From secret agents to an agent factory
Scaling the productivity gains of AI agents to the level of the entire organisation requires an industrial approach to producing them. The critical success factor is how agents are managed and led, Juho Nevalainen reminds.
First, the good news: Finnish companies have a strong understanding of the opportunities AI offers for boosting productivity and renewing business. As many as 45 per cent of companies participating in our market survey say that they have gained competitive advantage through AI. The frontrunners have already moved from simply leveraging generative AI into the third wave of AI transformation, and they are deploying AI agents that can carry out knowledge work tasks independently.
But then to the bad news: even among these leading companies, the significance of scalability hasn’t yet been fully understood. Although 55 per cent of companies have begun deploying AI agents, only one in ten has more than ten agents in production use. The same is can be seen in investment levels: 48 per cent of companies have invested hundreds of thousands of euros in clearly defined AI initiatives. However, none of the companies surveyed has yet made genuinely strategic investments exceeding two per cent of annual turnover.
The truth is that having only a handful of AI agents in use does not significantly increase organisational productivity. To paraphrase an old proverb, one agent doesn’t make a summer.
Trust in quality is key
Typically, achieving organisation-wide productivity gains requires a whole library of agents. Some agents generate documents or update databases, while others guide people to act in accordance with shared operating models.
To help our clients scale the benefits of AI agents, we have already established our own agent factory, where agents are produced using an industrial model: business needs drive priorities, output is consistent in quality, results are delivered quickly, and users are supported in adopting their new digital colleagues into everyday work. The agent factory is also part of our initiative exploring ways to improve youth employment through AI.
Trust is central to agent development, as people must always be able to rely on the quality of the work agents produce. This requires a clear framework, underpinning agent operations to ensure that agents do not hallucinate and generate inconsistent outputs for different users. In our factory, this framework is the Business Technology Standard, developed for business technology management. Its structure ensures that every agent performs precisely the right task, in the right way, and in alignment with other agents.
For us, a factory-style, industrial approach to producing AI agents goes far beyond traditional DevOps-based development models – it represents a much more holistic mindset. The production line begins with a deep understanding of business needs and an assessment of the suitability of different technology platforms. After development, it extends to supporting everyday adoption, as well as the governance and continuous improvement of each agent.
From manager to leader of people and agents
Establishing our own AI agent factory and building an agent library based on the Business Technology Standard has provided us with valuable practical lessons on how to manage and lead a large portfolio of agents while ensuring consistently high-quality work.
One key lesson is that tasks assigned to agents must be broken down into sufficiently small components so that the work of each individual agent can be validated and tested for quality assurance. A common pitfall in defining agent behaviour is attempting to force a single agent to handle too broad a scope of work.
For example, agents can be trained to assist knowledge workers in shaping and evaluating requirements related to a new business system. Instead of trying to train one agent to manage all related tasks, the work should be divided among a handful of agents.
What matters most is that agents operate seamlessly together as part of business processes and structures, rather than as isolated solo performers scattered across the organisation. Dividing work into appropriately sized components also ensures that agent quality can be maintained during further development and with every update.
The importance of ownership cannot be overstated. Agents must be managed just like human employees – and the more of them you deploy, the more crucial leadership becomes. Every agent should have a clear owner who thoroughly understands how it operates. In our factory, product managers lead the development of agents from the initial definition phase through to release and continuous improvement.
Download a free market study about where companies are on their AI journey and how to be one of the forerunners.
About the author
Juho Nevalainen works as an Executive CTO of Sofigate’s BT Integrator business, he is also the owner of Sofigate’s AI Agents programme and the leader responsible for Sofigate’s AI offering.
BT Integrator supports growth-driven organisations in complex business transformation projects by bringing together Sofigate’s management and technology services from strategic planning to project management and continuous IT service operations.
Juho has extensive experience in IT management and business development consulting, concept development and commercialisation, as well as in leveraging artificial intelligence for business planning.
Building purpose through partnership
Jobs for young people through artificial intelligence
