When should an AI system be allowed to fix an IT problem without waiting for a human to approve the action?
In this episode of Tech Talks Daily, I speak with Matt Tuson, General Manager for Europe, the Middle East, and Africa at LogicMonitor, about the practical path from reactive IT operations toward autonomous IT. The promise is attractive: fewer repetitive tasks, faster incident resolution, less disruption, and systems that can correct familiar problems before users feel the impact. The difficult part is deciding when the available data, controls, and evidence are strong enough to trust an automated action.
Matt argues that observability is taking on a different responsibility. It once helped engineers answer what happened and why. In an autonomous environment, it must also give AI enough connected context to recommend or execute what should happen next. That means understanding how infrastructure, networks, cloud services, applications, and other operational signals relate to one another. A fast decision based on an incomplete view can resolve the wrong symptom, create duplicate incidents, or make the original problem worse.
Trust therefore begins with the information supplied to the system. Many IT environments contain separate monitoring and management tools introduced by different teams to solve specific problems. Those specialist products may continue to do useful work, but their signals often remain disconnected. Matt describes a business combining nine organizations, each bringing its own tools and operational practices. The challenge was not simply the number of products. It was the inability to connect their outputs into a reliable view of what was happening across the combined environment.
This matters because AI can process poor information as quickly as good information. Matt warns that adding autonomous remediation to incomplete, duplicated, outdated, or poorly contextualized data risks creating automated confusion. False positives multiply, root causes are missed, and teams repeatedly address symptoms while the underlying fault remains. His advice is to start with the business outcome, identify the information needed to support that outcome, and improve the quality and consistency of the data before increasing AI authority.
We also discuss how leaders can decide which actions belong with AI and which should remain under human control. Matt rejects the idea that organizations must choose between full autonomy and complete human approval. He recommends graduated levels of authority based on the action, confidence in the recommendation, business impact, and whether the change can be reversed easily. Ticket enrichment, alert correlation, and routine remediation may offer lower-risk starting points. A change affecting a major service should receive closer human oversight until the system has established a reliable record.
That creates a practical model for earning trust. If an AI recommendation repeatedly matches the action an experienced engineer would have taken, the organization gains evidence that it may be safe to automate that decision. The process can move gradually through support levels and operational complexity, with explainability, governance, and audit records maintained throughout. Human judgment remains where accountability and business consequences demand it.
Matt also addresses the difference between companies making progress with AI and those stuck in pilot purgatory. The stronger performers begin with an operational problem, such as reducing incidents, improving uptime, or accelerating resolution. They invest early in data quality, visibility, collaboration, and governance. AI becomes part of an existing workflow with a defined path from insight to action, rather than a detached experiment looking for a reason to exist.
For IT leaders who want greater autonomy but do not yet trust their environment, Matt's starting point is simple: establish broad visibility, improve operational data, connect isolated sources, and define the actions AI can take independently. Teams must also be able to understand why a system reached its conclusion and inspect what it did afterward. Autonomy should grow through repeated evidence, not through hope or one large leap.
There is a commercial reason to get this right. Faster resolution can reduce the cost of service, protect customer experience, and limit the business impact of downtime. There is also a human benefit. Engineers spend less time sorting duplicate alerts or joining war rooms to prove which team was innocent. They can concentrate on decisions that require experience, judgment, and knowledge of the business.
As observability starts supplying evidence to machines as well as people, what controls would make you comfortable allowing AI to take action inside your IT environment? Listen to the episode and share your thoughts.
[00:00:00] Do you need AI agents that you can trust? Well, with an AI data layer providing real-time connection within your data platforms, you can trust your agents to provide accurate solutions. So, scale your business by trusting your agentic AI accurately getting the work done for you. Trust its capabilities with Denodo. And you can do that by simply visiting denodo.com to learn more.
