Software engineering is being redefined by AI agents. Since March, an internal team at SolarWinds known as “Area 51,” has been building an AI agent designed to automate IT operations. The organization claims the agent has increased engineering output fivefold compared to human engineering teams at the company.

The project contributed to the release of Avior, an agentic AI system designed to automate remediation across IT environments which SolarWinds announced today. Avior has tasks encoded into the agent as goals or outcomes that IT teams can use to coordinate specialist agents across tasks including correlating related alerts, continuously auditing devices, remediating configuration drift, implementing configuration backups, tracking capacity and managing the asset lifecycle.

According to SolarWinds VP of Software Engineering, Peter Choi, Area 51 began as a team of four engineers located across the U.S who collaborated to build an agent over Zoom and Teams calls, and has since scaled to approximately 25 employees. The objective was to create an agent that could reduce customers “daily toil,” across IT operations, which the team attempted to achieve by building an agent that could automatically remediate these issues.

With the help of AI coding tools, SolarWinds developed a prototype agent within approximately two weeks, which was requested by a dozen trial customers within just two months of its development. The project highlights how AI agents are transforming development and IT operations across the enterprise.

AI Agents As The New Bottleneck

Many organizations are now looking to AI agents to streamline IT operations. For example, technology advisory firm The Futurum Group, estimates that 49.2% of organizations are planning agentic AI deployments in IT operations and cybersecurity.

Coding agents like Claude Code and Codex provide engineers with access to powerful tools they can use to increase product delivery and eliminate the traditional productivity bottlenecks that human engineers faced. “I think my personal observation is that the implementation time and effort no longer seems to be a bottleneck. It’s about figuring out what is the right way to build, right? and what to build," Choi told me in a video interview.

SolarWinds looked to use coding tools to build an agent that could reliably remediate operations incidents. More specifically, Choi told me that SolarWinds opted to use codex and Claude Code as its primary coding agents, working to build a harness to enable the agent to intelligently select which models to use for the tasks it needs to perform.

“We want to shift the conversation from how intelligent a model is to how much useful work can we reliably trust agents to complete in these complex environments,” Krishna Sai, senior vice president and chief technology officer at SolarWinds told me in a video interview.

Capturing Domain Knowledge

In order to try and build an agent that could reliably automate IT operations tasks, SolarWinds collected domain knowledge across its product, engineering, sales and customer engagement teams, which was captured within the organization’s internal systems and fed into the agent as context.

The team looked at patterns in historic investigations, analyzing how often a given event occurred, while looking at the most frequently encountered customer escalations from the past 12 to 24 months and assessing how incidents were handled historically. Users also reviewed the agent’s recommendations on how to handle the issue.

At the same time, SolarWinds also gave the agent access to internal systems including CRMs, source code, Jira tickets and Confluence pages. Choi says the agent could then “synthesize,” that data to develop its skill sets. The approach aimed to build up “confidence,” that “the agent is behaving as expected and doing the right thing.”

Building An Agent Factory

As part of a broader effort to streamline the organization’s software engineering process, Choi has shifted from a management role back to hands-on coding, jumping from 3 GitHub commits to over 1,400 in a 6 month period. Coding agents have been integral to his workflow, with Choi noting that when he builds he codes using Claude Code, but when he needs to do a review of an implementation or pull request, he asks Codex to do a review.

From a leadership perspective, Sai says SolarWinds is undergoing a “builder transformation where every engineering or development team is evolving from a team that builds applications or features," “into teams that are building agents,” as part of what Sai calls an “agentic factory.” But what does this broader transformation look like? Sai says SolarWinds has a system of agents that every team is adopting, where every aspect of the product development lifecycle, from design agents that can produce SolarWinds compliant visual design, to agents that produce code, to review agents that look at code being produced and auto approve.

“All of these aspects of product development, we’re building agentic systems that power each of those and each of the teams,” Sai said. He also highlighted increased bandwidth across the organization’s engineering operation. "It’s as if they have a team of new junior engineers working for them, and they can allocate more and more work.”

Across the industry, many organizations are struggling to keep up with IT operations alerts. In May 2026, Production ops agent company NeuBird released the results of a survey of more than 1,000 SRE, DevOps and IT operations professionals and found that 44% of organizations experienced an outage in the past year that was directly linked to suppressed or ignored alerts. It also found that 83% of organizations report their teams ignore alerts at least some of the time.

Agentic deployments like Avior are an attempt to address the persistent challenge of alert fatigue through autonomous remediation. “The concept and strategy makes perfect sense. IT Must move past contextually absent device management and adopt a business outcome driven IT operational mentality. The characteristics that SolarWinds speaks of when referring to Avior are that of iteration toward a goal, auditable operation and fine-grained autonomy without being/becoming an unrestricted autonomous operator,"Carlos Casanova, VP principal analyst at Forrester told me via email.

Casanova added that these capabilities should be well received, but he expects enterprises to keep a very close eye on autonomous operations to ensure they don’t get outside their control structures. The pressure to adopt AI agents also creates new security and governance concerns.

As Casanova says, “this new era requires more rigor than just simple and traditional reactive controls,” highlighting that an hour of runaway AI activity is not the same as an hour of manual mistakes, but “magnitudes worse.” He adds that agents need to operate in a space of least agency, continuous risk management, and secured intent.