Skip to main content

What Is Real User Monitoring in an Observability World? It Is Not APM "Agents" - Part 1

Eric Futoran
Embrace

Agent-based approaches to real user monitoring (RUM) simply do not work. If you are pitched to install an "agent" in your mobile or web environments, you should run for the hills. The world is now all about end-users. This paradigm of focusing on the end-user was simply not true a few years ago, as backend metrics generally revolved around uptime, SLAs, latency, and the like. DevOps teams always pitched and presented the metrics they thought were the most correlated to the end-user experience. But let's be blunt: Unless there was an egregious fire, the correlated metrics were super loose or entirely false. Instead, your teams should prioritize alerts, monitoring, and work based on impact to the end-user, as it directly affects your businesses. And your developers and DevOps teams should collect data, monitor, prioritize, and resolve issues accordingly.

The agent-based RUM problem

"Agents" are a mechanism that does not work in the current end-user centric world. They were born out of shimmying the principles of the backend to mobile, web, and the myriad of other ways users interact with the world. Let's compare the difference between user environments and backend environments: ■ User environments are open, unstructured, and uncontrollable as they are unowned devices and browsers with the central figure being an unpredictable user. ■ Backend environments are closed, structured, and controlled as they are composed of relatively homogenous physical and cloud applications. With closed systems that have fewer external variables, agents focus on a known set of errors to monitor and to trigger data collection for resolution. However, monitoring systems outside of the backend is complex because there are a multitude of types of errors way beyond crashes, error logs, network traces, and API errors. In an observability world, real user monitoring is about collecting "all" the data for every session — good or bad — and not just a sampled set based on predefined error types. Only by collecting the entirety of every session can the best vendors have the opportunity to analyze and provide the utmost value to your teams. These vendors have evolved beyond agents to surface every type of user-impacting issue, help resolve them by comparing against good sessions, and prioritize overall impact across the complete set of issue types. For example, the same crash for two different users could have different root causes because of the environments, third-party SDKs, and API timeout parameters. To hit the difference home, watch a developer, outside of DevOps, open a RUM dashboard for a vendor who uses the agent-based approach. The core dashboard will have the following: ■ A geographical map laying out the incidents ■ A generic list of error logs and crashes ■ Some sort of mapping of network errors ■ A single health score The developer reviewing this dashboard will not come back to it regularly or at all. And it's not hard to see why. The dashboard does not tell them which users are affected, where to prioritize their efforts, or the types of bugs and optimizations that they should care most about. It's not built for them from the data collected to data organization and display. There is a reason why these developers always implement and use other vendors — even for simple concepts like error logging and crashes — alongside those application performance monitoring vendors. Let's deep dive into the core differences between these approaches and explore what a true real user monitoring methodology looks like. That way, you will know it when you see it and can create the best experience for your end-users as well as your developers and DevOps team.

The spider web problem

To illustrate the core implication of an agent mentality, let's focus on the "spider webs." You know the ones I'm talking about. You've seen the cool demos with a picture connecting nodes across your systems to demonstrate "visibility" across all the apps running on your servers and machines. Everything is connected by an ever-expanding spider web of nodes and lines — every app, compute instance, API call, etc. Oh, it's very pretty to see all the apps and API calls going to and from each other. It's also a nice source of confidence that the agents are collecting the data required to monitor, identify, and resolve potential issues. However, the very nature of this mental model of a spider web is it assumes all the issues occur on the lines between the nodes or on the nodes themselves: ■ An increase in network latency means you should look at the connected database, server, or service calls. ■ An increase in downtime means you should look at the connected servers to see if they're under heavy load. ■ An increase in transaction failures means you should look at the connected service calls for a point of failure. The paradigm of agents is one of looking for a closed set of known symptoms for broken apps, failing processes, and poorly designed code. To help resolve these symptoms, the agents collect samples of app and process information, so that when an API throws an error or a process has downtime, the agent collects the corresponding data in reaction to the error. And this approach works … on the backend, for a known set of errors, in a controlled environment, with little external pressure from the outside world. But when applied to the client side of web and mobile, what happens when the complexity explodes?  What happens when there are an infinite number of unknown pressures, from the users, the devices, the operating systems, the app versions, the network connectivities, and the other apps running? How do you truly understand your team's effectiveness when the biggest issues are not related to downtime or following individual service calls throughout a distributed system?