[00:00:26] When should an AI system be trusted to fix an IT problem without waiting for a human to approve the action? In this episode today, I'm going to be speaking with Matt Tooson, General Manager for Europe, the Middle East and Africa at LogicMonitor.
[00:00:47] We're going to be talking about what must happen before autonomous IT becomes a responsible operating model, rather than just another ambitious AI promise. And he will explain today why observability now needs to provide evidence for machines as well as people. And we'll also discuss how fragmented monitoring data can actually lead to automated confusion at machine speed.
[00:01:15] But also why different actions require different levels of authority. And if we've got time, we'll also cover tool sprawl, human oversight and the operational signs that indicate an organisation is actually ready to automate further. So, anyone who has spent an evening in a crowded incident war room might find his mean time to innocent observation painfully familiar.
[00:01:43] But enough from me. Let me introduce you to Matt now. So, thank you for joining me on the podcast today, Matt. Can you tell everyone listening a little about who you are and what you do? Of course. First of all, thank you very much for having me. It's always good to have a chat. Brief introduction about me. I am a General Manager of Europe, Middle East and Africa for LogicMonitor.
[00:02:08] As a company, helps organisations understand what is happening across complex IT environments. So they can identify and resolve issues quicker, faster and more intelligent. And what we really do, is the way I try and look at it, we see it, we predict it, we fix it, and then provide outcomes to make sure it doesn't happen again.
[00:02:35] And that's quite a cool thing, especially in this modern world. And as powered observability is now becoming more of a thing, we are leading in taking people through the journey to autonomous IT. And autonomous IT is really about ensuring that as we learn about things that we fixed before,
[00:03:04] we might as well say, if that's happened again, let's just go and fix it without the need of a human in the loop. Companies really are in the process of making those business decisions. I just was in front of a company this morning, and we were talking about where their risks were associated to what their thoughts are about autonomous IT. And they want to take it as far as possible to a point where their risk factor is OK.
[00:03:33] And then once it hits that risk factor, then put a human in the loop. And that really, why are we doing that? It saves companies a lot of money. It enables us to provide fixes faster and quicker, therefore making companies money. And obviously, we want to reduce risk and protect the brand, which is fundamentally what we're trying to do as an organization to ensure that everything is up all the time.
[00:04:01] Thank you so much for sitting down with me today. And I think looking back over the last 10 to 15 years, IT has been on a journey of its own from being known as the blockers that say no far too much to business enablers that they're seen as now that every organization needs, of course. And the reason I bring this up is when doing a little research on you, I was reading you said trust is now becoming one of the biggest blockers in IT operations.
[00:04:28] So what is it that IT teams are increasingly reluctant to trust? The data, the automation, the AI making decisions? Or is it just the ability to understand what is happening across all these increasingly complex environments? What are you seeing and hearing here? It's really interesting. And obviously, AI has been front and center in conversations around the globe, including Donald Trump over the last week or so.
[00:04:57] Trust is really important in every single way. But if you think about technology, 25 years ago, I was selling technology that would automatically respond to an email that a customer had sent with a prescribed answer to service that customer and provide them a great near-term experience with that company. Now, don't you think it's amazing?
[00:05:26] And that was available 25 years ago. And we're only just starting to put this in the back office of IT operations today when the technology has been around for so long. And yes, it's improved. And yes, we're getting smarter. And there's more data than ever. But I think it's incredibly, it just amazes me that we can treat our most important, which is our customers in a way which is automated.
[00:05:54] And yeah, we've had automated voice in call centers for years to where we are now, where we are starting to do autonomous IT based on where AI is in the back office IT operations. Now, why do we think that is? Because if there's a mistake, it's not just one customer. It's multiple customers. And therefore, it's all about protecting the brand.
[00:06:23] There's obviously security concerned at the back end, which is all again, we don't want to be hacked. We don't want malware. We want to make sure that we are running smoothly to protect the broader thing that is our brand and our company.
