)
25% of Spendesk staff have built their own AI agent this year – solving real challenges across our business
With just weeks to go in 2025, I wanted to celebrate one of the most important initiatives of my first full year as CEO of Spendesk.
Twelve months ago we launched a push to integrate AI into every part of Spendesk, Europe’s leading spend management and procurement platform. Today, I’m proud to share that 83% of Spendeskers are using AI agents each week. When you consider that in any given week a few people will be out of the office, this is very close to full company saturation.
Most importantly, after trying out an agent, I haven’t heard a single colleague say they prefer their old way of working.
Our colleagues aren’t just using agents; they’re building them too. After two recent hackathons, which showed how easy it can be for even non-technical staff to create their own AI agents, we now have 213 agents live in our internal library, up from just 10 in January.
Of these, 135 are being used actively. Employees have also built more than 1,000 private agents for personal use (more on that later). Fully one in four of our employees has created an agent this year.
)
These agents are tackling real-world tasks across our business. One of my favourites can give me a full rundown of clients – or potential clients – and their relationship with Spendesk. Before attending a conference or a sales call, I can just drag a list of companies I’m going to meet into the tool, and it can give me a full briefing and suggest relevant talking points for each. I could, of course, ask a member of our sales team to pull this data, but it might take them a couple of hours. This frees them up for more strategic work that can add value.
I don’t know if I should be sharing this publicly, but I actually built my own AI agent to help me prepare for board meetings. It helps me stress test my presentations by predicting what questions each board member will have. It can be scarily prescient!
One of my colleagues also created a “What would Axel say?” agent that they can bounce ideas off when I’m out of contact – or just to help them better prepare for a meeting with me. Another favourite is the agent that mines Slack and email to wrap up everything you missed after a few days off.
My own AI journey
AI has come so far, so quickly. When ChatGPT launched three years ago, my first reaction was to dismiss it as a fancy autocomplete. I also had concerns about sharing confidential, professional data. For true transformation, it felt like the tech wasn’t quite ready.
Over the past couple of years, as AI tools have matured, I’ve become a full convert. I wanted to build a culture at Spendesk where we can all learn and experiment together. No one should be ashamed. Everyone can proceed at their own pace. Dust, which helps create custom AI agents, has become a key partner for us on this journey. After using AI, I’ve never seen anyone who said, “That was nice, but I prefer my old way of working.”
So what’s next for AI at Spendesk? In 2026, I want us to focus on finding new sources of internal data so our agents can become ever more useful – basically making our knowledge management AI-ready. Grabbing relatively structured data from Slack and Salesforce is already table stakes. The game-changer will come when we can mine insights from unstructured data from internal meetings or Notion brainstorming sessions.
In short, we need to find ways so that our AI agents can get better access to our brains. This work of structuring and cleaning our own goldmine of internal knowledge isn’t necessarily sexy. But it will be essential.
)
)
)
)
)