Engineering @ Scale — 2026-04-18#

Signal of the Day#

Figma’s implementation of the Model Context Protocol (MCP) demonstrates that reliable LLM-driven features require exposing strict, deterministic APIs for state extraction rather than relying on generative guessing. By injecting capture scripts to extract running DOM data and programmatically mapping it to native canvas layers, they solved the chronic fragility of code-to-design pipelines.

Deep Dives#

Rewritten Fiber Runtime for Resource Efficiency · Effect · Effect v4 Beta: Rewritten Runtime, Smaller Bundles and Unified Package System Effect completely rewrote its core fiber runtime for the v4 beta release to heavily optimize memory usage and drastically reduce final bundle sizes. To solve dependency hell—a massive operational drag in scaling TypeScript monorepos—they consolidated their entire ecosystem of packages to share a single, unified version number. The team also formalized an “unstable modules” boundary, creating a safe architectural path for rapid feature iteration without compromising the stability of core API contracts. This approach highlights a mature framework lifecycle: prioritizing low-level runtime efficiency while aggressively managing dependency sprawl and API versioning.

Eliminating Node.js Dependency Overhead · Pulumi · Pulumi Adds Full Bun Runtime Support Pulumi escalated Bun from a simple package manager option to a first-class execution environment in version 3.227.0. Infrastructure teams can now declare runtime: bun in their Pulumi.yaml configuration and execute their entire infrastructure-as-code graph natively, without installing Node.js at all. Eliminating the Node.js runtime requirement is a significant operational win, as it structurally removes a heavy dependency from CI/CD pipelines and reduces the overall toolchain surface area. For teams managing massive deployment footprints, this shift minimizes initialization overhead and streamlines containerized deployment workflows.

Generative AI for Operational Control Planes · AWS · AWS Announces General Availability of DevOps Agent for Automated Incident Investigation AWS has moved its generative AI-powered DevOps Agent into general availability, positioning intelligent automation directly within the operational workflow. The assistant is architected to parse environments and automate routine incident investigation, deployment analysis, and troubleshooting tasks across AWS systems. By integrating AI directly into the infrastructure control plane rather than treating it as a disconnected chat interface, AWS provides a blueprint for reducing mean-time-to-resolution (MTTR) for system failures. This reflects a growing industry mandate to automate complex operational runbooks by combining large language models with deep, programmatic access to platform telemetry.

Agentic Workflows via Model Context Protocol · Figma · EP211: How the JVM Works Figma tackled the deeply complex design-to-code and code-to-design synchronization problem by exposing deterministic tools through a Model Context Protocol (MCP) server. Rather than relying on naive visual scraping, the agent calls an MCP tool that injects a capture script directly into a running browser. When a user selects a UI component, the script extracts structured DOM data, and the server maps it deterministically into native Figma primitives like auto-layout groups and editable text. Conversely, for design-to-code, the agent explicitly requests get_design_context to pull exact layout definitions before generating React, Vue, or Swift code. This is a masterclass in AI integration: utilizing LLMs solely for translation and code generation while enforcing strict API boundaries for state extraction.

Patterns Across Companies#

A clear pattern this period is the aggressive reduction of intermediate layers to improve execution speed and operational predictability. Both Pulumi’s native Bun execution and Effect’s rewritten fiber runtime demonstrate that engineering organizations are stripping away dependency bloat to achieve faster cold starts and smaller deployment footprints. Concurrently, platforms like Figma and AWS are proving that the next wave of AI capabilities relies entirely on deeply integrated, structured context APIs—replacing ad-hoc prompt engineering with deterministic data extraction.