[00:06:39] But then today, as companies are starting to use agents in every sort of part of their business, it is a little bit of a, Claude does this, Chachi Peter does this for me. People are doing their own stuff. And there is this little bit of a lack of control in an organization. So there has to be governance.
[00:07:05] And I think there isn't a distrust of AI once the data is cleaned and correct. It's where that AI can see the information it is using and then create the actions it needs based on the data it has. Now, what that means to me is you've got to get the data right.
[00:07:27] And getting the data right across multiple clouds, different premises, SaaS platform, dozens of tools, it's fragmented. And confidence in those decisions, because you're not getting the complete view of the world, is more complex. IT teams are unlikely to give AI more authority until they can understand the evidence behind its recommendations or actions.
[00:07:56] Therefore, it's all about proving that you can trust the data and the results and what AI does. And therefore, then take the human out the loop. And I think that's really important is you have to trust the results because your brand relies on it. And we've got to make sure that reliable data is doing what it's meant to be doing, continuing that we get the decisions right first time.
[00:08:26] So, in the simple terms, you can't build trust in an AI system if you don't trust the information it's working from. And ensuring that you have as much information as possible to make intelligent decisions is the most important thing. It really is. And you made so many great points there.
[00:08:49] And if we look at observability, for example, it was traditionally about helping humans understand systems and troubleshooting problems. But as we move towards this AI-driven and autonomous operations that we're seeing here, does observability become something different, essentially providing the evidence machines need before they can safely act? Is there a big change here? It feels like there is.
[00:09:14] I suppose historically, observability helped humans answer what happened, why. We've all seen the huddles. We've all seen the war rooms. Everyone's on at night. It's not my fault. It's their fault. Everyone's fault. It's somebody's fault. And everyone wants to, I think a few people have vented this term to me, a mean time to innocence. It wasn't me. It was somebody else.
[00:09:40] So, yeah, today it's increasingly helping AI answer what should happen next. So it's not what happened and why. It's based on a business outcome of what should happen next. And if you focus on business outcomes rather than technology, I'm a massive believer in the technology will drive those business outcomes.
[00:10:04] But you're focused on what the outcome should be and not just what technology is. So in a more autonomous environment, observability becomes a foundation of trust that is built.
[00:10:19] And that foundation of trust is the most important thing because the amount of data that we can see enables us to make really intelligent business decisions and recommend or execute on actions that the AI agent can do.
[00:10:40] And therefore, not only do we need to get the data, we need to know how all the areas of the complex IT infrastructure, network, cloud, etc., application services, how they interact. And therefore, we understand the bigger picture and broader picture to enable the right resolution and outcome for a business.
[00:11:11] So before AI can act with confidence, it needs to be able to see and observe and understand with confidence to make those critical business decisions that provide the best outcomes to serve a company in the way they'd expect, which is make money, save money, protect the brand.
[00:11:38] Okay, so many good points there, especially around the example of an IT team when things go wrong. It does feel like that Spider-Man meme with all these identical Spider-Men pointing to each other from the infrastructure, operations, applications, development teams, all pointing at each other when something goes wrong. And of course, every organization will have a different appetite for automation and indeed risk.
[00:12:02] So for leaders listening, how should they decide which IT decisions should be handed over to AI, which require human approval and which should remain firmly under human control? Because I think this is something that many teams are wrestling with at the moment. So I've been speaking to six customers this week alone, proper conversations about where they see their future. And it really does depend on the risk profile.
[00:12:32] So defining what they want their brand to be defines them on what clear risk profile they probably is acceptable to them. And not every task carries the same risk profile. But it also doesn't carry the same business impact and benefit or not, as the case may be.
[00:13:02] And I think that's where we're doing a session with circa 50 people over the next six to eight weeks called AI Heroes. And AI Heroes is really trying to get them to understand the risk profile of their own business.