The problem with uncontrolled environments

Uncontrolled environments are any digital experience that's external to data centers. Beyond just smartphones and web browsers, they're point of sales, VR and AR devices, tablets in the field, and smart cars. And the world is increasingly one of uncontrolled environments for business-critical touchpoints. The most effective developer and DevOps teams monitor these client-side environments with early warning systems to determine when users are impacted so they can triage and resolve issues. They flip the traditional application monitoring paradigm. ■ Traditional application monitoring: Sample data by looking for a known set of errors, then gather context around them. ■ Modern application monitoring: Gather data without knowing its full value, correlate those data points to user impact from the end-user vantage point, then determine the error, measure the impact in order to prioritize it, and route it accordingly. In order to collect, identify, and resolve errors correctly, DevOps teams must understand the challenges that come along with running apps in these types of uncontrolled environments. After all, the assumptions about where failure points can happen are vastly different. Start with: What Is Real User Monitoring in an Observability World? It Is Not APM "Agents" - Part 2

Eric Futoran is CEO of Embrace

The Latest

The final two installments of the series, Exploring AI Observability, cover predictions from the experts about where AI observability is heading ... In the next few years, tracking your AI won't just be a nice-to-have tool, it will be a survival requirement for businesses ...

Cloud absolutely changed IT for the better. But after more than a decade of using it, companies have also learned something else. Where an application runs matters, and there isn't one environment that makes sense for everything ...

Automation often starts with the people closest to an immediate need. A developer may schedule a script with cron or Amazon EventBridge Scheduler; a business team may adopt an RPA bot, low-code platform, or agent. Many tools stay useful within one domain. Some become consequential across organizational boundaries because other systems and teams depend on them or business outcomes rely on their timing and recovery. Those cases may call for broader coordination ...

The conventional wisdom in our industry says that AI demands a new operating model and specific data tailored for it. I disagree. AI is changing the rules, but not in the way people think ...

Most organizations invest heavily in security tools, policies, and controls. Yet despite these efforts, one reality persists: employees often find ways around them. This isn't always malicious. In many cases, it’s a response to friction, urgency, or misalignment between how work actually happens and how systems are designed to protect it. When employees bypass security, it creates blind spots. Sensitive data may move through unmonitored channels, access controls may be weakened, and risks that security teams intended to mitigate reappear in new forms ...

Enterprise AI projects are often justified with a familiar metric: hours saved. A team automates a manual process, employees spend fewer hours reviewing documents or entering data, and the productivity gain becomes the headline. Time savings matter, but they capture only part of the financial impact AI can have inside an organization ... A more useful way to evaluate enterprise AI is to examine the financial decisions and risks surrounding the workflows it supports ...

We run an internal AI assistant that answers engineering questions in Slack ... Each question kicks off what we call a turn ... We could already audit a single turn after the fact. Every reply carries a short lookup token in its footer, and a slash command pulls up the full receipt for that turn. What we could not do was answer questions about the whole fleet. How long do turns take? What do they cost? Which tools does the agent reach for? Where does the time go inside a single answer? Those are APM questions. So we treated the agent like any other service and instrumented it with OpenTelemetry, end to end. This post covers what we built, the design choices that turned out to matter, and the things that surprised us ...

Applications have long used bounded, deterministic logic to control what happens within their own functions. Process orchestration coordinates activity across people, applications, decisions, and automation tools. Workload orchestration coordinates executable work and dependencies across applications, platforms, and infrastructure. As agentic capabilities enter applications and domain systems, more planning and decision-making will also occur inside those environments. These forms of automation will coexist. Sometimes they will overlap, and they will increasingly need to exchange triggers, state, and results. But they are not becoming interchangeable ...

For many organizations, the AI agenda has landed squarely on the CIO's desk. At first glance, that makes perfect sense. AI depends on data, infrastructure, security, integration, governance and scalable technology platforms — all areas where the CIO plays a central role. But there is a problem with treating AI primarily as an IT initiative: AI is not simply a technology deployment. It’s also a business transformation ...

In MEAN TIME TO INSIGHT Episode 28, Shamus McGillicuddy, EMA VP of Research, Network Infrastructure and Operations, and special guest Sanjay Gidwani, Founder & CEO of Kosmos AI Labs, discuss root cause analysis ... 

