Building trustworthy agents for Operational Excellence reviews<br>Metadata<br>authorAaron Wirick<br>published2026-08-18<br>read_time16 min

On this page

Tags<br>agentsoperational-excellenceopexllmengineering-intelligence

Building trustworthy agents for Operational Excellence reviews<br>How and why we built an agent to drive an engineering process

Overview

At Cortex, we run our own weekly Operational Excellence (OpEx) reviews to continually improve our engineering processes and empower our engineers to deliver an awesome experience for our customers. Since our main product at Cortex is a platform that helps organize engineering teams, it’s natural that the next evolution of our product would include a toolkit for other companies to build their own OpEx program.

The best way to do that in 2026 is to build an agentic capability that understands engineering, has access to metrics and data about our engineering processes, and can support regular OpEx reviews through data, interpretation, and open-ended question answering.

This post is a look at how our new OpEx Review Agent works. As engineers, both the “why” and the “how” are kind of interesting so rather than just focusing on the technical bits it starts with a little background on where this product came from and how it helps serve engineering as a process. From there it covers why an agentic system fits well for the problem space and wraps up by addressing a key challenge, trustworthiness, that we had to overcome when building it.

So what is Operational Excellence and why should I care?

If you’ve never heard of Operational Excellence before I guarantee that you have still felt its impact. OpEx is a business strategy that focuses on continual improvement, empowering everyone in an organization to deliver value to customers and reduce waste or bottlenecks. The most famous example of OpEx in action is the Toyota Production System (TPS) which is the system that gave us Kanban.

OpEx at Cortex

I’ve worked at Cortex for a little over two years now and, starting on my second day of work, I’ve been participating in the weekly Operational Excellence (OpEx) process that my colleague Shawn leads. It follows the same format every week. We look at our top-level reliability metrics, we follow up on initiatives, we talk about recent incidents, and we keep a pulse on what’s working and what needs more attention. That regular focus sets the tone for how our leadership (both ICs and Managers) thinks about making our product resilient, scalable, and awesome.

The success of that program has led to legitimate, material improvements in our product. Despite serving increasingly greater scale, system latencies are down, and we’ve added 9s to our availability SLOs. Our on-call rotations subjectively have become much less crazy and I no longer feel like I need to sleep with my laptop under my pillow. In an industry where it seems like AI has led to significant quality issues, we’ve improved on most of the key engineering metrics that we track.

This success of Cortex’s OpEx process is why our CTO, Ganesh, distilled down the principles that worked for us into the DRIVE Framework. That framework was built from real lessons not only at Cortex but at other major engineering teams in the industry.

API requests / month<br>111.9M<br>Sep 2025

313.3M<br>Jul 2026

Catalog API p99 latency<br>4.50s<br>Sep 2025

0.72s<br>Jul 2026

What our OpEx Agent does is allow you to capture these same gains without building this whole process from scratch. By just configuring the agent as a Cortex customer, you too can start driving towards these types of improvements. For example, many customers care a lot about Delivery (the D in DRIVE) and they will pay attention to things like deployment frequency and review cycle times.

Before you run your meeting, the agent wakes up, pulls key metrics that you’ve identified and runs an interpretation workflow that builds you a report highlighting what is improving and what needs focus. It ties back to the rest of the Cortex ecosystem so that teams can drill into specific problem areas with our Engineering Intelligence tools or take other actions like creating Initiatives (the I in DRIVE) to improve a trouble spot.

What does Agentic Capability even mean

In 2026, agentic capabilities are all the rage for good reason. Used here, the term refers to a system that uses a Large Language Model (LLM) to reason about a domain and feeds that reasoning back into a product experience. Not just a chat bot, but something that proactively works to help you understand things and can exercise autonomy in making decisions about what to do next.

Even before we wrote a line of code, we knew there were some properties that our system required in order to effectively support OpEx reviews. It had to be 1) trustworthy and based on solid data, 2) dynamic and customizable to meet the needs of a highly varied customer base and 3) incorporate temporal awareness and memory so that as issues come in and out of focus...