[00:13:20] Because low risk, repetitive activities, such as ticket enrichment, alert correlation, routine remediation, they're all simple places to begin. Whereas a high risk business decision probably needs, and I would expect it to today, need a human in the loop and human oversight.
[00:13:47] But that doesn't necessarily mean that's the future as well. Because once you've confirmed that you've done it right, 30, 40, 50, 60, whatever number of times, then if it happens that it's right every single time and the human didn't need to be in the loop, is there a way of speeding up that process and providing the end customer, so our customer's customer, the best return, which makes the brand look better in the future.
[00:14:16] So I think as the world is changing, the level of autonomy should depend on the potential consequences of incorrect decision and the quality of the available data, right? This is, it's really quite obvious. If I've got all the accurate data, I can make better business decisions where my decisions are going to be right 100% of the time. But to be able to do that, you have to build the trust.
[00:14:43] And that trust needs to be built incrementally. And as AI systems agents demonstrate consistent and explainable results, then we are in a far better place for the future. So the level of autonomy should really match the level of risk. And that conversation with the six companies I've had this week, it's pretty much similar in every one of those conversations.
[00:15:10] We want to do as much as we can as quickly as possible, but understanding the risk profile. And once we've proven it multiple times, then we can go on to the next level and make us more autonomous and more on the decision process, because the quicker we resolve, the quicker the benefit to the business.
[00:15:36] And I think it's also worth highlighting that IT teams in general have spent years adding things like monitoring, security, cloud, endpoint, and management tools, etc. So at what point does that very tall sprawl become part of the problem too, that is creating conflicting signals and making it harder for both humans and indeed AI to understand what is actually happening across that? The conversation I had this morning, which is really interesting,
[00:16:06] their company is bringing nine organisations together. And this is pretty normal across many parts of the world at the moment. MSPs is a great example of this as the PE backers are bringing them all together and trying to improve business outcomes and enable them to make money and reduce cost as they do this. But these organisations have all added different tools,
[00:16:35] and those different tools have resolved specific problems at the time. And different people had different ownership, and they were on the hook for the problem. So the result is dozens of monitoring and management systems, each are producing their own alerts in different places, which are supporting different parts of the business.
[00:17:00] Now, the problem is not necessarily that specialist tools exist in teams, is that struggle to connect those signals across them is becoming more and more important every single day. Now, providing a tool that can do all of that is not necessarily the correct answer, because you don't want jack of all if you need specific information to support.
[00:17:31] But what you do need to do is ensure that they are connected in a way that enables the decision of the business to have greater information to facilitate the appropriate answers. Organisations need a unified operational view that can correlate information across them. And as such, make super intelligent decisions
[00:17:59] because they've got more information. So having more data is good as long as you understand how they interact and how you understand what is going on with it. More data gives you more power as long as you understand how they correlate it, and then AI can make better decisions. And as we've discussed there,
[00:18:28] automation is only as reliable as the information that feeds it. So I'm curious, from what you've seen and heard here in your many conversations, what happens when an organisation introduces AI and autonomous remediation on top of incomplete, duplicated, outdated, or poorly contextualised operational data? I'm sure you've got a few war stories from the field there. But anything that you can share? Of course. Yeah, and it's really interesting.
[00:18:55] Organisations risk automated confusion rather than improving the outcomes because they've focused on the tech rather than the business outcome. And if you focus on the business outcomes, what are you trying to achieve? Then you can make intelligent decisions about the data you need to give you the answer. AI can process information quickly, but that speed amplifies the effect
[00:19:23] of both good and bad inputs. So if the data is bad, you're going to get a bad output. And if you haven't thought about what you're trying to achieve, you're going to then have a problem. And we've seen it in many places across many, many organisations where they started with technology rather than starting with a business outcome. And the thing I would, the people that are listening to this today, it's all about,
[00:19:52] have you understood what you want to achieve from the business outcomes to make the appropriate data information decision and where you're going to get it from? Incomplete or outdated information can lead to false positives, misruit causes, actions that address a symptom without resolving the underlying issue. We see that all the time. You fix it, but guess what? It's continuing to break. And we just don't need that in our business. We need to fix it and solve it for the future.
