How Merge lifted weekly output 37%—and built AI-enabled workflows for migration & performance uplift

Merative's Merge Imaging division builds medical imaging software trusted by 6 of the 10 largest U.S. health systems. Their solutions power cardiology, radiology, and hemodynamics workflows in hospitals globally. Engineers at Merge maintain complex, mission-critical systems where reliability isn't optional.

Merge wanted more than access to AI tools. They needed a repeatable way to use AI safely in a complex legacy environment, where reliability is not optional.

Early GitHub Copilot experiments showed promise, but also highlighted a common limitation: AI tools cannot reliably account for business and system constraints that are not captured in the code. In this environment, performance "optimizations" can easily introduce regressions when the tool does not understand workload patterns, processing requirements, or rules the team has learned over years.

The urgency was not about one feature or one codebase. Merge needed to increase capacity across multiple teams without adding headcount, and to move from ad hoc experimentation to consistent workflows engineers could reuse.

DevClarity partnered with Merge on a structured engagement designed to turn AI tool access into repeatable, reliable workflows. The work followed a clear sequence: environment setup to prepare the team with their AI coding tools, foundational training across the engineering organization, and then two focused hands-on efforts aimed at common legacy modernization challenges.

In addition to DevClarity's AI Coding Foundations training the Merge team was interested in applying their learnings in two hands-on efforts where the workflows were tested against real constraints:

Performance optimization workstream. The team optimized a legacy component responsible for high-volume medical data processing. Each processing cycle carried significant overhead, and performance testing required multi-day validation. Earlier optimization attempts had also introduced regressions. The goal was to improve efficiency while maintaining reliability.

Data access migration workstream. After the performance work proved the approach, Merge brought DevClarity back to modernize a repeated SignalR data-access migration pattern across multiple services. The key challenge was consistency: creating a workflow engineers could reuse across the codebase without reinventing the approach each time.

The overall solution applied by DevClarity helped catalyze AI coding adoption across the team and delivered impact in key areas:

• • AI Coding Foundations training. Trained engineers across multiple teams on repeatable AI practices, prompt basics, and environment setup, including context rules that reduced bad suggestions.
• • Context rules for legacy systems. Captured business and system constraints that are not obvious from the codebase, including processing requirements, patterns to avoid, and known limitations.
• • Plan-first workflow. Required a written plan that the AI and engineers refined before any code changes. In the performance workstream, the team spent significant time refining the plan in Markdown before touching code. In the migration workstream, early planning surfaced architectural mismatches before they could spread across services. Human review helped prevent production issues.
• • Behavior change through iteration. Reinforced that adoption is cultural as much as technical. Engineers learned to invest in planning documents and written iteration, which can feel slow in the moment but compounds into faster, more confident execution over time.

Throughout the engagement, each challenge turned into a documented addition to the team's AI coding playbook.

Merge reported measurable productivity gains across the organization. Developers reported 48% faster completion of repetitive tasks and 37% higher overall weekly output. AI tool use became part of day-to-day work for many engineers, with adoption spreading across teams.

Beyond the headline metrics, the durable outcomes were:

• • Two reusable, AI-native workflows. A performance optimization workflow (identify → plan → execute) and a migration workflow that can be reused for similar modernization work across the codebase.
• • Context rules that reduced bad suggestions. Documented constraints the AI could not infer from code alone helped steer work away from approaches that had caused problems in the past.
• • Legacy code review that surfaced issues. The workflow helped identify two database connection leaks in older code that had been missed in manual review, and can extend to security scans and other audits.
• • A planning-first habit. Engineers spent more time writing, refining, and reviewing plans and prompts as part of normal delivery, which improved confidence and reduced rework.
• • Artifacts committed to repositories. Prompts, context rules, and workflow patterns were checked into team repositories so they can be reused, taught, and improved over time.

What Merge received. Merge moved from isolated AI experimentation to repeatable, production-ready ways of working. The engagement left the team with reusable workflows, documented constraints that make AI assistance safer in a legacy environment, and a growing set of repository-backed artifacts that can be shared and improved as new scenarios come up.