Podcast Summary: Airbnb’s Open-Source GraphQL Framework with Adam Miskiewicz

Podcast: Software Engineering Daily
Host: Gregor Vand
Guest: Adam Miskiewicz (Principal Software Engineer at Airbnb)
Date: February 5, 2026

Overview

This episode explores Viaduct, Airbnb’s open-source, large-scale GraphQL platform. Adam Miskiewicz discusses its journey from concept to critical internal infrastructure, the architectural principles behind its development, their transition from "classic" to "modern" Viaduct, and the technical and organizational lessons from scaling GraphQL at Airbnb. The conversation also touches on open-sourcing the framework and the evolving landscape of backend engineering in an AI-driven world.

Adam Miskiewicz: Background and Role at Airbnb

• Experience Journey ([02:48])
  • Adam transitioned from agency work and small startups to big tech at Airbnb, bringing a generalist mindset.
  • Has been at Airbnb for nearly eight years, witnessing the engineering org’s growth from 500 to 3,000 people.
    • “I have been a software engineer for...pushing 20 years...I went from small company to big company instead of big company to small company.” ([02:48], Adam)

The Origins of Viaduct and Airbnb’s GraphQL Story

• Early Days: Moving from Monolith to Microservices ([05:20])

  • Airbnb’s evolution began with a Ruby on Rails monolith, transitioning to microservices with layers: data services, derived data services, and presentation services.
  • Early GraphQL adoption focused on the "presentation" service layer, enabling strongly-typed APIs for client engineers.
  • Initial approach: automatic conversion of existing Thrift schemas to GraphQL and schema stitching for a unified endpoint.
    • “Our first general approach...was just kind of convert the thrift schema to GraphQL and stitch all of the presentation service schema together into one GraphQL schema and one GraphQL endpoint.” ([07:26], Adam)

• Internal Communications and Migration ([11:21])

  • Conversion from Thrift to GraphQL was mostly seamless for backend teams.
  • Client engineers (iOS, Android, web) were eager adopters, championing the change.

Architectural Philosophy Shift: From Service-Oriented to Data-Oriented

• The “Data Architecture Working Group” and Viaduct’s Foundation ([12:56])
  • In 2019, a working group with diverse disciplines at Airbnb was established to solve fragmented data challenges, leading to the Unified Data Store (UDS) and the early Viaduct prototype.
  • Viaduct’s first major usage: powering features like Airbnb’s “Trips” and “Wishlists.”

What Is Viaduct? Naming and Core Principles

• What’s a Viaduct? ([17:01])

  • It’s a bridge (misconceptions about water-carrying clarified!).
  • Name metaphor: Viaduct connects services, bridging disparate systems.

• Philosophy and Key Principles ([18:00])

  • Three guiding tenets:
    • Essential schema
    • Hosted business logic
    • Re-entrant API
  • Hosted business logic: Teams are encouraged to host business logic directly in Viaduct, a contrast to best practices in many GraphQL systems.
    • “From the beginning, we’ve been encouraging teams to host their business logic directly in Viaduct. This runs counter to what many consider best practices in GraphQL.” ([18:42], Adam)

• Unified Schema vs. Federation ([20:16])

  • Viaduct enforces a single schema but subdivides logic with “tenant modules,” unlike Apollo Federation’s microservice subgraphs.

Technical Architecture Breakdown

• Three Key Layers ([25:05])
  • Engine: High-performance execution (possibly replaceable/swappable in future).
  • Tenant API and Runtime: Strongly-typed interfaces (Kotlin), defines boundaries for multi-tenant operation.
  • Hosted Application Code: Where business logic lives, separated and structured for scale.
    • Analogy: Engine space vs. tenant space (kernel vs. user space).

Classic vs. Modern Viaduct: The Modernization Journey

• Classic Viaduct ([28:20])

  • Rapid organic adoption led to architectural issues: feature bloat, fuzzy abstractions, and runtime challenges.
  • Reliability took priority over developer experience as Viaduct became mission-critical (serving ~80% of Airbnb traffic).
    • “At this point, something like 80% of all traffic runs through Viaduct... it can’t get much more critical.” ([32:35], Adam)
  • “A victim of its own success,” needing iteration for scale, isolation, and maintainability.

• Modern Viaduct ([35:45])

  • Clear boundaries between engine, tenant API, and code.
  • Shift from dynamic engine API to statically-typed tenant API for a simpler, less leaky abstraction.
  • Core design principle: Everything is a resolver (decluttered mental model).
    • “Everything’s a resolver. We build that reentrancy capability into that concept and that’s how you write your code.” ([36:34], Adam)
  • Async memoization: Avoids duplicate work within a query by caching execution of identical resolver calls.
    • “Most GraphQL servers...I’ve never really seen one do that. Turns out, it has a lot of performance wins.” ([36:34], Adam)
  • Modern supports enormous queries and schema sizes unique to Airbnb’s scale (millions of QPS and ~25,000 types).