What Is Real User Monitoring in an Observability World? It Is Not APM "Agents" - Part 1

Eric Futoran
Embrace

Agent-based approaches to real user monitoring (RUM) simply do not work. If you are pitched to install an "agent" in your mobile or web environments, you should run for the hills. The world is now all about end-users. This paradigm of focusing on the end-user was simply not true a few years ago, as backend metrics generally revolved around uptime, SLAs, latency, and the like. DevOps teams always pitched and presented the metrics they thought were the most correlated to the end-user experience. But let's be blunt: Unless there was an egregious fire, the correlated metrics were super loose or entirely false. Instead, your teams should prioritize alerts, monitoring, and work based on impact to the end-user, as it directly affects your businesses. And your developers and DevOps teams should collect data, monitor, prioritize, and resolve issues accordingly.

The agent-based RUM problem

"Agents" are a mechanism that does not work in the current end-user centric world. They were born out of shimmying the principles of the backend to mobile, web, and the myriad of other ways users interact with the world. Let's compare the difference between user environments and backend environments: ■ User environments are open, unstructured, and uncontrollable as they are unowned devices and browsers with the central figure being an unpredictable user. ■ Backend environments are closed, structured, and controlled as they are composed of relatively homogenous physical and cloud applications. With closed systems that have fewer external variables, agents focus on a known set of errors to monitor and to trigger data collection for resolution. However, monitoring systems outside of the backend is complex because there are a multitude of types of errors way beyond crashes, error logs, network traces, and API errors. In an observability world, real user monitoring is about collecting "all" the data for every session — good or bad — and not just a sampled set based on predefined error types. Only by collecting the entirety of every session can the best vendors have the opportunity to analyze and provide the utmost value to your teams. These vendors have evolved beyond agents to surface every type of user-impacting issue, help resolve them by comparing against good sessions, and prioritize overall impact across the complete set of issue types. For example, the same crash for two different users could have different root causes because of the environments, third-party SDKs, and API timeout parameters. To hit the difference home, watch a developer, outside of DevOps, open a RUM dashboard for a vendor who uses the agent-based approach. The core dashboard will have the following: ■ A geographical map laying out the incidents ■ A generic list of error logs and crashes ■ Some sort of mapping of network errors ■ A single health score The developer reviewing this dashboard will not come back to it regularly or at all. And it's not hard to see why. The dashboard does not tell them which users are affected, where to prioritize their efforts, or the types of bugs and optimizations that they should care most about. It's not built for them from the data collected to data organization and display. There is a reason why these developers always implement and use other vendors — even for simple concepts like error logging and crashes — alongside those application performance monitoring vendors. Let's deep dive into the core differences between these approaches and explore what a true real user monitoring methodology looks like. That way, you will know it when you see it and can create the best experience for your end-users as well as your developers and DevOps team.

The spider web problem

To illustrate the core implication of an agent mentality, let's focus on the "spider webs." You know the ones I'm talking about. You've seen the cool demos with a picture connecting nodes across your systems to demonstrate "visibility" across all the apps running on your servers and machines. Everything is connected by an ever-expanding spider web of nodes and lines — every app, compute instance, API call, etc. Oh, it's very pretty to see all the apps and API calls going to and from each other. It's also a nice source of confidence that the agents are collecting the data required to monitor, identify, and resolve potential issues. However, the very nature of this mental model of a spider web is it assumes all the issues occur on the lines between the nodes or on the nodes themselves: ■ An increase in network latency means you should look at the connected database, server, or service calls. ■ An increase in downtime means you should look at the connected servers to see if they're under heavy load. ■ An increase in transaction failures means you should look at the connected service calls for a point of failure. The paradigm of agents is one of looking for a closed set of known symptoms for broken apps, failing processes, and poorly designed code. To help resolve these symptoms, the agents collect samples of app and process information, so that when an API throws an error or a process has downtime, the agent collects the corresponding data in reaction to the error. And this approach works … on the backend, for a known set of errors, in a controlled environment, with little external pressure from the outside world. But when applied to the client side of web and mobile, what happens when the complexity explodes?  What happens when there are an infinite number of unknown pressures, from the users, the devices, the operating systems, the app versions, the network connectivities, and the other apps running? How do you truly understand your team's effectiveness when the biggest issues are not related to downtime or following individual service calls throughout a distributed system?

