40x Faster: How Depot Powers Modern Software Builds With Honeycomb

From faster container image builds to their newly launched CI engine, Depot gives software teams the speed and reliability to ship more, wait less, and build with confidence.

They built their early observability stack the way most high-velocity engineering teams do: pragmatically. Logs went to Amazon CloudWatch, metrics lived in Grafana, and custom alerting code written in-house flagged issues to the right people.

This approach served the team well in the early days, but as Depot scaled, the manual effort required to maintain that patchwork began to outpace the business. Alert fatigue crept in. On-call engineers sometimes needed to spend weeks investigating signals that turned out to be noise, and the team found itself burning time simply verifying that things were okay when they could have been building.

"We had a disorganized and noisy way to know when things were broken, and we were spending too much time just checking that things were okay,” said Jacob Gillespie, CTO and Co-Founder at Depot.

As the customer base grew, the weight of their homegrown approach became impossible to ignore. Because several team members came from observability companies, Depot understood the landscape well and knew exactly what they were looking for. They needed a system that could:

• Define what healthy looked like across each product
• Surface problems before customers felt them
• Cut through noise so on-call engineers could act on signal
• Support SLOs as a first-class concept
• Scale with a team that had no intention of slowing down

With deep roots in the space, the team possessed the technical maturity to know exactly which solution fit their needs. Jacob had tracked Honeycomb for years and recognized that its philosophy on traces and native SLO support aligned perfectly with Depot's requirements. The decision was straightforward.

"Honeycomb is the best tool I have seen for traces, and the best at the SLO concept, which is exactly what we wanted," Jacob said.

Depot integrates deeply with GitHub and GitHub Actions. When an issue arises, the team must quickly determine whether the problem is internal to Depot or related to an external service.

Before Honeycomb, answering that question meant roughly an hour of manual log correlation. Engineers had to build up context for one error at a time to determine whether a pattern pointed somewhere actionable or to an intermittent external service issue. Often it was the latter, meaning an hour was spent just to conclude the issue was out of their hands.

With Honeycomb BubbleUp, that same triage now takes seconds.

By selecting an anomaly, engineers can see the differentiating attribute and move on. Jacob described the workflow plainly: pull up the heatmap, select the problem window, and BubbleUp surfaces which attributes distinguish the bad traces from the good. If the differentiating attribute points to GitHub, the investigation is over before it starts.

The downstream effect extends beyond time savings. Depot can now navigate customer conversations during GitHub incidents with certainty, and that clarity builds trust. "We want to be the resource customers turn to so we can tell them exactly what they’re seeing. Honeycomb lets us deliver that. We understand what is going on and we can be that authority for our customers,” Jacob explained.

Depot runs five distinct products, and before Honeycomb, the engineers didn’t have a single pane of glass to verify the health of the entire platform. Keeping tabs on every product required manual checking and reactive responses to alerts.

With Honeycomb, Depot now has a single dashboard with an SLO for each product, updated in real time.

• SLOs built for each product's reality
  Each SLO reflects what healthy means for that specific product. For CI, it is queue time, or how fast Depot starts fulfilling a build request. For the cache product, it is error rates across reads and writes. These are thresholds that are tuned, tested, and tied to burn-down alerts.
• Peace of mind for the whole company
  The dashboard is widely used by engineers and leadership alike. Depot's CEO, a former engineer who could absolutely dig through logs, no longer needs to. One look within Honeycomb confirms everything is running as it should.

"Honeycomb is where you go for an authoritative understanding of the system. If something you need to know exists, it’s there. And if it doesn’t exist there yet, we add it,” Jacob noted.

Depot's customers are technical, so when they open a support ticket, it’s rarely a simple question. It’s usually a specific and complex debugging request rooted in their own build environment. For a small team, meeting that bar consistently takes the right infrastructure.

Depot connected their AI coding agents, Claude via Cursor and Claude Code, directly to Honeycomb through the MCP integration. When a support ticket comes in, the agent reads it, pulls correlated traces from Honeycomb, and assembles a root cause analysis with supporting evidence. A human engineer still writes the final customer response, but they arrive at that conversation already knowing the answer.

• Structured data makes AI fast and accurate
  Honeycomb's structured, standards-based trace data is easy for AI agents to navigate. There is no ambiguity or manual parsing of unstructured logs.
• Small team with a big-team customer experience
  Some Depot customers have described the company as feeling like an extension of their own engineering team. This workflow is an example of how Depot earns that reputation.

"We connect a customer support ticket to Honeycomb, have the agent pull the traces of what the customer saw, and can arrive at the answer before we even pick up the conversation,” Jacob shared.

Customers choose Depot to build faster. Delivering on that means knowing, at every layer of the control plane, whenever there’s something slowing things down.

Depot runs across AWS ECS, Fargate, and Lambda. Without distributed tracing, each of those services produces its own isolated log stream. By unifying these services into a single coherent picture, Honeycomb's distributed tracing makes cross-service boundaries transparent and debugging nearly instantaneous.

• Moving from 100ms to 50ms and beyond
  With full trace visibility, performance improvements stopped being guesswork. Engineers can examine a trace, spot a span that takes longer than it should, and act on it.
• Meaningful improvements for the customer
  Those gains compound at the customer level. Some Depot customers have seen build times drop by as much as 40x.

Jacob shared, "As we make performance improvements, customers see that drop in time. They message us to ask if we did something because their builds just got faster. That kind of customer delight comes directly from the data Honeycomb gives us."

Depot CI, the company’s CI orchestration engine built to reduce dependency on external systems like GitHub Actions, launched in 2026. Having seen the value of early instrumentation, Depot shipped the new engine with Honeycomb tracing baked in from day one. This has now become the Depot standard: every new product is built with Honeycomb from the first line of code.

At the center of how Depot connects observability to their AI tooling stack is the Honeycomb MCP integration. Combined with custom API wrappers, it lets any AI agent in their environment access Honeycomb data without friction. As the pace of AI-driven development keeps accelerating, the value of that connection grows with it.

The bigger vision Jacob describes is a developer platform built for the way software is made today and where it is heading. Tools that support modern workflows, not just the ones that worked a decade ago. Honeycomb is the observability foundation that makes that possible.

"The rate of change today is significantly higher than it used to be. That makes observability critical. To keep up, to understand whether things are working, to make sure systems are performant enough to handle new demand, Honeycomb is how we do that," Jacob observed.

1. Invest upfront because it compounds
   A focused instrumentation effort frees your team from firefighting. The time you put in pays back with interest.
2. Be intentional about what you trace
   Don’t blindly collect data. Know what you’re watching for and build your instrumentation to match. You’ll get more value from deliberate tracing than from a firehose of noise.
3. Make Honeycomb your authoritative source
   If the data is not there, add it. Consistency is what gives the tool its power across your entire team.
4. Measure everything you care about improving
   Visibility creates its own momentum. Once engineers can see the numbers, they want to fix them.