Agentic AI is opening up a new chapter in the development of artificial intelligence (AI), moving beyond systems that simply respond to instructions towards technology capable of carrying out more complex tasks with greater autonomy.
As investment and interest in this field grow, organisations across different sectors are beginning to consider how these systems could fit into their operations. However, turning the potential of Agentic AI into practical applications requires more than technological progress. It also raises important questions about expertise, trust, accountability and the pace at which businesses and public institutions can adapt.
For companies working at the intersection of AI, geospatial technology and disaster management, these developments present both opportunities and challenges. Understanding how the technology can be applied responsibly, while ensuring that it delivers meaningful value, will be central to its wider adoption.
In this interview, Guy Schumann, CEO and Founder of RSS-Hydro, shares his perspective on the evolution of Agentic AI and what its growing capabilities could mean for the future of technology, business and public safety.
We’re seeing rapid advances in Agentic AI and large language models. What potential do you see from combining these technologies with geospatial data, and how could this transform disaster management?
I think these AI agents are tools that are almost like another person. You can build an agent to interpret information within your business and carry out tasks for you. The important thing is implementing it where you think it will be most useful, because you could build an agent for almost anything.
This is already transforming the geospatial sector. If you want an AI agent to work effectively and save time and money, you need to make sure it can interpret the information in your databases and understand the data provided by your suppliers.
We’re seeing platforms become increasingly agent-driven. You can upload an image, tell the system what you want to understand, and the agents behind it will interpret the image and extract the relevant information.
This is happening across different countries, with start-ups emerging in Earth observation. The European Space Agency (ESA) is also moving in this direction through its Agentic AI workshops and funded programmes. We’re seeing a significant push towards both digital twins and Agentic AI, in publicly funded projects and commercial industry.
I think the technology is already here and widely accepted, but it needs to be supported by the new technologies coming through. It will be transformative, and I don’t think we’ll have to wait long to see that.
Traditional disaster alerting often relies on historical data, forecasts and predefined warning thresholds. How could AI agents combine real-time information from satellites, ground sensors and Earth observation to predict developing hazards and provide more actionable insights?
There is a whole range of information that can be brought together. You have predictions, historical flood mapping, stochastic modelling to interpret risk and hazards, vulnerability curves and even parametric pricing models.
Imagine bringing all those data streams together in one place. That’s where platforms are heading. Organisations are building systems that allow customers to upload their own information while also accessing data from third-party suppliers.
These platforms are often designed for particular sectors. You have systems for insurance and reinsurance companies, banks, municipalities and defence organisations. The challenge is that someone still needs to interpret all that information.
You would need a person who understands practically every data stream involved. You could build a team of 50 people to access the platform and analyse the information, and the resulting analysis might be very good, but it would take time.
Now imagine having an Agentic AI model that is constrained to the resources and environment of a particular business or organisation. It can access all that information and produce a report on an approaching event, potentially within seconds. The report could be generated in any language and structured so that the decision-maker receiving it can understand the situation.
It might explain which areas are likely to be affected, how many previous events have occurred there, how many people could be exposed and what economic assets are at risk. That can now be done in around 10 seconds.
The agents would be developed by people and experts who understand the different data streams. There is no realistic way for 50 different specialists to produce one comprehensive report for senior decision makers in that amount of time.
The more you structure your data so that an agent can interpret it correctly, without hallucinating or making things up, the better the reporting becomes.
That’s also why we develop our PIN approach. We provide parameters and intelligence in the form of values, text and numbers. This is much easier for a large language model or AI agent to interpret because the information has clear boundaries. The numbers are already there, so the agent has less scope to invent them.
Businesses need to accept that these agents are coming. They are already being developed and integrated across industries.
As organisations embrace these tools, could they also free up employees to focus on other areas of their work?
Yes, and I think that’s wonderful news. Take our own company, for example. We officially have 15 full-time employees, plus contractors. For a company of our size, that’s a substantial team.
We could decide to employ an in-house accountant, for example, or someone to prepare reports for senior management covering what happened during the previous week, what was discussed in meetings and what is happening across the business. Hiring people with those skills can be expensive.
