Article content
In briefShow moreShow lessWe are moving from generative AI that answers questions to agentic AI that independently plans and performs tasks to achieve a goal.
- We are moving from generative AI that answers questions to agentic AI that independently plans and performs tasks to achieve a goal.
- The trend is moving toward integrated solutions in which multiple specialized AI agents collaborate as digital employees.
- Sjøholt recommended starting with one task and checking whether the agent handles it well.
The story
We are moving from generative AI that answers questions to agentic AI that independently plans and performs tasks to achieve a goal.
What this means for the work
The trend is moving toward integrated solutions in which multiple specialized AI agents collaborate as digital employees.
Many AI initiatives fail because they lack a clear business rationale, not because the technology is immature.
The announcement or interview records the capabilities described at the publication date. NIST’s January 2023 AI Risk Management Framework recommends governing, mapping, measuring and managing risk throughout the lifecycle, providing a more concrete frame than broad productivity promises.
In December 2023, OpenAI argued that agentic systems need clear responsibilities and safety practices. Anthropic’s 2023 Responsible Scaling Policy similarly tied stronger capabilities to stronger safeguards. Greater autonomy requires greater control.
Leaders should define the task, data access and permissions before an agent is tested in operations. Actions should be traceable and stoppable, with uncertain cases escalated to a person. Benefits and failures need to be measured in the real workflow.
Sources and perspectives
Primary source: Jets Group sparer minst 400 timer i året med AI-støttet ordrebehandling
NIST: AI Risk Management Framework 1.0 (26 January 2023)
OpenAI: Practices for Governing Agentic AI Systems (14 December 2023)
Anthropic: Responsible Scaling Policy (2023)
Practical follow-up
Sjøholt recommended starting with one task and checking whether the agent handles it well. The organisation must decide which data it can use, which actions it can take and when a colleague should step in. The interview offers implementation advice without quantifying its effect.
Join the discussion
What would you investigate first before using this? Share your questions or experience in the discussion.
