Across three of the most complex domains Emids handles for Surescripts, Claude Code is now the standard way of working. A program-level playbook pairs each engineering track with its own Claude Code Skills, so Skills and agents take on the routine work and 27 engineers spend their time on the complex stories that move the roadmap.
CHALLENGE
Three of the hardest workstreams in the portfolio
Surescripts operates the network through which prescribers, pharmacies, and payers exchange clinical and benefit information. Emids runs a broad portfolio of work across it. These three were chosen for this story because they are among the most complex: a regulated national network, high transaction volume, complex business rules that are critical for patient care. A way of working that holds here holds in most places.
- Electronic prior authorization (platform and new product). A mature, high-volume platform had to be modernized while new coverage capability was built on top of it. Emids expanded coverage to payers and PBMs not natively connected to the network, brought prior authorization access inside the prescriber’s EHR workflow, migrated off a legacy commercial database to managed cloud PostgreSQL, and modernized the runtime and operating system.
- Customer onboarding (owned end to end). Onboarding depended on a manually coordinated certification process for every partner. Emids led development of the self-service certification capability, from requirements through implementation, review, QA, and release, so customers now certify onto the network themselves.
- Formulary and eligibility (technical initiatives). The domain was midway through a move to cloud infrastructure. Emids provides targeted engineering on those services, including the migration itself, and on the transaction-processing and audit applications that support them.
Each needed engineers who could work safely inside regulated healthcare systems and unfamiliar legacy code. Certification also meant building a new product from scratch, so the same operating model now covers both brownfield and greenfield work.
SOLUTION
One playbook, 27 engineers, agents paired to each track
Emids brought a team of 27 engineers across all three areas, and each one of them works in Claude Code every day. Emids was also deliberate about how Claude entered the work. A program-level playbook, rather than individual prompting habits, defines the practice, and each track applies it to its own codebase with track-specific Claude Code Skills and slash commands. Emids engineers wrote those Skills alongside the ones Surescripts had already built.
About 80% of the code these teams deliver is now generated with Claude Code. Governed Skills, instructions, and review steps, refined over time, are designed to ensure that code meets organizational standards and prevent rework and functional or technical defects.
That creates a clear division of labor. Skills and agents take on repetitive, mechanical work. Engineers keep the thinking work, including architecture, domain judgment, and edge cases, and carry more complex stories per sprint. Outcomes are logged continuously in a shared record as the work happens. The playbook, the Skills pattern, and the connector set carry over to other engagements, where only the codebase changes.
Each needed engineers who could work safely inside regulated healthcare systems and unfamiliar legacy code. Certification also meant building a new product from scratch, so the same operating model now covers both brownfield and greenfield work.
Claude Code across six stages of delivery
The Skills encode team standards once for all 27 engineers, and the Atlassian MCP connectors put Claude inside Jira and Confluence, where requirements and design already live. At every stage, the engineer makes the call. Claude proposes, the engineer validates, and nothing merges without review.
Stage 01: Requirements
Engineers work against Jira stories and Confluence specifications in place, through the Atlassian MCP connectors. Working with Claude, they break features into epics and stories, validate the definition of done, and settle the conformance conditions a certifying customer must satisfy.
Stage 02: Design & tech design
Architecture documents and context-aware data-flow diagrams are drafted into Confluence from the source requirements. The engineer reviews and settles on the best possible design.
Stage 03: Development
Code is generated to the team's coding standards. An AI review pass runs before human review, so reviewers receive changes already checked. Reviewer comments are applied back through Claude Code, cutting review-to-merge iterations.
Stage 04: QA
Test planning, test-case generation, and end-to-end UI automation are written to the project's established patterns, then refined by the engineer. Coverage arrives with the pull request, so testing no longer waits for sprint end.
Stage 05: Deployment
Emids prepares migration scripts, pipeline configuration, and release documentation. Surescripts owns the deployment window and executes the cutover.
Stage 06: Support
Separation of duties and strict data provenance ensures offshore teams like Emids or AI Tools do not own production or staging tickets. Claude supports that work indirectly through analysis, reconstructing event sequences and summarizing findings for the teams that own them – all while protecting PHI and Customer data.
OUTCOMES
25–30%
FASTER RELEASE CADENCE
25–30%
HIGHER STORY THROUGHPUT
45–55%
SHORTER DEV CYCLE TIME
Results beyond the headline numbers
Emids logs outcomes in the shared record as the work happens. These results cover the whole delivery lifecycle, and they come from the same division of labor: engineers spend their time on complex work while Skills and agents handle the repetitive tasks.
6 Services migrated with one pattern
A repeatable rewrite approach, applied across the engagement.
Once Standards encoded for 27 engineers
Skills run identically in all three domains, whoever is at the keyboard.
Pre-checked Shorter pull-request review cycles
An AI pass runs before human review, and reviewer comments flow back through Claude Code.
40–50% Less test-authoring effort
Reported independently by two QA engineers on separate workstreams.
80–82% AI-generated code
Claude Code now generates most of the code these teams deliver.
~18 REUSABLE SKILLS BUILT BY THE EMIDS TEAM
The Emids team built 17–18 reusable Skills that work across initiatives, on top of dozens of existing individual and Surescripts-specific Skills and commands.