Performance at Scale and Developer Experience

• Operational Scale ([45:05])

  • Multi-million lines of code.
  • Handles over a million QPS in GraphQL operations.
  • Most of Airbnb is on Classic Viaduct; Modern is being rolled out for new use cases, with AI-powered migration strategies planned.

• DevEx for Viaduct Modern ([45:41])

  • Open source repo provides the core of Viaduct, but not all Airbnb-scale glue/infrastructure.
  • “What is there is truly the Core of Viaduct and what we run internally.” ([46:25], Adam)

On Open Sourcing Viaduct

• Open Source Motivation & Process ([42:01])
  • Supported by Airbnb’s CTO as a way to increase accountability, strengthen abstractions, and engage the tech community.
  • “It’s an accountability mechanism almost...to get validation on those ideas and be able to talk about them openly.” ([42:01], Adam)
  • Breaking internal dependencies and making Viaduct usable by others was a non-trivial engineering effort.

The Future: Backend Engineering & AI

• How AI Will Change Backend Development ([47:01])

  • Adam foresees more code being generated or managed by AI, with engineers shifting toward oversight and pattern enforcement.
    • “Software engineering is going to change a lot. It’s not going to go away...But we’re still going to be responsible for code.” ([47:01], Adam)
  • Strong, opinionated patterns and unified platforms like Viaduct will become more essential as development responsibility shifts and teams (and agents) scale.
  • “I see a pretty clear future for...simplifying architectures in large companies. Services don’t die, but maybe there’s a considerable collapse of how many services there are.” ([47:01], Adam)
  • The importance of managed platforms for both people and AI/agent workflows.

• GraphQL’s Place in an AI World ([51:49])

  • Strongly-typed, unified data graphs like GraphQL’s schema are advantageous for both humans and machines.
    • “The general idea of having a strongly typed data oriented schema that represents all of your core business data...that seems pretty powerful, for sure.” ([51:49], Adam)
  • Adam predicts a possible "GraphQL resurgence" as AI and agents increasingly need flexible, discoverable data backends.

Notable Quotes & Moments

• On Viaduct’s Role at Airbnb:
  • “At this point, something like 80% of all [API] traffic runs through Viaduct, so it can’t get much more critical than that.” ([32:35], Adam)
• On Re-entrancy and Composability:
  • “As the graph grows...the less that you need to go elsewhere to get some data, and the more that you can use the data that you already have in the graph to build features.” ([22:10], Adam)
• On Open Sourcing:
  • “It’s an accountability mechanism almost...to get validation on those ideas and be able to talk about them openly.” ([42:01], Adam)
• On AI and Backend Engineering:
  • “Having these managed platforms be like critical pieces of infrastructure in large companies...I don’t think you can do it without it.” ([49:59], Adam)
• On GraphQL’s Future:
  • “There could be a bit of a resurgence in GraphQL… The easier we can make it to write scalable backends using that technology, I think it’ll actually benefit folks in the AI world.” ([51:49], Adam)

Key Timestamps

• [02:48] Adam’s path to Airbnb and early engineering context
• [05:20] Airbnb’s monolith-to-microservice journey and GraphQL’s introduction
• [12:56] Data Architecture Working Group & Unified Data Store
• [18:00] Viaduct’s naming and core guiding principles
• [25:05] Technical architecture: engine, tenant API, application code
• [28:20] Classic Viaduct’s evolution and growing pains
• [35:45] Modern Viaduct: abstraction boundaries and performance improvements
• [42:01] Open sourcing Viaduct: motivations and process
• [45:05] Viaduct at scale: devex, migration plans, and multi-million QPS operation
• [47:01] The future: AI-driven backend development, the role of managed platforms
• [51:49] GraphQL and AI: why unified schemas matter, possible resurgence

Resources & Further Engagement

• GitHub Repository: github.com/airbnb/viaduct
• Adam’s Socials: X/Twitter @skevy
• Community: Viaduct Discord (link in repo/readme)
• Recommended Reading: Adam’s Viaduct blog post (linked in the episode)

Closing Tone

The conversation was technical yet accessible, blending organizational storytelling with practical engineering insights. Adam’s candid reflections on scaling pains, platform evolution, and the future of backend engineering offer value for both enterprise engineers and tech architects considering GraphQL or unified data platforms at scale.

Summary prepared for Software Engineering Daily listeners and those interested in open-source infrastructure, large-scale GraphQL, and the future of backend systems in an AI-driven era.