| This is a guest post from Wilmer Yan, Co-Founder of Coverwatch |
|---|
Coverwatch is an AI-native commercial insurance brokerage startup based in San Francisco. The company combines in-house risk specialists with its internal automation platform to help companies manage their risk portfolios, uncover coverage gaps, and negotiate better insurance pricing.
Commercial insurance is still a highly operational business. A broker gathers information from calls, emails, documents, and customer conversations; translates that information into carrier-specific applications; submits those applications across different channels; and manages underwriter follow-up until a policy binds. The work is highly operational, asynchronous, and full of edge cases.
Coverwatch built its platform to make that process faster and more reliable. It uses Temporal to coordinate long-running insurance workflows that combine AI agents, human review, carrier communication, and operational follow-up.
"At every other team I've been on, we've had to hand-roll workflow orchestration infrastructure with complex queues, retry mechanisms, state-machine tracking, and a lot of operational glue," says Wilmer, Coverwatch's founder. "Temporal Cloud abstracts away that complexity while staying reliable and intuitive to use."
The challenge#
The challenge in commercial insurance automation is coordinating long-running tasks and tracing the many handoffs that happen before coverage is bound.
A submission can begin with a customer call, continue through document intake and application preparation, pause while a broker reviews missing information, wait days for an underwriter response, and then resume when new questions, quotes, or bind instructions arrive. The system holding that together has to survive deploys, outages, third-party API issues, and long stretches of silence while people on the other end review paperwork.
Traditional queue-based background jobs and DAG-based workflow tools were a poor fit. Insurance submissions are not just a set of tasks to run in order; they wait on customers, brokers, underwriters, carrier systems, and retryable external services. Coverwatch needed a durable way to track the submission process over time, including what had finished, what was waiting on an outside party, what should retry, and where a broker needed to step in.
The team also wanted to avoid the burden of building and maintaining orchestration infrastructure. Without Temporal, they expected to need message queues, retry logic, scheduled polling, database-backed state reconstruction, and bespoke debugging tools just to make long-running automation safe enough for production.
The solution#
Coverwatch models each insurance workflow as a durable Temporal Workflow. Instead of scattering state across queues, cron jobs, and service-specific tables, developers can describe complex workflow steps all in one place. These steps can run immediately, pause for review, wait for an external event, or retry when a dependency is temporarily unavailable. Temporal supports all of this out of the box. Having the entire workflow trace of a carrier submission on the Temporal Cloud UI makes auditing and debugging workflows simple. The team knows when a workflow is stuck awaiting a customer's response within seconds from the dashboard.
"One underappreciated benefit is that Temporal centralizes workflow history and activity logs. When our AI coding agents have access to the Temporal log context, they can diagnose failures and iterate much faster."
Within the platform, Coverwatch combines Temporal's durable execution guarantees with agentic automation through Pydantic-AI's native integration. One agent may extract structured risk data from unstructured data, whilst another may prepare carrier-specific application data. Another may support submission or follow-up workflows with customers and underwriters. Temporal keeps those steps coordinated and cleanly traced throughout the lifecycle of the carrier submission process.
That durability is especially useful for agentic workflows, where model providers, tool calls, carrier systems, and other integrations can be unpredictable. If one of those dependencies has a transient issue, Temporal retries the affected activity according to the workflow's rules and preserves the context needed to continue. Instead of reconstructing what was in flight from logs, the team can inspect the workflow state and understand where the work paused or recovered.
"Long-running agentic communications and carrier submission workflows can survive LLM disruptions and outages without any manual intervention," says Wilmer.
Broker review fits into the same durable flow. When an agent produces a low-confidence output, or when a business-critical step needs human judgment, the workflow can pause and wait for human-verified data. Temporal Signals let Coverwatch resume from the same state once that review is complete, keeping automation and expert judgment connected instead of forcing them into separate systems.
The results#
Today, Temporal is core infrastructure for Coverwatch's backend asynchronous workflows. For the engineering team, that means faster development velocity, cleaner abstractions and stable foundation to scale their platform on.
"Temporal Workflows just work out of the box, it's also feature rich enough to do everything that we need for the foreseeable future."
For a small team operating in a high-stakes industry, that reliability compounds quickly. Workflows that once would have required bespoke orchestration infrastructure now ship faster, fail gracefully, and give the team full visibility into every step of the submission lifecycle.
The takeaways#
The team walked away with these lessons:
- Durable Execution is the right foundation for agentic insurance workflows. When AI agents, human reviewers, carrier systems, and external APIs all share the same workflow, failures in any one layer need to be isolated and retried; not manually reconstructed. Temporal gives Coverwatch that guarantee out of the box, without building custom retry or state-management infrastructure.
- Temporal Signals make human-in-the-loop automation composable, not bolted on. Rather than routing broker review into a separate system and reconciling state afterward, Coverwatch pauses workflows in place and resumes them with a Signal. Human judgment and automated steps live in the same execution graph.
- Workflow history is a first-class engineering asset. Coverwatch's team, including its AI coding agents, uses Temporal's execution logs to diagnose failures and accelerate iteration. Centralized, structured workflow history replaces ad hoc log scraping and guesswork.
- Startups can skip the orchestration infrastructure tax. Queue management, retry logic, scheduled polling, and state reconstruction represent a significant hidden cost in any long-running automation platform. Temporal Cloud removes that burden so a small team can focus on building domain logic instead of distributed systems plumbing.
To learn more about how Coverwatch is using Temporal to modernize commercial insurance, visit their site. If you're building long-running AI workflows and want the same durable foundation, get started with Temporal Cloud.