[00:20:21] See it, predict it, fix it, stop it happening again is something that needs to happen in organisations. So when you think about the correlation that really matters, if related signals are not corrected, AI may interpret one instance as several unrelated events, which causes problems as you go into your NOC
[00:20:51] because there are so many events and tickets that have been created. Improving the quality and consistency of contextual data and operational data should come before giving AI greater authority to make decisions. The key takeaway really is the faster way to automate the wrong decision is to give AI bad context. 100% with you.
[00:21:21] And I think there's somewhat of an interesting tension here between speed and certainty because if an AI system can detect a problem and believes it knows the fix, waiting for a human approval could prolong an outage, but equally allowing it to act alone could make things even worse. So how are you seeing organisations designing for that trade-off in practice? Because it almost feels like a paradox of sorts. organisations do not need
[00:21:51] to treat this as a choice first and foremost. Yeah. Should we have full autonomy or should we have complete human control? I think you've got to establish graduated levels of action based on the type of action, confidence in the recommendation, potential business impact, and whether the action is easily reversible. Again, focus on the business outcome and you are going to make the right business decisions.
[00:22:19] AI may be permitted to carry out a low risk and easily reversible action while requiring approval for a change affecting a critical service. That's sort of common sense, right? But guardrails are important and they're explainable and they give you audit capabilities. Now, they're becoming as important as the automation itself. You've got to prove in many ways what you're doing and why
[00:22:48] because you're treating this to get the appropriate business outcome. And the aim is not to remove people from the loop. I think human in the loop for important decisions has to be governed. And it's governed by action where judgment and accountability add the most value. And people with that focus, judgment, and accountability are there because it's really important for the appropriate business outcomes.
[00:23:17] The question isn't simply whether AI should act. Yeah. It's knowing which actions you've decided it is safe for AI to take. And I think that makes a lot of sense. I'm curious. As someone that works closely with enterprises and MSPs across UK and Europe there, what are you seeing separating the organisations that are achieving consistent results with AI and automation from those that are still
[00:23:47] stuck running experiments and proof of concepts and almost caught in pilot purgatory? Are there big differences that you're seeing between those that are getting right and those that are not? Pilot purgatory. What a great turn. It was interesting. I was in Munich last week, another event, and we were talking about AI and where you go and what you're meant to be doing. And yeah, how far do you want to take where you're going with these projects? I think successful
[00:24:16] organisations start with the operational problems they have. Yeah. Not an AI project, right? And they're focused on the business outcomes. God, I'm sorry, I sound really boring. But the focus on business outcomes as reducing incidents, improving uptime and accelerating resolution times has an impact to the bottom line. Yeah. Yeah. And that's, but do nothing is also incredibly, right, purgatory
[00:24:45] using your term because if you're not doing it, your competitors are and they will be, you will be left behind in, yeah, reducing your cost to serve, improving the way you service your customers, providing a better customer experience. And as an organisation, you don't really have a choice now. Companies are going down this route. It really is about the speed and how you get going.
