The pace at which artificial intelligence has evolved and been adopted has been truly remarkable over the past three years. Just look at AI’s penetration into popular culture: seemingly everyone has heard something about it, something which marketers can only dream of – believe me, I have spent the last two decades failing in my attempts to explain managed file transfer (MFT) to friends and family!
Businesses and individuals alike have immediately been able to understand how this technological leap forward could improve productivity and innovation to levels that are sometimes described as the next big leap after the industrial revolution of the 1800’s.
In fact, belief in its abilities is such that 42% of enterprise-level businesses report that they are actively using AI, and another 40% are exploring its potential. Perhaps even more telling, 90% of businesses view AI as crucial for gaining an advantage over competition.
See also: Enterprise AI Implementation: From Hype to Real Business Results
And yet, it is suggested that we are very much at the beginning of AI development and the realization of how we might benefit from it. Today, the world is excited by the use of generative AI — a version of AI that creates text, videos, imagery, and even music based on prompts, questions, and conversations with human input.
However, it is the future of agentic AI which is even more promising.
What is agentic AI?
In plain speak, agentic AI refers to AI in the form of an autonomous agent. Rather than acting on the prompt of a human, AI would be given access to one or more applications, network devices, or services, along with some rough guidance on the boundaries of behavior.
The agent would then manage what it has been given responsibility for autonomously and without human intervention, unless conditions breach the defined behavioral boundaries.
The expectation amongst IT leaders is that in future, multiple AI agents will look after large parts of the IT ecosystem, communicating with each other and dynamically reacting to conditions to ensure that the network is optimized and meeting business needs. In this scenario, humans would only respond to those outlying moments in which they are absolutely needed.
This coming change has been acknowledged by Microsoft CEO Satya Nadella as recently as January 2025, in which he said that traditional business and SaaS applications will fundamentally change in the agent era. And that these agents will operate across multiple apps or databases, independent of the back-end systems, coining it the “AI Tier”.
Comparing generative AI and agentic AI
So how is this any different from the tools we have today?
Compared to agentic AI, in which there is less human intervention and more autonomy, the world of generative AI is very subservient to humans. Generative AI simply creates things when asked by a human, whether that be:
- Text
- Responses to questions
- An analytical answer
- Am image or video
- Programming code
It can provide these responses due to its ability to consume huge amounts of pre-existing data, often from the internet. In effect, generative AI is an upgraded version of a search engine, able to join complex topics, consolidate multiple results, and reproduce content by combining sources.
Interestingly, it never really creates anything new. It relies on the existence of something(s) which it has consumed beforehand to generate its output – and as many are learning today, generative AI’s output is only as good as what it has been trained on.
Generative AI may seem like it can produce something novel, but the puzzle pieces which make up its perceived novel output already exist – they just benefit from advanced processing power and pattern –matching, giving the illusion of creativity.
| Generative AI | Agentic AI |
| Produces something based on data sources it has learned from Relies on human input to execute Statically returns a response to a human prompt | Truly dynamic, continuously learning from its environment Executes autonomously within human-defined boundaries and without prompting Adapts to conditions Able to work with other AI agents for coordinated outputs |
Today, generative AI has already started making its way into IT and managed file transfer technologies. MFT can benefit from generative AI, primarily as a source of information, help, and onboarding.
Examples include:
- Chatbots which have consumed product documentation and are able to answer configuration questions
- Tools which create automated workflow patterns using the most optimal model
- Simplified onboarding processes by pre-creating accounts or prospectively setting up client side authentication objects like certificates
As is characteristic of generative AI, these examples require human prompting and return answers based on materials they have been trained on.
See also: The Evolution of Managed File Transfer AI: Automation to Autonomy
How could agentic AI be used in MFT?
Like most operational IT, MFT stands to be a significant beneficiary of agentic AI. While IT and automation has helped to improve the velocity and accuracy of business operations of the past four or five decades, AI will take this further by removing some of the productivity bottleneck: the human.
Here are four ways agentic AI could be used in managed file transfer:
Autonomous scaling
Today, we largely build virtual environments using fixed levels of resource and capacity, based on minimum software requirements or the level of demand we expect to see.
In an agentic AI world, resource and capacity would be relative to the demand at a particular point in time, and could be adapted and changed dynamically as conditions change or where there is a predicted need.
For example, if an AI agent knew that important large files were transmitted at a period of high traffic, it could dynamically alter the file transfer environment to prioritize and dedicate more resource to this activity, ensuring faster transmission and processing – then return to a base level once complete.
Intelligent routing
Following on from the example above, AI agents could also dynamically route files according to prioritization, transmission speeds, and to avoid interruptions and problems caused by other changes or degradation in the network.
