When an agent executes and a human guides, the bottleneck is not the technology
Most companies use artificial intelligence to complete legacy work steps slightly faster. The real value is unlocked only when processes are redesigned so that AI agents perform the work and humans steer it. The technology is already ready for this, yet the vast majority of organisations are not.
A typical AI rollout looks like this: A specialist opens a chat window, asks for a summary or draft, copies the output into another system, and continues working as before. An individual step becomes faster, but the process remains unchanged. A human remains the one transferring information from one place to another, verifying, compiling, and reporting.
This is why many organisations wonder why their AI investments do not reflect in the bottom line. They do not reflect because the way work is done has not changed.
From assistant to agent
The next step is agentic processes. An AI agent does not merely answer a question; it executes a multi-step task from start to finish. It gathers data across various systems, makes decisions within defined rules, and updates the systems accordingly. It escalates only exceptions and situations requiring human judgement.
The difference is fundamental. In the assistant model, the human executes and the AI assists. In the agentic model, the agent executes and the human supervises, directs, and decides.
Almost every organisation has processes where the bulk of the effort involves moving, reviewing, and compiling information. Examples include order and invoice processing, reporting, customer request classification and routing, quote preparation, and data maintenance across systems. In these instances, the question should not be how AI can help the employee. We should be asking why a human is doing this at all.
This transition has already taken place in software development. At Crasman and across several of our client projects, AI agents now handle the majority of routine work. They write detailed task descriptions for tickets, write code, generate tests, and conduct most code reviews. The role of developers has shifted from manual execution to direction. They define what is built, own the architecture, and ensure the quality of the final outcome. The same team now delivers significantly more than before.
We have implemented the same transformation in our customer support. An agent resolves routine inquiries from our clients, investigates the background of support requests, and routes more complex cases to the appropriate specialist with full context prepared.
Why we settle for polishing the old
If the technology is ready, why do companies remain stuck incrementally refining the old? Because it feels safe. No one's role changes, no one takes a risk, and no one has to make a major decision. Four underlying factors drive this.
Fear for job security. Reducing manual work means roles will evolve. If leadership does not address this openly, people will protect their current tasks—and one cannot fault them for doing so. Transparency works better than promises that nothing will change. Communicate clearly what the objective is and how the freed-up time will be used—whether for client engagement, innovation, or business growth without linear hiring needs.
Excessive risk aversion. An error made by an agent feels more egregious than one made by a person, even though humans make mistakes in manual processes continually. Risk management is essential, but it must not lead to paralysis. Inaction also carries a cost: processes stay slow, expenses remain high, and competitors who dared to move ahead pull away.
Lack of clear guidelines. Agentic processes demand decisions that have not yet been made. Which systems may the agent access? What can it decide autonomously? Who is accountable for the end result? Without these policies, every initiative stalls on the same questions.
Capabilities. Designing processes around agents is a new discipline. It requires people who understand both the business and the capabilities of AI, alongside time to experiment in practice.
Each of these is a matter for leadership, not IT.
What can leadership do today?
Select one process and design it from scratch. Ask what would remain for a human to do if the process were built today.
Set an ambitious goal. Not "improve efficiency by ten per cent", but "eliminate the vast majority of manual work". Small goals yield small changes.
Define the rules of engagement for agents. What may the agent do autonomously, where does a human approve, and who carries overall accountability?
Speak openly about the evolution of work. What will people focus on when routine work diminishes?
Measure the process, not the tool. Lead time, cost, and quality are far more informative metrics than user counts.
Incremental improvement is a sound way to learn, but a poor strategy. Organisations that dare to redesign their work today will achieve substantially more in the same timeframe than those still fine-tuning the old.
Considering where to begin? We help identify the areas where agents deliver the greatest reduction in manual work.
Olli Maksimainen
Chief Operating Officer