The problem with uncontrolled environments

Uncontrolled environments are any digital experience that's external to data centers. Beyond just smartphones and web browsers, they're point of sales, VR and AR devices, tablets in the field, and smart cars. And the world is increasingly one of uncontrolled environments for business-critical touchpoints. The most effective developer and DevOps teams monitor these client-side environments with early warning systems to determine when users are impacted so they can triage and resolve issues. They flip the traditional application monitoring paradigm. ■ Traditional application monitoring: Sample data by looking for a known set of errors, then gather context around them. ■ Modern application monitoring: Gather data without knowing its full value, correlate those data points to user impact from the end-user vantage point, then determine the error, measure the impact in order to prioritize it, and route it accordingly. In order to collect, identify, and resolve errors correctly, DevOps teams must understand the challenges that come along with running apps in these types of uncontrolled environments. After all, the assumptions about where failure points can happen are vastly different. Start with: What Is Real User Monitoring in an Observability World? It Is Not APM "Agents" - Part 2

Eric Futoran is CEO of Embrace

The Latest

The final two installments of the series, Exploring AI Observability, cover predictions from the experts about where AI observability is heading ... In the next few years, tracking your AI won't just be a nice-to-have tool, it will be a survival requirement for businesses ...

Cloud absolutely changed IT for the better. But after more than a decade of using it, companies have also learned something else. Where an application runs matters, and there isn't one environment that makes sense for everything ...

Automation often starts with the people closest to an immediate need. A developer may schedule a script with cron or Amazon EventBridge Scheduler; a business team may adopt an RPA bot, low-code platform, or agent. Many tools stay useful within one domain. Some become consequential across organizational boundaries because other systems and teams depend on them or business outcomes rely on their timing and recovery. Those cases may call for broader coordination ...

The conventional wisdom in our industry says that AI demands a new operating model and specific data tailored for it. I disagree. AI is changing the rules, but not in the way people think ...

Most organizations invest heavily in security tools, policies, and controls. Yet despite these efforts, one reality persists: employees often find ways around them. This isn't always malicious. In many cases, it’s a response to friction, urgency, or misalignment between how work actually happens and how systems are designed to protect it. When employees bypass security, it creates blind spots. Sensitive data may move through unmonitored channels, access controls may be weakened, and risks that security teams intended to mitigate reappear in new forms ...

Enterprise AI projects are often justified with a familiar metric: hours saved. A team automates a manual process, employees spend fewer hours reviewing documents or entering data, and the productivity gain becomes the headline. Time savings matter, but they capture only part of the financial impact AI can have inside an organization ... A more useful way to evaluate enterprise AI is to examine the financial decisions and risks surrounding the workflows it supports ...

We run an internal AI assistant that answers engineering questions in Slack ... Each question kicks off what we call a turn ... We could already audit a single turn after the fact. Every reply carries a short lookup token in its footer, and a slash command pulls up the full receipt for that turn. What we could not do was answer questions about the whole fleet. How long do turns take? What do they cost? Which tools does the agent reach for? Where does the time go inside a single answer? Those are APM questions. So we treated the agent like any other service and instrumented it with OpenTelemetry, end to end. This post covers what we built, the design choices that turned out to matter, and the things that surprised us ...

Applications have long used bounded, deterministic logic to control what happens within their own functions. Process orchestration coordinates activity across people, applications, decisions, and automation tools. Workload orchestration coordinates executable work and dependencies across applications, platforms, and infrastructure. As agentic capabilities enter applications and domain systems, more planning and decision-making will also occur inside those environments. These forms of automation will coexist. Sometimes they will overlap, and they will increasingly need to exchange triggers, state, and results. But they are not becoming interchangeable ...

For many organizations, the AI agenda has landed squarely on the CIO's desk. At first glance, that makes perfect sense. AI depends on data, infrastructure, security, integration, governance and scalable technology platforms — all areas where the CIO plays a central role. But there is a problem with treating AI primarily as an IT initiative: AI is not simply a technology deployment. It’s also a business transformation ...

In MEAN TIME TO INSIGHT Episode 28, Shamus McGillicuddy, EMA VP of Research, Network Infrastructure and Operations, and special guest Sanjay Gidwani, Founder & CEO of Kosmos AI Labs, discuss root cause analysis ...