Spec-Driven Development at Scale: Claude for a Hospital Group's Data Team
Hospital data teams sit at the intersection of two unforgiving demands: clinical stakeholders who need reliable data products, and an environment where an unvetted change can have real consequences. For one large hospital group, Lineate's engineers pair Claude Opus 4.8 with Amazon Kiro, AWS's spec-driven agentic IDE, to bring structure and speed to the data team's delivery. It is a natural fit: spec-driven development is the same principle behind our artifact-driven workflow, built into the tooling itself.
Share:
Service:
- AI Product Development
Got a project?
Lineate designs and builds the data pipelines, governance, and infrastructure that power AI.
Contact UsShare:
Service:
- AI Product Development
Got a project?
Lineate designs and builds the data pipelines, governance, and infrastructure that power AI.
Contact UsProblem
The data team's backlog looked like most in healthcare: a steady stream of change requests where the hard part was rarely the code. Scoping a request against a sprawling data estate, planning implementation across multiple steps and systems, and validating that a dashboard or interface actually works for clinical users all consumed more senior time than the implementation itself. Refactoring aging pipelines kept getting deferred because nobody could spare the hands. The result was a team spending its most experienced capacity on analysis and verification instead of judgment.
Solution
Lineate configured Kiro with project-specific AGENTS.md instruction files, encoding the team's conventions, architecture, and constraints so Claude Opus 4.8 operates with full project context from the first prompt. Every request now follows a spec-driven path: Claude produces a scope analysis and a multi-step implementation plan as reviewable artifacts before any code changes, and an engineer approves the spec before execution begins.
From there, the agent handles the work that used to sit in the backlog. Automated refactoring runs against the approved plan, keeping pipelines current instead of deferred. For user-facing work, UI/UX validation loops run through Playwright: Claude drives the browser, checks the rendered result against the spec, and iterates until it passes, with the evidence captured for human review.
Results
-
Senior capacity shifted from scoping and verification to decisions, with Claude producing the analysis and plans they review
-
Refactoring moved from perpetual backlog item to routine work executed against approved specs
-
UI validation automated through Playwright loops, catching issues before clinical users do
-
Every change backed by a spec, a plan, and validation evidence, the paper trail a healthcare environment demands