What if you could build a very good AI agent to carry out some of that work for a fraction of the cost? It would still involve an expense, but the people you might otherwise have hired could be brought into different roles, or existing employees could spend more time on other tasks.
I don’t think AI agents will reach a point where companies simply introduce them and fire people to replace them. I see them as a valuable addition. Businesses can use their resources to recruit elsewhere or allow existing employees to concentrate on work where they are more productive.
Agents could take on tasks that are important but that nobody particularly enjoys doing. We are certainly thinking about some of these applications ourselves.
For example, our technical team might be developing a service for a customer. I could ask them to summarise their work for the board, but why should they have to do that? They are trained to code and deliver services to customers, not necessarily to produce reports for senior management.
An AI agent could prepare that report instead. The company could then use the money saved to invest in other areas that support growth, rather than simply keeping it as additional profit. For a smaller company like ours, that could mean reducing costs in one area while creating opportunities to recruit and expand in another.
There are some roles where I think a human relationship remains important. I wouldn’t necessarily build an AI agent to act as an HR manager, for example. But there are other tasks where an agent could be very useful.
If you use an external accounting office, you still need to submit all the relevant paperwork, but an agent could potentially summarise your accounts and help prepare the information. It could also reduce the risk of errors or missed submissions.
For small and medium-sized enterprises, I think there are some fantastic opportunities to implement these systems.
Beyond identifying hazards, could agentic AI analyse developing situations, anticipate how they might change over time, and potentially issue warnings or initiate actions?
I think so, and it’s developing very quickly, particularly when it comes to actions.
When all these different data streams are available, AI can interpret them more effectively. It might inform a decision-maker that there is a certain percentage chance of an event affecting another location, even if that information isn’t necessarily contained in the original data provided to the system.
More importantly, it could help identify what actions should be taken. An agent could analyse the available information, produce a list of 10 possible actions and rank them according to what it considers the most appropriate response.
It could say: given the data available, these are the actions I would recommend. But ultimately, a human decision-maker must take responsibility for what happens. The responsibility and liability have to remain with a person.
It is difficult for someone to understand every possible action they could take based on complex information, particularly when they may not fully understand all the data involved. Having an agent provide that assistance could be extremely useful.
What surprises me is the speed at which these agents can now perform tasks that produce very reasonable results. They can help with warnings and prioritising actions, but those recommendations should remain suggestions.
The human decision-maker might choose the second option because the first recommendation doesn’t work in their particular country or situation. There are complexities around disasters that only people may fully understand, including cultural differences.
For example, an agent might recommend issuing a warning through television, but that might not be effective in a particular country because people do not watch or trust the news. Radio might be a much better way to communicate the information.
An AI agent may not understand those nuances. That’s why I think it is essential to retain a human being as the final decision-maker.
What are the biggest technical and practical challenges in developing agentic AI systems for disaster management?
The first challenge is understanding how to build them. These are very specific agents that need to be trained to understand the details of disasters and the different datasets involved.
For a company like ours, we have experts in early warnings, hydrology, modelling and simulation. But you cannot simply ask those people to build an AI agent. It doesn’t work like that.
You need someone who can actually develop the agent you require. One option is to work with external specialists who have some background in geospatial technology, disasters and modelling, but who focus specifically on building agents.
Companies offering these services are starting to emerge, and we are looking at those options. In the meantime, we’re exploring whether we can develop some agents ourselves to interpret results.
For a smaller company, it may not be possible to build everything in-house. You might need to rely on external expertise to make the technology work properly.
Another challenge is making sure the agents don’t go rogue. They need to be restricted to the resources available and the data they are supposed to interpret. If you give them too much freedom, they can become a liability.
I often compare them to children. If you give them too much freedom without proper guidance, things can go wrong. You need to train them properly so that they become part of the system you’re building rather than creating problems.
There are also legal obligations, liability and cybersecurity issues. Those are three areas where we need to be particularly careful when working with agents.
Despite those challenges, I think Agentic AI is bringing us some wonderful tools.
AI technology and regulation are both evolving rapidly. Is it difficult for businesses to keep pace with those developments and adapt?
