Arize CEO: AI Agents Will Automate Software Fixes

Jason Lopatecki, founder and builder at Arize, envisions a future where AI agents autonomously fix software issues, transforming the observability landscape. Speaking at the AI Engineer World's Fair, Lopatecki explained the evolution of observability from human-driven analysis to an agent-centric approach, where telemetry data fuels AI agents to identify and resolve problems.

From Human Debugging to Agent Automation

Lopatecki highlighted that traditional observability relied on human developers manually sifting through logs and dashboards to debug issues. This process is often slow and energy-intensive, especially during critical outages. He stated, "Observability used to be for humans. Used to be a UI you click, a graph you click, something you look at."

The shift, he explained, is towards "Software 2.0" which combines coding agents with robust telemetry. These agents, powered by data such as traces and logs, can trace execution paths and identify root causes. Lopatecki emphasized that "telemetry is like this smoke thrown off of your system that can allow these agents to go make fixes."

The 'Signal' Platform and Self-Improving Agents

Arize's new platform, 'Signal,' is designed to harness this AI-driven approach. Lopatecki described it as a "next generation" tool that "automate[s] a bunch of things which we do every day." The core idea is to build systems that can "autonomously fix themselves."

The process, as outlined by Lopatecki, involves an event occurring, gathering context through skills, and then triggering an agent to investigate and propose a fix. This creates a loop where the agent can identify issues before a human even notices them, and the evidence is already compiled by the time a human reviews it. "Your job kind of moves from responder to reviewer," he noted.

Bridging the Gap: From Building to Improving at Agent Speed

A key challenge identified is that while developers can now build at agent speed, they still struggle to improve systems at that same speed. Lopatecki explained, "You can build at agent speed, but today you can't improve your systems really at this agent speed." The bottleneck, he argues, is not the fix itself, but the confidence and evidence gathering required before deploying a fix.

Arize's approach inverts the traditional loop. Instead of humans debugging and agents fixing, their system allows agents to investigate and write fixes, which are then reviewed by humans. This model aims to significantly accelerate the feedback loop for system improvement.

The Future of Observability: Data-Driven and Autonomous

Lopatecki's vision for the future of observability is one where systems generate significantly more traces and logs, feeding a continuous loop that allows agents to self-correct. "The future observability actually looks a lot more like this than it does clicking around a Grafana UI," he stated.

He showcased a demo of their agent, Alex, identifying and proposing a fix for a "stream canceled" event caused by a retry loop. The agent was able to pinpoint the issue, which required a simple one-line code change, demonstrating the potential for AI to streamline debugging and maintenance processes.

Open Architecture and Product Lines

Arize offers its platform through its SaaS product, AX, and also provides Phoenix, an open-source version. Lopatecki emphasized an open approach to agent architecture, allowing users to select their preferred "harnesses," "sandboxes," and "skills." He noted that many clients prefer to install Arize's sandboxes within their own VPCs to maintain data security, especially when dealing with sensitive production systems.

The presentation concluded with a look at the Arize Agent Swarm, illustrating how multiple agents can work together. Lopatecki reiterated his belief that by logging and tracing orders of magnitude more data, systems can achieve continuous self-improvement loops, fundamentally changing how software is maintained and operated.