AI Agents in Practice: How Henrik Kniberg Sees the Future of Collaborative Work - Product at Heart 2025
When working with interns, the experience usually varies from “Please take a break. It’s better if I do this by myself.” to “I want to have you as a full-time employee right now!”. Imagine a world where the latter is the case for every intern you hire. Every intern super-smartly executes according to your guidance. And they are available for anyone in need. This is the world we already live in! Let me explain why.
At Product at Heart 2025, Henrik Kniberg, an AI whisperer, trainer, product developer, and author who created the viral video "Generative AI in a Nutshell" and the book with the same name, explored practical approaches to effectively leveraging AI agents. He provided a positive outlook for a future where teams of humans and AI agents collaborate seamlessly.
In Between Code and Humans
AI agents are autonomous entities characterized by combining Large Language Models (LLMs) with clearly defined missions, autonomy, and specialized tools. They sit at the intersection of traditional code (fast, predictable and not intelligent) and humans (slow, unpredictable and intelligent).
Instead of seeing this as a competition, Kniberg highlighted how integrating code, agents, and human capabilities creates powerful, synergistic systems.
Practical Demonstrations of AI Agents
To ground concepts in reality, Kniberg presented practical examples, notably demonstrating an Invoice Router Agent developed with abundly.ai and designed to monitor emails for invoices, analyse and categorize them efficiently to route them towards the right department and flag those needing human attention.
The demonstration vividly illustrated how AI agents could streamline traditionally tedious workflows, drastically improving productivity and accuracy. Here is how Kniberg outlined the steps:
Clarifying the Mission
He first set the agent’s mission clearly by explaining the situation and what he needs help with. That conversation resulted in a clear mission: classify incoming invoices automatically, identifying invoices that need immediate human attention or further investigation.Enabling Capabilities and Autonomy
Kniberg then enabled the agent to access and interpret invoice data. He emphasized striking a balance between granting autonomy (allowing the agent to act independently) and maintaining human oversight (ensuring accuracy and reducing risk).Training and Fine-tuning
One major advantage of AI agents is their ability to absorb internal knowledge quickly. You can provide them with documents like invoice processing guidelines, and they’ll read and follow them - no onboarding sessions required. Recognizing the importance of context-specific knowledge though, Kniberg explained that initially, the agent - much like a knowledgeable intern - required feedback and fine-tuning.
By implementing this Invoice Router Agent, Kniberg demonstrated a substantial reduction in manual effort, faster processing, and more accurate categorization of invoices. Employees, relieved from repetitive manual sorting, could focus their efforts on strategic, high-value tasks requiring human intelligence and judgment.
In summary, the Invoice Router Agent example provided a compelling demonstration of how product managers can practically and effectively integrate AI agents into business workflows, enhancing both productivity and employee satisfaction.
Identifying Optimal Use Cases
One of the most critical takeaways: the importance of identifying tasks suitable for AI automation. Kniberg introduced a clear "Automatibility Scale," ranging from highly predictable tasks (like payroll calculation) to unpredictable tasks (such as writing complex business proposals). Kniberg pinpointed the sweet spot for AI agents-workflows that are repetitive, time-consuming, yet not overly complex, such as support ticket classification or investment screening.
His investment screening example, in particular, demonstrated remarkable efficiency gains, reducing manual effort by 95% and freeing teams for more strategic tasks.
Essential Agent Design Principles
When designing an AI agent as a Product Manager, consider the following actionable tips:
Think of agents as extremely knowledgeable interns: AI agents possess significant general knowledge but require specific context and iterative human feedback to excel. Ask AI for feedback if you get stuck.
Encourage human-agent collaboration: AI agents perform optimally in shared spaces with humans, clear task delegation, and aligned goals. Ask them what they did, review and fine-tune their process for optimal output.
Prioritize iterative development: Building effective agents requires continuous iteration. A minimally viable agent can be deployed quickly but requires ongoing refinement.
Balance power and responsibility: Greater agent autonomy brings higher rewards but demands rigorous testing, careful prompting, comprehensive monitoring, and appropriate guardrails to mitigate risks.
Limit complexity: Similar to humans, AI agents perform best when they aren't overloaded with excessive instructions or tools. Keep it simple and rather design multiple agents (that could talk to each other) with focused tasks.
The Agent Design Canvas
Struggling to get started? Don’t worry, there is a canvas for designing agents - the Agent Design Canvas, which provides product managers and teams with a structured framework for designing and developing AI agents effectively.
This visual template helps to define an agent's purpose and mission clearly while systematically addressing key areas, including triggers for action, required knowledge and context, intended outputs and deliverables, necessary tools and integrations, risk mitigation strategies, and essential human-agent collaboration points.
By explicitly outlining success metrics and indicators, the canvas ensures ongoing evaluation of the agent's performance and value contribution. This methodical approach not only streamlines the design process but also encourages holistic thinking, minimizing risks, and enhancing clarity around roles and responsibilities between humans and AI agents.
Embracing a Collaborative Future
Kniberg's keynote underscored that the integration of AI agents within teams represents not a replacement of human roles, but an augmentation. As a Product Manager, you should view AI agents as collaborative partners, enhancing productivity, creativity, and strategic capacity. If you do that, you’re not just automating tasks - you’re amplifying your team’s potential.
Ready to bring your own “super-smart intern” onboard? Pick one repetitive workflow from your backlog: support ticket triage, release notes, or internal FAQ creation. Then sketch your first agent using the Agent Design Canvas. You don’t need to be an AI expert to start. Curiosity and a collaborative mindset works perfectly.
Remember, it’s not about choosing between humans and machines. It’s a symbiotic partnership between them, unlocking unprecedented levels of innovation and efficiency in Product Management. A system where both can thrive - together.
