model, and it can move.
Sources: The Register, 12 November 2026; Wikipedia, article “2026 Microsoft outage”
Microsoft 365 is an important place for agents. It shouldn't be the only one. We build on what you run and keep a second way open: your own server and open models.
If you stop, the agents, the rules and the data stay with you.
Most firms try one of these first, and each has its strengths.
| Question | Waiting | A strategy consultancy | Doing it yourself | PromptPartner |
|---|---|---|---|---|
| Does something go live? | Waiting: No | A strategy consultancy: Often a report that nobody builds | Doing it yourself: Sometimes, when your power users find the time | PromptPartner: Yes, agents working in your systems |
| Who keeps it running? | Waiting: Nobody | A strategy consultancy: Nobody, once the project ends | Doing it yourself: Your power users, on top of their job | PromptPartner: We look after your agents every month |
| Where does the data stay? | Waiting: Often in personal AI accounts | A strategy consultancy: Described in a policy paper | Doing it yourself: It depends on the tool | PromptPartner: Where you decide, set out in a data agreement |
| Who trains your people? | Waiting: Nobody | A strategy consultancy: A workshop | Doing it yourself: Your power users, when they can | PromptPartner: We do, every week while we build |
| Who owns it if you stop? | Waiting: There is nothing to own | A strategy consultancy: You keep the report | Doing it yourself: You do | PromptPartner: You do. The agents, the rules and the data stay with you |
Firms with 10 to 250 staff start with the trial run. Larger firms start with the AI assessment.
We run an agent, read only, over cases you've already finished, for example 10 of your quotes. Afterwards you know whether the job suits an agent, and you keep:
You know where an agent fits in your firm, in one department or across several. You keep:
You decide after each step whether to go on. From the AI assessment on, each step has a fixed price, and we bill no hours.
While we set up your agents, we train the people who will work with them. Every week they bring cases from their own work to the AI office hours and leave with a better way to handle them.
We spent twenty years building software that firms use to automate their servers. Today we bring AI agents into firms like yours.
Lukas Hertig takes that call himself. We're based in Schindellegi, in the canton of Schwyz.
Two tools from our work are open source under the MIT licence: AgentSmith and a Bexio MCP server for Swiss accounting.Our open-source tools→
Companies on our path
At first the agent only reads, through an access your IT partner approves. In the trial run we also check in writing what your software can already do. If it can do the job itself, we tell you, and there is nothing for us to build.
Always someone on your team. For quotes it's your estimator, for partner documents your QA lead. Nothing is sent, booked or filed until they've approved it.
Then the agent can run in your Microsoft 365, where your team already works. We build in your prices and rules, test the agent on cases you've already finished and look after it every month.
Where you decide, depending on how sensitive each file is. We put that in writing in a data agreement before anything runs.
Three things: which job the agent prepares, who checks its work and after which step you can stop. If the board wants more, the AI Value Canvas from the AI assessment shows who is responsible for what.
We tell you the price of the trial run in the first call. From the AI assessment on, each step has a fixed price, set by the size of your firm and by how many jobs it covers. We bill no hours.
In 30 minutes we look at one job with you. If your software can already do it, or your data can't carry it, we tell you and stop there.
Cookies let us see which pages work and show our ads to the right people. Nothing is stored until you choose. Privacy policy.