[00:25:15] And if you're looking at a team of work colleagues, and let's just say there's 20 people in this room, one of them will be an AI hero. Yeah. As a minimum. And ultimately, we understand that AI will become the ability to either make room for growth or reduce cost. And if you're the person that is actually
[00:25:44] doing the work and becoming an AI hero for your business, you are more likely to be the person that leads the success of AI in the organisation. And what those people are doing typically in the organisations that are doing it well is they're investing early in data quality, investing early in getting visibility and cross-team collaboration. they are treating AI as a capability embedded into workflows
[00:26:13] rather than a standalone experiment. And yeah, the proof of concept. Importantly, they create a path from insight to action where clear controls around how much authority the AI has. Organisations that struggle often underestimate the importance of context, governance and organisational readiness. And it does depend on the speed of that organisation, how they work and what they do and what their
[00:26:43] risk factors are. But the organisations that are truly making progress today are treating AI as part of their operating model, not a science experiment. And more importantly than that, they're focused on how they can make money or save money protecting the brand as part of what's in their brain and how they're thinking. And so much of what you said there will resonate with people listening around the world. I don't know,
[00:27:12] let's say an IT leader who could be listening and they want to move towards more autonomous operations but they don't yet trust their environment enough to remove humans from the loop. Any foundations that you would tell them to fix first and what would tell them that they are fine and nearly ready to give automation more authority. What would you advise to that person listening? Start with the basics, honestly. You begin by establishing comprehensive visibility across the technology estate.
[00:27:42] That's really important. An organisation can't safely automate what it cannot see, bluntly. If it can't see it, how can you fix it? How can you resolve the problem? Data quality is obviously really important so get that improved. We have silos of data everywhere trying to reduce those silos, make them interconnected so the people or your AI all together are working from the same information to define the appropriate
[00:28:12] business outcomes. Define governance, policies and clear rules covering actions that AI can take independently which require approval and should remain under human control. That data governance we are seeing more and more in organisations and we are prescribing to AI governance when we're doing contracts as an example. And those are the people that have understood that AI is here, it's
[00:28:42] here to stay and we need to have governance around it. Yeah, got to ensure that the process and teams can understand what the systems recommend and make it auditable. So why has the AI reached that conclusion and is that the conclusion we wanted to reach? One sign that's signed truly that is an organisation is ready to grant AI more authority is when its recommendations
[00:29:11] consistently match the actions of its most experienced engineers they would have taken and we're sort of seeing from a level one to a level two from a level two to a level three how much of level one can be automated to how much of level two and how much of level three and that's the sort of the growing concern that happens and actually the benefits that happen is as you go from those it's becoming more business critical downtime is
[00:29:41] typically more affected and therefore we want to resolve them quicker right and that's where we get smarter and that's where we start getting the real business benefits so greater autonomy should be earned through evidence as you're moving through those levels of complexity to ensure that we have the evidence rather than just a hope strategy or a single leap so like everything autonomy
[00:30:10] should be earned through evidence and the AI agents need to prove that they can do it consistently and successfully to resolve outcomes quicker and more effectively than a human and I think that is a powerful moment to end on and anybody listening here concerned or frustrated or excited about this AI era now and infrastructure cloud internet
[00:30:40] services applications and yes AI agents and they want to find out more about anything we talked about today and how you might be able to help at logic monitor as well where's the best place of keeping up to speed with network with you and discuss many of these things obviously the company website we've got an event in November the 4th which is our user conference which is prospects are obviously
[00:31:10] allowed as well which is elevate and would love to see you there so please go to our quicker than you could ever imagine making sure you're making the right decisions based on business outcomes is how we can improve both your company and to be honest the world
[00:31:40] in greater one of things I should
[00:32:27] be used to reduce lessons from Matt today was that autonomy should be earned through evidence especially before giving AI greater authority organisations need visibility across its tech estate reliable contextual data connected teams and clear rules for what AI can do on its own and low risk and reversible actions typically provide the most sensible starting point obviously decisions affecting major
[00:32:57] services should retain human approval until the system has been able to prove itself consistently and I think this approach avoids the false choice between full autonomy and total human control instead we're creating graduated authority based on risk impact confidence and reversibility so thank you to Matt for joining me and rescuing business outcomes from becoming just another
[00:33:26] phrase we say without testing so logic monitors website Matt's LinkedIn profile and the elevate event I'll include links to them all but over to you what evidence would your organisation require before allowing AI to act alone tech talks network dot com easiest place in the world to find me and
[00:33:57] bye bye for now