Skip to content
Agentic·Foresight

Questions · Glossary

Questions on Agentic Foresight

Short answers first, explanations after. The key terms are collected in the glossary at the end.

01 / Questions

Ten questions, ten answers

What is an AI agent?

An AI agent is a program that breaks a goal down into steps on its own, uses tools to carry them out and checks its intermediate results until the task is done.

At its core there is usually a language model. Around it sit tools such as search, databases or files, a memory of where the work stands, and rules for when the agent stops or asks a human.

How is an agent different from a chatbot?

A chatbot answers a question; an agent works through a task.

The chatbot waits for the next input. The agent plans, calls tools, reads the results, corrects itself and delivers a piece of work at the end, for example a list of verified signals with their evidence.

What is Agentic Foresight?

Agentic Foresight is foresight work in which specialised AI agents search for signals, build scenarios and test strategies against them, while humans ask the questions, judge and decide.

The three agents are called Scout, Outlook and Tunnel. Each covers one stage of the textbook sequence and hands its result on to the next.

How does it differ from classic and from continuous foresight?

Classic foresight is a project, continuous foresight is ongoing scanning, and Agentic Foresight is an ongoing process in which agents also keep scenarios and strategy checks up to date.

In a classic project, experts do almost everything themselves. In continuous foresight a team scans constantly, but scenarios are only revised at intervals. In Agentic Foresight, agents take on research, scenario building and the strategy check; humans ask, judge and decide.

Do agents replace foresight experts?

No; they shift experts' work from producing material to judging it.

Expertise is needed at different points: framing the question, selecting the drivers, checking plausibility and translating scenarios into decisions. This division of labour is called Flipped Foresight.

What can AI not do with the genuinely new?

AI struggles to recognise the genuinely new because it has not yet left traces in the data the AI learns from and searches.

What is stable can be extrapolated; what is changing leaves signals. What does not exist yet has to be considered possible by a human before an agent can look for it. Language models also lean towards the probable and smooth out deviations.

What data do the agents need?

Scout and Outlook work with publicly available sources; only the strategy check in Tunnel needs internal information about your own organisation.

That includes the business model, resources, dependencies and the assumptions the strategy rests on. The more precise this model, the more meaningful the check.

How is confidential information kept confidential?

By processing internal data only where the organisation decides, if in doubt on its own infrastructure.

Public research and the internal strategy check can be separated technically. For the confidential part, options include your own servers, models with contractually governed data processing, or locally run models. Which option fits is the organisation's decision.

How do you get started in an organisation?

With one concrete strategic question and one lever, usually Scout on a narrowly defined search field.

Once the signals prove useful, scenarios with Outlook follow, and then the strategy check in Tunnel. The system grows with experience instead of starting as a large project. How Operation Zukunft builds agents for innovation and foresight (in German).

Who is behind Agentic Foresight?

Prof. Dr. Ulf Pillkahn, professor at FOM University of Applied Sciences and lecturer at TU Munich.

With Operation Zukunft he develops AI-supported methods for corporate foresight and scenario development and tests them in his own agent cluster. More about Prof. Dr. Ulf Pillkahn.

02 / Glossary

Terms

Signal
An observation with evidence that points to a possible change, such as a study, a draft law or a product announcement. Weak signals are early, isolated hints.
Driver
A factor that shapes the future in the field under study, for example energy prices, regulation or customer behaviour.
Key factor
A driver with high impact and an open outcome. Key factors span the scenarios.
Impact matrix
A table in which every driver is rated against every other. Active and passive sums show which drivers push and which are pushed.
Wilson matrix
A grid that places drivers by impact and uncertainty. Drivers with high impact and high uncertainty set the scenarios apart.
Scenario framework
The combination of key-factor states from which the scenarios are built, often a scenario cross with two axes.
Strategic fit
How well a strategy matches an environment. In the wind tunnel it is tested for each scenario separately.
Early-warning indicator
An observable measure that shows early on whether the world is moving towards a particular scenario.
Agent
An AI system that breaks a goal down into steps on its own, uses tools and delivers a piece of work.
Flipped Foresight
The division of labour in which machines design futures and humans decide: producing material goes to the machine, human time goes into judgement and decision. The method at a glance.

Published: 16 November 2026 · Last updated: 16 November 2026