A VP of operations once showed me a Confluence page with 47 steps, embedded screenshots, and a note that the process only works because a senior analyst manually checks each screen. That is the state of many automation portfolios: a few stable RPA bots that handle high-volume, rule-based work, and dozens of processes that never make it past the planning phase because they are too fragile for traditional RPA. Traditional RPA relies on selectors, xpaths, and object IDs. When the UI changes, the bot breaks. When the process involves exception handling, legacy desktops, or Citrix, RPA often fails before it even starts. Computer use agents see the screen and act like a human. They can follow a standard operating procedure written in plain English and still recover when the process goes off-script. The result is an automation pipeline that starts with a Confluence page, not a flowchart.
Why RPA breaks here
Traditional RPA works best when the source system is stable and every button has a predictable selector. In many enterprise environments that assumption does not hold. A 2023 industry survey found that 60 percent of RPA initiatives experience at least one bot failure within the first year, and the most common cause is a change in the UI that breaks selectors. The cost of that fragility is not just lost time. Each break means a developer must rebuild the bot, update documentation, and retest the process. For a mid-sized enterprise with 40 active bots, that can mean two full-time developer hours per bot per change, which adds up to weeks of maintenance every quarter. In processes that involve exception handling, legacy applications, or virtualized desktops, the failure rate is even higher. RPA bots are built for deterministic flows. When a field is missing, a popup appears, or the URL shifts, the bot halts and requires human intervention. That is why many SOPs are never automated at all.
What changes with computer use agents
- Computer use agents survive UI changes because they see the screen, not selectors.
- No brittle selectors or xpaths to maintain when the application updates.
- They recover from exceptions and unexpected states instead of halting.
- An SOP written in plain English is already almost a prompt for an agent.
- They work across any application, including legacy systems and Citrix.
- You can run multiple agents in parallel for higher throughput.
Traditional RPA is brittle and expensive to maintain, but computer use agents follow the SOP as written and adapt when the UI changes.
The SOP-to-agent pipeline
You can move from a Confluence page to running automation in a few steps. First, choose a process that is high-priority but currently manual or only partially automated. Look for processes with clear steps, some variability, and at least a few exceptions. This is where traditional RPA struggles. Next, document the process in plain language. You do not need to create a flowchart. You simply write each step as a sentence or two, referencing the screenshots and system names that already exist in your Confluence page. Then, send that documentation to a computer use agent. The agent reads the steps, interprets the context, and performs the actions on the screen. If a field is missing or a window behaves differently than expected, the agent can reason about the situation and recover, rather than crashing. After the first run, review the logs and refine the SOP if needed. Over time, you can expand the process, add more steps, and introduce parallel agents for different parts of the workflow.
How to move without the risk
A phased approach lets you test the model on a single process before scaling across the enterprise. Start with a pilot that targets the highest pain point: the process that people currently run manually. Set clear metrics, such as time saved, error reduction, or cost avoidance. Run the process with a human analyst first to capture the actual variation and exceptions. Then, run the same process with a computer use agent. Compare the results. If the agent handles the variability and recovers from exceptions, you have a proof of concept. From there, expand to similar processes that share the same applications and workflows. Keep your high-volume, stable RPA bots for backend tasks where selectors are reliable and the process rarely changes. Over time, you can migrate those processes to agents as well, but a staged rollout reduces risk. The goal is not to replace RPA overnight. The goal is to build a durable automation portfolio that grows with your applications and processes.
If your automation portfolio is stuck between brittle RPA bots and unwieldy SOPs, computer use agents offer a more durable path. They let you turn a Confluence doc into running automation, recover from exceptions, and adapt to UI changes without rebuilding every bot. To see how this works in practice for your use cases, book a demo with the Coasty team at https://cal.com/coasty/15min .
Want to see this in action?
View Case Studies