Agents and automation
Agents that read, sort and prepare the work, and a person who approves.
Examples
- Quotes from site visit notes
- Orders from email to ERP
- Notifications to customers
Technology partner for companies with 10 to 250 employees
For companies that need a flexible, hands-on technology adviser to modernise their workflows, internally and with their partners and customers.
What we deliver
Agents that read, sort and prepare the work, and a person who approves.
Examples
Accounting, CRM, inventory and specialist systems that share data, so nothing is entered twice.
Examples
Interfaces where your customers and employees find the answer themselves.
Examples
Data from operations, sensors and systems, brought together into overviews and alerts that management uses.
Examples
From conversations with clients
These statements come from conversations with Norwegian SMEs over the past year, in industries from construction to energy. We design automation in which controlled agents take over the repetitive work, with new, user-friendly interfaces where it makes sense.
Bids from up to 200 subcontractors arrive by e-mail in their own formats, and someone puts them together by hand.
An agent reads the bids whatever the format and lines up comparable items. Every item links to the line in the original document, so the buyer sees the source and confirms or corrects it.
Our customer desk spends the day on status questions and reminders, instead of contracts and customers.
An agent follows the status in the rental and finance systems and sends the customer a text or e-mail when something happens. The customer desk handles the exceptions.
Every department writes its quotes from scratch, so they take anywhere from five minutes to two hours and look different every time.
Notes from the site visit or the e-mail become a draft with prices, quantities and standard terms. The salesperson adjusts and sends it, and every quote won or lost makes the next draft better.
When stock runs short, we add more columns in Excel to work out which orders to pick first.
When stock runs short, the system proposes a prioritised pick list based on the rules you have set. Someone approves it with one click before it goes to picking.
The customer books a pickup by e-mail, the driver counts by hand, and the warehouse gives the office a handwritten list.
Pickups are booked and recorded in one system, and driver and warehouse confirm at the same moment. That gives one shared record instead of three paper copies and an argument about the numbers.
Equipment faults can go unnoticed for months, even though everything is in the logs.
An agent reads the logs continuously and flags idle chargers, low output and faults within the hour. Every deviation becomes a case with an owner and a priority, not a row in a report.
Only a few experienced people know which machines a new part should go through and how long it takes, and the number is typed in by hand.
An agent reads the drawing and proposes machines and time based on earlier orders, straight into the production system, and the experienced person approves. The knowledge stays in the company when someone leaves, is ill or retires.
Purchases we re-invoice are typed in again as a sales order and an invoice, entirely by hand.
The incoming invoice becomes a sales order and invoice automatically, with the right mark-up and contract reference. Pricing rules are checked before anything goes live, and exceptions go to a person.
The invoice basis is put together by jumping between Excel and the line-of-business system, and nobody sees the invoice before it goes out.
Finance gets one workflow where an invoice can be previewed, approved or held back, with a notice to whoever owns the assignment. Manual rows are flagged for review.
Mapping and context
Through conversations and access to your systems we see how the work actually flows today.
Orders typed in twice, reports built by hand, and routines only one person knows. Pinpointing where the time goes.
Orchestrated agents move information between the systems, and a person approves before anything goes live.
We review what has been delivered with you, and build further where it pays off most.
Context engineering
Agents write code fast. Whether that code solves the right problem depends on what they know about you. So we keep everything we learn about your business in one place, and keep it current for as long as we work together. Our people use that context when they specify the work, and the agents use it when they build.
Orchestrated agentic development
Many have let AI write code and ended up with something that works in a demo, but not in production. What decides the outcome is how the agents are directed. With us, they work from a specification and a design you have approved, and with the context we have gathered about your business. Every change is tested against the requirements and reviewed by a developer before you approve it.
Our advisers write the specification and design the solution with you, in your own visual identity. You approve both before anything is built.
An orchestrator splits the work between agents with separate roles. The agents follow our own skills, written instructions for how we build, test and secure.
The agents test against the requirements, and a developer reviews every change.
You approve before anything goes live, and see the result in Bridge.
How we work
We go through your systems, workflows and ambitions. You get a prioritised list of what to connect and automate first, and what it will return, whether or not you continue with us.
We work through the list in the order you decide, one task at a time or several in parallel. The fee is the same every month, and you approve the specification before we build.
We regularly review what has been delivered with you, and adjust what misses the mark. New requests and improvements go into Bridge as tasks.
Bridge, the collaboration space
In Bridge, your employees, our advisers and the agents work in the same space. The agents prepare, build and test. The people decide, approve and follow up. Everyone sees the same thing: what is waiting on whom, what has been decided, and the status of the roadmap you steer by.
We hunt for gains on your behalf
The research group METR measures how large a task an AI agent can solve on its own, counted in the time it would take a person. In 2019 that was a few seconds. Today the best agents handle tasks that take a person more than 16 hours, two full working days, and since 2023 the length has doubled roughly every four months.
We keep up and hunt for gains on your behalf. Every quarter we review where the next gain is biggest, and we tell you when something is not worth the money. That leaves your time for the core of your business.
Three promises
The code, the contract and the decisions are yours. We do the work, you own the result.
We take on following the field and translating it into your day-to-day, so you can stay focused on running the business.
You see how we work and why we make the choices we do. That builds capability on your side instead of moving it out.
Questions we often get
If you have developers who can lead agents, that is a good choice. Most companies we work with do not. We take responsibility for understanding the problem, writing the specification, building, testing and running it, and we answer for the result. The tools are the same. The experience and the accountability are what you get from us.
In most cases, yes. Where a system has an open interface, as Tripletex and Business Central do today, we use it. Where there is none, we can read the database or the documentation, and in some cases let an agent work through the user interface. We always build in a separate test environment, and nothing touches production until a person has approved it.
We only get the access the work requires, and personal data is kept out wherever possible. You get a data processing agreement before we start. When the solution talks to your users, it has to be clear that they are talking to an AI, as the EU AI Act requires.
The review shows where the time goes and where there is most to gain, before you decide on the next step. Along the way we review what has been delivered with you, and improve what misses the mark.
You do. The code, documentation and decisions belong to you. While we work together, the code usually lives in our repository, and you have full access to it. If you end the partnership, we hand everything over to you.
Contact
The first meeting takes an hour. Tell us how you work today, and we will tell you where we think there is most to gain, and what a mapping would give you.
Or write to info@sevenpeaks.no