I imagine it is difficult. Fortunately, regulatory issues aren’t my area of responsibility, but I think one thing that could help is starting to consider agents in a similar way to employees within a company.
I recently saw a post from someone talking about a new hire. When I read it, it became clear that the new hire was actually an AI agent. They had included it as part of their hiring structure.
I think that is an interesting way of looking at it. A company might decide it needs 10 people, including one person to carry out administrative work. Now, it could decide it needs one additional entity to do that work, and that entity is an agent.
There is a cost involved, just as there is with employing someone, although the money obviously doesn’t go into the agent’s bank account.
I’m not suggesting that agents should be treated exactly like people, but looking at them as another team member could make it easier for businesses to understand where they fit into their structure.
As the technology develops, how important will human oversight remain?
I think it is essential to keep a human in the loop because, ultimately, there needs to be a responsible person who can bear the consequences of a decision.
You can hold an agent responsible to a certain extent. If it doesn’t behave properly or carry out its tasks, you can remove it. But I don’t see how you can make an agent bear the consequences in the same way as a human being.
You can delete an agent, but you cannot take legal action against it in the same way you could against a person. There needs to be someone above the agent who remains accountable.
As the owner of a company, I see that responsibility in the same way whether an employee or an agent makes an error. I am still the person ultimately responsible for what happens. I think that distinction is crucial.
Moving on to your FloodPin service, it also explores integrating large language models to deliver personalised emergency messages in different languages and dialects. How important is it to make this information specific to different areas, and how could it improve public safety?
I think the best way to improve public safety is to integrate the information into things that people already use every day.
I’ve been thinking about how we could incorporate it into everyday devices, whether that’s a television, a household appliance or, particularly, a car.
I would like to see people able to access geospatial information about hazards, risks and predictions as part of their everyday journeys. Whether they are driving or using another form of transport, there is an opportunity to provide warnings through systems they already interact with.
Modern vehicles are increasingly guided by AI through their navigation systems. When your car speaks to you, it is already using technology to communicate information and provide guidance.
We could potentially integrate our PINs into those systems. They are short, message-based pieces of information, which is important because you cannot load gigabytes of geospatial map data into a vehicle’s navigation system.
Instead, you could transmit small amounts of information, potentially in real time, or preload messages that are then interpreted as the vehicle travels.
If the information is predictive, it could also be streamed directly into a connected vehicle. I think that is a very exciting possibility because it would put public safety information into something people use every day.
We aren’t there yet, but we’re moving in that direction and are talking to people about how we might open up those possibilities.
FloodPin focuses on flooding, but the same approach could be applied to other disasters, such as wildfires. Could we eventually see universal AI-powered public safety platforms, and what role does RSS-Hydro hope to play in their development?
I think it would be great to see that happen, and it would be even better if we could play a part as one of the early adopters and developers of that technology.
I don’t necessarily need RSS-Hydro to be fully involved in a public safety platform. It would be nice to be recognised as one of the companies that helped start something that could be genuinely game-changing.
I do think there is a place for AI and AI agents in public safety platforms. We already see these technologies being used in defence, and I think we will see them increasingly applied to public safety.
However, we need to be a little more careful in this area because we’re dealing with civilians who need to be protected and kept safe. That’s different from maintaining military or defence capabilities.
As we develop digital twins for public safety, which is something we’re already working on, I think the inclusion of AI agents will become inevitable.
At the EU level, do you think these technologies are being adopted, and are policymakers keen to see them developed and scaled?
I think so, definitely. We’re already seeing chatbots being adopted as a form of AI assistance, helping people answer questions and access information. I think that will expand very soon into agents carrying out more serious tasks.
I’ve seen the beginnings of that development, and I think adoption is gathering pace. However, there are still regulatory issues that need to be addressed, which can slow things down.
We’re seeing a much faster pace of technological development than adoption. Adoption requires recognition and regulation, and those things are not necessarily moving at the same speed.
It’s not simply a case of buying a new iPhone. It’s buying a new iPhone with agents inside it. People need to understand what that means and how to use the technology.
Some of these developments are moving faster than people can adopt them or even become aware of them. That gap between technological progress and wider adoption is something that will need to be addressed.
