The Real Knowledge Base · from our own projects
Which tasks do you leave to AI, and which stay with your team?
Choose per task, not per tool. There are three ways to handle a task. AI does it on its own, AI supports your team, or the work stays with people. Four questions guide that choice. How often does it recur, what is technically possible, what does a mistake cost and what can your team handle? What fits your brand and whether it is good enough is always decided by a person.
Agentic and augmented, what the difference is
You hear both words more and more often. With augmented AI, AI supports your team, including helping to carry out the work. Your team makes the decisions. With agentic AI, an AI agent carries out a task on its own, within agreed limits. They are not two kinds of AI. It is about how much room you give AI. The same task can first run with support and later on its own.
Three ways to handle a task
In our own company all three sit side by side.
- On its own. Our ads are monitored, analysed and improved every day by our own AI.
- Supporting. Our AI recognises website visitors, qualifies them and prepares a draft email. The groundwork is done. A person chooses what is sent.
- With people. What fits a brand, and whether work is good enough. At most, AI helps explore.
It can also differ within one process. Our bookkeeping runs without data entry, and we only check the exceptions. So choose per action which one may run on its own and which needs a review.
Four questions per task
- How often does it recur? Work that recurs often in a fixed form suits the machine. Something that happens twice a year usually does not.
- What is technically possible? Some tasks AI cannot yet do well enough. In those cases, AI can only assist.
- What does a mistake cost? If a mistake could reach your customer or affect your brand, a person must check the work first. What you can fix without harm, the system may do.
- What can your team handle? Your team must be able to judge the outcome and keep the system up to date. If AI takes over more than your people can follow, you lose control.
None of the four decides on its own. For recurring tasks where mistakes can cause serious harm, AI remains in a supporting role.
How much room AI gets
The right level of independence for AI depends on proven quality and the consequences of mistakes. AI should only work independently where your team can recognise and fix mistakes in time. Any content that represents your brand is always reviewed by a person before it goes out.
Where it goes wrong
- Independence as a goal. More room for AI is not always better. It depends on what a mistake costs.
- Letting go too early. An agent that has not yet proved itself makes mistakes that are noticed late.
- Not counting the checks. Even with a final check, AI can save manual work. Count that check before you choose how much to let go.
- Handing over judgement. Whether work is good enough is decided by a person. That is a design choice, not a consequence of the technology.
What this means for you
For recurring processes such as quotations, administration and customer correspondence, the future-proof scan looks at what is ready for AI. For your creative team, the CID does this in the assessment. It examines what your people already do with AI and how people and AI divide the tasks.
Frequently asked questions
What is the difference between agentic and augmented AI?
With augmented AI, AI supports your team, including helping to carry out the work, and your team decides. With agentic AI, an agent carries out a task on its own, within limits you set. You choose the form per task.
Which tasks do you give to an agent first?
Actions that recur often, have a fixed form and where you spot and fix a mistake in time. Within a process, choose which action may run on its own and which needs a review.
Who supervises when AI works on its own?
Your team, with regular spot checks and an owner per process. Agree what that owner does when an outcome deviates.
Sources and accountability
- This article describes our own way of working, as set out on our CID and TID pages.
- The examples of independent and supporting tasks come from our own systems on the TID page.
- What agentic and augmented mean is our explanation.
- Our advice on these trade-offs is based on this way of working.
- These sources give no measured result for how much time each form saves. That is why there is no figure here.
Which task are you unsure about?
Tell us where your team gets stuck. We discuss whether the future-proof scan or the assessment is the right first step for you.
Talk to us