One Company’s Blueprint for Taking a Whole Engineering Org Agentic
This Kilo customer isn't waiting to see how the shift to agentic engineering shift will play out.
Two things you don’t usually associate with moving quickly: legacy codebases and government applications. Ida Infront, which builds software for Swedish and Nordic government authorities, is challenging that with an ambitious goal: doubling the company’s development speed by the end of the year—and they’re using Kilo to do it.
Ida Infront builds secure communication systems, case management tools, and digital archiving solutions for customers including national agencies with demanding data handling requirements and long records retention obligations. The company was established in 1984, and the platform underlying everything they build is over 25 years old.
In these environments, compliance concerns, risk analysis, and legacy infrastructure can hamstring experimentation.
Magnus Grimsell, CTO at Ida Infront, knows that embracing AI is how the company will stay competitive: “The value of a large, established platform can decrease when new products can be written faster and cheaper,” he says. “We want to address that risk and take advantage of the opportunity by being early adopters.”
So the company has been on a structured AI journey since GPT 3.5 launched. They started with Sourcegraph’s Cody, but within six months they were already looking beyond coding assistants to the next frontier: agentic engineering.
After evaluating several tools, including Cline and Roo Code, Ida Infront chose Kilo Code and began rolling it out in January of this year, starting with three forerunner teams and 70 seats across developers and other technical roles.
The Rollout Plan Is Just as Critical as the Choice of Tool
Magnus has watched peer companies in the industry tell developers they can use whatever AI tools they want, then wait. “Nothing happens,” he says. “There’s no progression in AI adoption, no next step.” Random experimentation isn’t good enough; without standardizing on anything, you won’t get real results.
So Ida Infront did the opposite: they selected one tool, committed to it organizationally, and built a structured onboarding process to bring every team through it together.
You Can’t Throw Tools at People and Expect Something To Change
I say this to engineering leaders constantly, and I’ll say it here too: buying AI licenses and handing them out is not an AI strategy.
Each pilot team goes through a series of three workshops, developed in close partnership with the Kilo team:
The first covers agentic engineering fundamentals, with the team mob-programming a real backlog item using Kilo. Before the session, a team ambassador—an AI champion appointed from each team—has already prepared the repo with an AGENTS.md file and relevant MCP tools.
The second workshop, held 1-2 weeks later, addresses what came up in practice: debugging issues, adding skills to the agent, connecting new MCP servers, refining workflow.
The third is non-technical. By this point the team has been using Kilo in production for about four weeks. The session walks through the four phases anticipated in Ida Infront’s AI adoption framework: AI Assistant, AI Agents, Multi-agent, and Software Factory. Teams discuss where they think they are, what they’re expecting, and what concerns them. There’s a structured exercise for surfacing fears and hopes.
“We had contact with the Kilo team throughout the first pilot,” Magnus says. “We didn’t just receive support for the product, but expertise that I’ve valued as well. It’s great to have somebody come in and say, ‘This is how we do agentic engineering.’”
The staggered rollout lets the company train their future AI trainers: the early adopter teams will then be able to coach others.
Early Indicators
Magnus describes a full spectrum of reactions to the agentic shift: excitement, skepticism, and a form of grief from developers who feel that the craft they’ve spent years developing is changing in ways they didn’t ask for. The final workshop is an opportunity for team members to feel heard and work through any resistance. “We have people who are on board but feel sad about it,” he says. “That’s a legitimate response. It needed space.”
Since the rollout began, the dominant attitude has been positive: teams are eager to get onboarded. Ida Infront’s developers who had been using Cody heavily were prepared for an incremental improvement when they saw Kilo Code in action for the first time. “There were some dropped jaws,” Magnus says. “With an assistant, you’re working the same way as before, just a little faster. With agents, the developer role itself starts to change.”
That shift is already visible in how Ida Infront is thinking about the future of their teams. Specialized agents for performance optimization or domain-specific tasks—filling skill gaps at the team level rather than requiring escalation to in-house experts—are on Magnus’s roadmap. Other companies in Ida Infront’s parent group are already looking to replicate their model.
The Proof Is Already Showing up in Their Products
Ida Infront is also building AI into the products they deliver to customers. ARN, the Swedish National Board for Consumer Disputes, introduced an AI-based function developed with Ida Infront in 2025 that automatically classifies incoming case documents like receipts, images, and correspondence—tasks previously handled manually. ARN’s summary of 2025 noted that AI investment freed staff from manual classification work and contributed to significantly shorter case processing times—an outcome Ida Infront is working to replicate across more of its customer base.
There’s a sense of excitement and possibility at the company. “I’ve been through the shift from waterfall to agile. That changed everything—not just how we wrote code, but how teams were organized, what roles existed, how we thought about delivery,” Magnus says. “This is the same kind of shift. It affects everything. And I’d rather be shaping what that looks like than waiting to find out.”