// the company logo, photographed in better light than your last incident
YESPRESS / COMPANY DOSSIER
Resolve AI.
An agentic AI Production Engineer that picks up the page at 3am so the on-call human doesn't have to.
An agentic AI Production Engineer that picks up the page at 3am so the on-call human doesn't have to.
Somewhere in a Slack channel at DoorDash, a checkout-service alert fires. Latency on a payment endpoint has crept past its threshold for the third time this week. Two years ago, this is where a human engineer would have rolled out of bed, opened six dashboards, and started typing "kubectl get pods" with one eye closed. Tonight, a Resolve AI agent gets there first. By the time the on-call engineer thumbs open Slack, the agent has already pulled the deploy timeline, narrowed the regression to a recent feature flag, and posted its working hypothesis.
That is the scene Resolve AI sells. Not slideware. Not a chatbot pretending to be helpful. An autonomous teammate that does the part of software engineering that nobody enjoys and almost nobody is good at after midnight.
The dirty secret of modern software is that it is increasingly held together by tired humans squinting at log lines. Microservices multiplied. Observability vendors got rich selling dashboards. Alerts proliferated. The number of engineers willing to be paged at 3am, however, did not.
Spiros Xanthos and Mayank Agarwal had front-row seats to this. They were the people building the observability tools - first at Omnition, which Splunk acquired in 2019, and then inside Splunk itself, where Agarwal served as chief architect for observability. They are also, somewhat awkwardly for the rest of the industry, co-creators of OpenTelemetry, the standard the whole sector now runs on.
So they watched, with the slightly weary patience of people who had built the previous solution, as that solution turned into a new kind of problem: more data, more dashboards, more alerts - and the same exhausted humans on the receiving end. The dashboards told you what broke. They did not fix anything.
Xanthos and Agarwal met at the University of Illinois Urbana-Champaign roughly twenty years ago. They have been collaborating since 2012, which in startup-time is roughly a geological epoch. By the time they founded Resolve AI in 2024, the bet was this: large language models had finally gotten good enough at reasoning over messy, real-world systems data that an agent could plausibly do the work of triage. Not perfectly. Not always. But often enough to matter.
Investors agreed. Greylock led a $35M seed in October 2024, with Fei-Fei Li and Jeff Dean writing personal checks - which is the venture-capital equivalent of getting a thumbs-up from both your favorite professors. Sixteen months later, Lightspeed led a $125M Series A at a $1B headline valuation, with DST Global and Greylock returning.
Resolve AI ships as a platform of agents organized around the actual rhythm of running production software. There is an On-call agent that joins the triage rotation - reading alerts, asking the obvious questions a senior engineer would ask, and proposing the obvious next step. There is an Incidents capability that brings a small team of agents into the war room with humans during live outages, helping with root cause analysis. There are background Operational Tasks, the kind of scheduled and trigger-based work nobody wants to own. And there is a Custom Agents layer - via MCP, API, and Skills - for the inevitable case where your company's internals are weird enough to demand a bespoke teammate.
Agents that triage alerts and ask the questions a senior engineer would, before anyone is paged.
Agent teams that collaborate with humans on live root cause analysis, in the channel.
Background agents for scheduled and trigger-based work. The unglamorous toil, automated.
Build proprietary agents via MCP, API, and Skills. For when your stack has, shall we say, character.
It is one thing to say "agentic AI." It is another to put it in front of an engineering team at Coinbase, where a wrong answer in production is measured in lawsuits, or DoorDash, where a wrong answer is measured in cold burritos.
It would be easy to read Resolve AI as a story about replacing engineers. It is, in fact, the opposite. The company's stated thesis - and the thing the founders keep returning to in interviews - is that software engineering has accumulated decades of operational tax, and that engineers got into this job to build, not to babysit alert queues. The agents are not the point. The freed-up humans are.
This is, of course, also a useful position to take if you would like enterprise engineering organizations to actually deploy you. It happens to be true and convenient at the same time, which Wilde would have called the highest form of honesty.
The optimistic version of Resolve AI's future is that "incident response" gradually stops being a job. The agents catch the regression before the pager fires. The humans wake up at 9am, read a tidy summary of the night, and merge a fix that the agent has already drafted. The dashboards still exist. They are just consulted less often, by fewer people, with less dread.
The skeptical version is that LLMs are excellent at the easy half of root cause analysis and humbling on the hard half - and that Resolve AI's value will depend on how much of the hard half they can credibly cover. Their bet is that with the right context (a "dynamic knowledge graph" of your systems, in their phrasing), the line keeps moving in the agent's favor. They have the team, the funding, and the customer base to find out.
The on-call engineer at DoorDash, in our opening scene, thumbs open Slack. The Resolve AI agent has already done the work. There is a hypothesis, a deploy timeline, a suggested rollback. The engineer reads it, agrees, hits a button, and goes back to sleep. The alert clears. The next morning, nobody writes a postmortem about heroics.
That, in the end, is the whole pitch. Less drama. Less downtime. Fewer 3am Slack threads that nobody remembers in the morning. Resolve AI is not promising to remove engineers from production. It is promising to remove the worst hour of the engineer's week. If they pull it off, the boring future is the good one.