Scaling AI-driven development through real-world experimentation

The challenge: Bringing AI into everyday engineering work

Following an Eficode-led git migration, Vaisala set out to build enterprise-wide AI capability, starting with a targeted trial. 

The objective was to fundamentally shift engineers' default habits —moving them from passive learning to using AI in their daily tasks. This would help Vaisala build the capability to integrate AI in its daily workflows and core products at scale.

Vaisala wanted to tackle the common challenges of sharing AI knowledge across team boundaries early on —ensuring custom Claude Code skills, hooks, and configurations built by one team could easily benefit others. 

To power this transformation, Vaisala selected Claude Code as its main AI coding assistant. The team used Anthropic models to power its AI agents, connecting directly through the Anthropic API and via AWS Bedrock. 

The solution: AI sprints built around real engineering challenges

Together, we created a hands-on AI experimentation approach focused on challenges from Vaisala’s engineers’ daily work.

Before the first sprint, we interviewed key stakeholders to understand Vaisala’s goals and make sure the sessions focused on relevant business and engineering priorities.

Instead of working through synthetic exercises, engineers brought real challenges from their day-to-day work into the sessions. With guidance from our advisors, teams followed a practical process to:

• Identify repetitive or time-consuming tasks in their daily work.

• Turn those challenges into clear, testable hypotheses for AI-assisted solutions.

• Build and test working agents using Claude Code and AWS Bedrock.

• Share reusable Claude Code configurations, hooks, and skills across team borders to solve the common challenge of isolated learning. 

• Demonstrate their solutions to internal stakeholders and senior leaders.

This gave participants the opportunity to experiment with AI in a setting directly connected to their own work.

The result: Practical AI experience and a model for scaling

The hands-on format gave engineers practical experience using AI to address challenges from their own work.

During the engagement, participants developed and validated hypotheses for reducing engineering friction and built working autonomous agents. Presenting those solutions to leadership also helped make the potential of AI more tangible across the organization. After eight four-week sprints, we tested over 40 hypotheses created by almost 300 engineers. 

Beyond immediate quick wins, the team also created a clear roadmap for future scaling and development across the broader organization, establishing a repeatable model for other functions and product lines.