# How Latent Health gets medications to patients faster with Devin ![Vimeo](https://devin.ai/_next/image?url=%2F_next%2Fstatic%2Fimmutable%2Fmedia%2Fcover.0lhpg6z57_2xq.webp\&w=3840\&q=75) 30% More PRs merged per engineer vs using only Claude Code 50 Automations running across engineering 100% Engineers actively using Devin ## About the company [Latent Health](https://www.latenthealth.com/) is an enterprise pharmacy intelligence platform whose clinical agentic engine helps health systems grow their pharmacy operations and serve more patients. More than 60 leading health systems, including Mount Sinai, Cleveland Clinic, Yale and others, together serving an estimated 70 million people, use Latent. The company has raised $80 million from investors including Transformation Capital, Spark Capital, and General Catalyst. Industry: Healthcare**[Visit site](https://www.latenthealth.com/)** ## Building software for a fragmented healthcare system Getting a patient onto a specialty drug can require pharmacy teams to read the patient's chart, find the insurer's coverage policy, assemble the required clinical evidence, submit a prior authorization, and appeal a denial. The process can take weeks, delaying treatment for patients who already have a prescription; 30–50% ultimately never get the drug at all. Latent automates much of that work so patients can get the medications their doctors prescribed faster. Its software reads clinical records and payer policies, determines what information is required, and helps pharmacy teams complete the next step. At Mount Sinai Health System, New York City's largest health system, prior authorization completion time fell about 75%, from 25–30 minutes to under five. Patients now get their medication in two days instead of four. At Yale New Haven Health, each team member can now process more than twice as many prior authorizations per month. There are serious engineering challenges involved in delivering these outcomes. There is no standard prior-authorization workflow. Each payer has different eligibility criteria, forms, and documentation requirements, while each health system stores the diagnoses, labs, and treatment history needed to satisfy them differently. And the rules themselves change. When Medicare opened the GLP-1 Bridge Program on July 1, 2026, four million members became eligible under new coverage rules. Within ten days, Latent had agents reading patient charts, determining which patients qualified, and assembling the right prior authorizations. > "There is so much operational context and process and preference that even if the base model is great, it still needs layers and layers of software around it to be truly useful." > > — Jiasi Zeng, VP of Engineering ## Latent needed an agent it could run on PHI Latent is an AI-native company. The team's philosophy is that they can't build the software their customers need unless every part of the business operates at the frontier: its people, its workflows, and the technology it uses. The urgency of the problem means Latent has to move at the speed of AI. But in healthcare, that speed has to come within the bounds of trust, with security harnesses and protections in place. That's why they chose Devin. By early 2026, Jiasi Zeng, Latent's VP of Engineering, wanted to start using software engineering agents seriously. But Latent handles protected health information, which narrowed the field: any vendor with access to its codebase needed to meet Latent's security and compliance requirements, including signing a Business Associate Agreement. In May and June 2026, Latent tested Devin alongside another vendor's cloud coding agents. > "It was just very clear that Devin was a much more mature product. The engineering team just naturally picked it up." > > — Jiasi Zeng, VP of Engineering Engineers started with individual coding tasks in the Devin web app. As they got more comfortable, they moved Devin into the systems where work already happened: Linear tickets, Slack threads, and scheduled automations. Now, they constantly use Devin. Every pull request Devin opens still goes through Latent's normal code-review process. The engineering metric Zeng watches most closely is pull requests merged per person. In the few months since Latent onboarded Devin, merged PRs per engineer have grown 30%. ## From individual tasks to 50 automations Over time, the team started using Devin for work that did not necessarily end in code. Because Devin can work across the same systems an engineer uses, a session can produce a pull request, a report, an analysis, or a Slack message. Latent now runs 50 Devin automations across engineering, including: - **Cross-functional project updates:** Devin pulls from Slack, Notion, and the codebase and sends weekly updates to the relevant teams. - **Security remediation:** Devin takes findings from Latent's existing security tooling and turns them into work the engineering team can act on. - **Recruiting briefs:** Devin prepares summaries of engineering candidates before interviews. - **Coffee-chat matching:** Devin pairs employees each week across the growing team. - **Knowledge maintenance:** Devin periodically cleans up its own knowledge base so outdated information does not accumulate. ## Devin moved beyond engineering Latent has also given people outside engineering access to Devin. Those teams primarily use it through Slack. When a salesperson recently needed a data pull, Zeng tagged Devin in the thread and pointed it to Latent's PostHog and Metabase connections. Devin pulled the data and returned the analysis without an engineer having to do the work manually. ## Going after hard problems in healthcare Using Devin gives Zeng's team more capacity to go after ambitious problems in healthcare. For example, the company trained CIPHER, its own 1.5-billion-parameter de-identification model, which removes patient identifiers from clinical notes with 97.9% recall at roughly one-thousandth the inference cost of a frontier model. Another is the work on the application layer: turning payer rules, clinical data, and health-system-specific workflows into software that can reliably take the next step for a patient. Latent is hiring at [latenthealth.com/careers](https://www.latenthealth.com/careers).