- Specification of integration solutions
- Implementing integrations and unit tests
- Test execution and version control
- Publishing software documentation through an agent and CI/CD pipeline
The result: our team delivers high-quality integration solutions faster, freeing up time for what matters most: understanding Fingrid‘s needs, complex architectures of the national grid, and business processes. This allows Fingrid to focus on its core mission of ensuring the stability and security of the main grid and enabling the clean energy transition, knowing its integration landscape is efficient and reliable.
What is an Agentic Harness?
We have built what we call an Agentic Harness. It’s a version-controlled AI toolkit shared across our team. It isn’t a static tool; it constantly evolves through iteration and practical delivery feedback. This harness includes:
Development guidelines: Team conventions and organisational knowledge, directly embedded into agent instructions.
- Skills: Reusable agent modules for specific, repeatable tasks, like automated code review.
- MCP servers: Both off-the-shelf and custom-built, these connect our AI to external systems and knowledge bases.
- Custom agent definitions: These orchestrate multi-agent workflows throughout the full development lifecycle.
When an agent misbehaves — for example, hallucinating in its output — we refine its instructions to make its behaviour more predictable. This is a deliberate practice: AI systems are probabilistic, and without precise guidance, they will confidently guess. Building this discipline into our daily work elevates the entire team’s understanding of how to work effectively with AI.
Crucially, this expertise isn’t tied to any specific AI model or development environment. Our consultants can adapt to any customer’s requirements on AI tooling, whether that means different model providers, local versus cloud deployment, or varying levels of AI governance.
Driving quality and cost-efficiency with specialised agents
As AI model operating costs continue to rise, cost-efficiency has become a primary focus alongside quality. Rather than always defaulting to the most powerful frontier AI model, we match the right model to the right task by using specialised agents.
This approach brings several key benefits for our customers like Fingrid:
Minimised errors: Each agent only interacts with files relevant to its development phase, keeping changes precise and reducing the risk of incorrect outputs.
Faster delivery: Design gaps are caught before any code is written, cutting rework and keeping projects on track.
Consistent quality: Automated code review enforces conventions and best practices on every pull request, without relying on individual reviewers to catch everything.
- Always accurate documentation: Docs are automatically generated from the live codebase, so they stay up to date without any extra effort from developers.
For a critical infrastructure operator like Fingrid, these aren’t just development conveniences; they directly reduce the risk of integration-related errors affecting grid operations and speed up the delivery of new integrations needed to support Finland’s clean energy transition.
Born from the trenches, built for impact
What makes this approach unique is that it was built incrementally by the developers doing the actual delivery work at Fingrid — not handed down as a top-down tool. This continuous feedback loop, where we build the harness and use it in production, keeps it sharp, practical, and genuinely useful.
This is hands-on, tested expertise that we consistently deliver to every customer engagement, directly contributing to their business success and enabling them to navigate complex challenges like the energy transition.