We did it in-house first.
At the turn of 2025 and 2026, we went through our own AI transformation. Today, AI agents cover the entire software development lifecycle as well as the company’s internal operations.
- −60% spend on software and cloud
- 71% of spend runs on our own compute
- 15 tools replaced
- ~6 months payback period
Fifteen tools we no longer pay for.
SaaS and cloud services we used to pay for, which we now run and, thanks to agents, are moving to our own compute, or which our senior people are building themselves with their agents.
| Area | Before | Today |
|---|---|---|
| 01Cloud infrastructure | public cloud | our own servers |
| 02Work management and company wiki | SaaS for tickets and documentation | kanban and documents in our own AI operating system |
| 03Production error tracking | SaaS | self-hosted |
| 04Monitoring and logs | SaaS | open-source stack on our own hardware |
| 05Container registry | SaaS | our own registry |
| 06Browser testing | SaaS | agents write and run tests on our own runners |
| 07In-app bug reporting | SaaS widget | our own widget; an agent files and triages the bug |
| 08Cookie banner and consent | SaaS | our own solution |
| 09Invoice approval | SaaS | an agent on top of invoices |
| 10Internal records | no-code database | our own app built by an agent |
| 11Diagrams | SaaS | agents draw diagrams straight into documents |
| 12Process mining | SaaS | an agent on top of process data |
| 13CRM | SaaS | an agent on top of e-mail and our own data room |
| 14Website hosting | SaaS | our own hosting |
| 15Design | SaaS | switching off, no longer needed |
- Investment
Our own servers and two people from the team for two to three months.
- Savings
High six figures in crowns every month, on cloud and software alone.
- Payback
Six months.
One source of truth.
Fifteen separate tools for sales, invoicing, approvals, and reporting, with thousands of hours in Excel in between. Every number in several versions. Answering “how much are we making on this project?” took hours or days across the team.
Today: one platform for running the company. Sales, projects, finance, and reporting in a single data layer. Hundreds of specialized agents on top of company data. Leadership gets an answer within minutes.
- 2 monthsfrom kickoff to switching off the old tools
- 15 → 1tools replaced by a single platform
- Thousandsof hours of Excel work nobody does anymore
- CRM
- Invoicing
- Approvals
- Reporting
- Excel
- + 10 more
15 tools, no connection. Every number in several versions.
- Sales
- Projects
- Finance
- Reporting
One platform. One source of truth.
Frontier models are the engine. The rest runs on our own metal.
We don’t replace frontier models. They are the engine of the whole system. Alongside them, we run open-source models on our own hardware wherever data must not leave the company or the client, and we can switch to them without touching the rest of the system.
What stays: communication, office tools, source control, and video and voice tools.
After the switch, roughly half of our remaining software and cloud spend goes to tokens. We pay for intelligence, not licenses.
- Communication
- Office tools
- Source control
- Video and voice
Why we don’t put a price on them.
Not because they are small, but because they can no longer be compared fairly. People who switched to agentic work are 10×, sometimes 100× more productive than two years ago. But comparing that with how software was delivered two years ago makes no sense.
That’s why we publicly stand by our cloud and software figures, which are hard numbers, and talk about processes in multiples, not in money.
The same way we work with clients.
- 01
Measure where money and time leak.
SaaS and cloud, repetitive work, paperwork, and reporting. Not “we’ll roll out AI”, but “here’s 15% of costs”.
- 02
Start where the impact comes in weeks.
Early results convince the team more than any presentation.
- 03
Every automation has an owner on the team.
The agent gets rules, memory, and feedback from the person who used to do the work.
- 04
Get people excited about change.
The technology is the easy part. The hard part is making the team want to work differently.
- 05
Offer clients what works for us.
Same math, same order: the biggest impact first, then the rest.
“Three quarters of what we used to pay for SaaS and cloud now runs on our own compute. Not because those tools were bad, but because our senior people can build and run them themselves with their agents. And we can deliver the same for our clients.”
Today, we use the same method to find opportunities for clients.
Often in industries far from software. We start with Launchpad: within 2–6 weeks, we’ll tell you where AI makes sense for your business and what return to expect.