AI Wrote the Code. Now comes the hard part.

Resolve AI has a production emergency its customers are writing checks to solve.

The company — which raised $125 million at a $1 billion valuation in February 2026, with total funding exceeding $150 million in sixteen months post-stealth — built what it calls an always-on agent layer for production systems. Its agents act as first responders for on-call alerts, typically triaging within five minutes before a human engineer is pulled in. Resolve says its new investigation architecture delivers more than double the root cause accuracy of its prior version on internal benchmarks — a claim the company measures against its own prior performance, not an independent benchmark.

The customer results are where the claim becomes concrete. DoorDash reduced time to root cause by up to 87 percent using Resolve's system, cutting investigation time from 40 minutes to one. Coinbase achieved investigation times 72 percent faster, with the company reporting it can now identify the root cause of an incident in under ten minutes. Zscaler, which handles more than 150,000 security alerts per month, uses Resolve's agents to handle alert volume that would overwhelm a human on-call rotation. The pattern is consistent: when the incident rate outpaces human triage capacity, the agent layer takes the first pass.

The underlying pressure is real and getting worse. One engineering organization Resolve works with ran an internal survey and found that 93 percent of its code was already being written by AI agents. At Resolve itself, the number is 95 percent. At the velocity Spotify's best developers are now merging hundreds of AI-generated pull requests per month through internal Claude Code tooling, the production surface accumulates faster than any human team can manually surveil it. The ops layer — the unglamorous work of diagnosing why code breaks and patching it before users notice — was always downstream of every coding decision. AI coding tools have now dramatically accelerated the top half, which means the bottom half is where the complexity concentrates.

Resolve is betting that agentic, autonomous investigation and remediation is distinct enough from conventional monitoring to justify a dedicated layer. Datadog, PagerDuty, and Splunk own the observability and incident management layer in most enterprise stacks. AWS, Azure, and GCP have their own alert tooling. Whether the ops agent sits above or alongside these systems — and who controls the alert-to-fix loop — is an open architectural question. StrongDM, a security infrastructure company, runs a three-engineer team where no human writes or reviews code — the limit case of what the model implies for team structure. If Resolve's approach becomes the ops standard, it owns the critical feedback loop between code deployment and failure diagnosis. If the monitoring incumbents close the gap first, the bet doesn't pay.

The customer evidence has limits. The case studies are Resolve-curated, and the accuracy and improvement claims come from Resolve's own benchmarks and customer self-reporting. The engineering org that reported 93 percent AI-generated code is unnamed. The next wave of AI tooling investment will likely flow toward code operations — the layer that determines whether AI-generated code actually survives contact with production systems — and Resolve is making the case that it gets there first.

What to watch: whether Resolve's agentic ops layer can move from triaging known alert patterns to diagnosing novel failures — the kind that require judgment rather than pattern match. If it can, the ops layer becomes the competitive moat. If it can't, the 87 percent improvement at DoorDash holds at 87 percent, and the ceiling of autonomous operations stays visible but unreachable.