Where agentic AI has access to multiple environments and solutions in the network – such as MFT software and network devices – it could result in collaborative changes which have not been seamlessly achieved before.
File classification and resultant behavior
Today’s file classification tools are mostly reliant on human-led processes. But given the right guidance and boundaries, agentic AI can classify data and files with a high degree of accuracy — just as humans do, based on an understanding of category rules.
Agentic AI could bypass the human element in most cases and do this autonomously, reducing errors and catching forgotten or misclassified files. It can also intervene when classification needs to be updated after changes.
This improves follow-up actions tied to each classification level, such as protective controls, prioritization, and tracking.
Vulnerabilities and configuration issues
Change control, by its very nature, slows the process of change down to reduce the negative impact of changes, ensure that compliance to standards is maintained, and pay due consideration to business operations. It is human-intensive and deliberately not dynamic. But what if it could be?
Consider a scenario in which a vulnerability had been detected in an SSH cipher. Today’s processes would dictate that someone be assigned the task of discovering whether that cipher is in use; submit a change request to remove it; once approved, remove it; and once removed, monitor the impact.
Agentic AI would be able to do all those steps instantly, including detecting any vulnerability disclosures, and reduce attack surface immediately while monitoring and notifying about any impact.
Will agentic AI take our jobs?
All this talk about speed, effectiveness, and the lessened requirements of people when using agentic AI certainly invokes fear of mass replacement of IT operations with machines – and there may be some truth to it.
Agentic AI doesn’t mean the wholesale removal of the people from operations. There is still a human need to define working parameters and to respond to scenarios within which the parameters the agent can work are breached. But the need for humans in IT operations will be reduced.
Ultimately, AI augments human capabilities by making us more productive and, in the short term, an increase in productivity will mean that the output of multiple people can be reduced to one.
Whether or not this is an immediate reaction to new technology, or if in the longer term a new baseline of productivity will mean greater business operational requirements and an increase in humans alongside AI agents in response, is something we will have to see with time.
What is clear is that while there may be some disadvantages, we sit on the precipice of a huge change in IT operations; and as business leaders, consumers, and human beings, we are set to benefit from an age of rapid technological advancements not seen by humankind so far.
Frequently asked questions
How is generative AI being used in IT operations today?
Generative AI has wide opportunities and usage in IT operations today, with 42% of enterprise businesses in 2025 making use of AI according to GPTZero.
Much of the use is centered on analytics and insights into large data sets, such as chatbots which can return answers to natural language questions found in business document libraries; or speeding up content creation (whether it be pitch decks, sales enablement materials, or public facing blogs and brochures). It is important to note that generative AI returns a result based on human invocation or inputs, and therefore functions in an assistive capacity.
Why is agentic AI important for IT and MFT operations?
Where generative AI creates based on existing data, agentic AI autonomously takes actions based on the conditions and events it detects. If an AI agent was able to access an MFT solution through a fully featured set of REST APIs, it would be able to effective control, run and adapt that solution as needed, improving its efficiency.
If the agent had the ability to connect to dependencies, upstream/downstream services and network devices, it could take parallel actions across the ecosystem to achieve its aims. Agentic AI improves the efficiency of MFT but also helps to solve the challenge of maintaining MFT skills, by replacing the need for a 24-hour worker.
How does agentic AI behave in critical situations, which haven’t been learned or defined?
The ability of agentic AI is not limitless. It requires less human input, but it acts within a boundary defined by its owner. With curtailed power, its ability to make devastating changes accidently or in response to a false positive are limited.
In scenarios outside of its programming or defined thresholds, it should flag for human guidance on how to proceed. However, as more extreme circumstances are discovered, and if it can learn in these situations, it will become decreasingly dependent.
How likely is it that the entire MFT operations requirement is handed to an autonomous AI agent?
It is hard to say, and it depends on the appetite of the business. The rapid development of AI comes with an assumption that everything developed will be adopted. Of course, this is not entirely the case, and handing over swathes of operations to an unnamed entity raises questions around security and trust – both areas which are slowly developing fields in AI.
Businesses want to adopt AI on the basis of promised cost-savings and efficiency, but at what risk is still to be determined.
How are MFT solutions preparing for the world of agentic AI?
Most IT vendors are clamoring to add AI capabilities into their solutions, in some cases building their own agents which act in isolation. I believe the real future lies in agents which can manage an array of solutions and tools across the IT estate, with each of these acting like tools in the toolbox of an agent.
Most MFT solutions today have REST API endpoints which can be used to manage and monitor the solution. It is the enrichment of these APIs and the creation of a headless architecture which will best serve the immediate- and medium-term future of agentic AI.