Building AI Developer Tools Using LangGraph for Large-Scale Software Development

uber · Tech · year n/a · source Agentic: yes · promotional bias: low · source-backed · extracted with prompt 0.5.6 (schema v2)

Generated from extraction.json — do not edit by hand. Every claim cites verbatim source evidence; stated means the quote directly asserts the claim, implied means it supports it indirectly.

Problem

Domain: LangGraph platform for large-scale developer tools

Uber needed a standardized way for teams to build AI developer tools for 5,000 engineers without every product reinventing agent orchestration and internal-system integration.

Constraints:

• Agentic patterns were proliferating across Uber.
• Multiple products needed shared primitives and internal-system integration.
• The company had hundreds of millions of lines of code.
• Tools had to fit existing developer workflows.
• The organization wanted intentional tech transfer from product-specific agents into shared infrastructure.

Architecture

Uber created Lang Effect, an opinionated framework wrapping LangGraph and LangChain for internal systems. Product teams build AI tools such as Validator and AutoCover as composed graph workflows, then extract reusable primitives so other teams can build faster on standardized agent infrastructure.

Diagram confidence: reconstructed (reconstructed from prose).

flowchart LR
  lang_effect_framework["Lang Effect framework"] --> langgraph_workflows["LangGraph workflows"]
  lang_effect_framework --> uber_internal_systems["Uber internal systems"]
  lang_effect_framework --> reusable_primitives_library["Reusable primitives library"]
  langgraph_workflows --> validator_product["Validator product"]
  langgraph_workflows --> autocover_product["AutoCover product"]
  validator_product --> reusable_primitives_library
  autocover_product --> reusable_primitives_library
  autocover_product --> validator_product

Components

Data flow

A developer workflow such as code validation or test generation is modeled as a LangGraph graph. Specialized agents and deterministic tools execute parts of the workflow, internal Uber integrations are handled through Lang Effect, and reusable components are extracted from successful products into shared primitives.

Why this architecture

The graph/platform architecture addressed organizational scale: product-specific agents could ship quickly, while useful pieces could be transferred into reusable infrastructure instead of fragmenting across teams.

Models used: LangGraph · LangChain · Lang Effect · LLM-based agents

Scale

• Uber supported 5,000 developers.
  “With 5,000 developers working on hundreds of millions of lines of code”summary, stated
• Uber serves 33 million trips daily across 15,000 cities.
  “serving 33 million trips daily across 15,000 cities”summary, stated
• Validator drove thousands of daily fix interactions.
  “thousands of fix interactions occurring daily as developers address code quality issues”summary, stated
• AutoCover saved an estimated 21,000 developer hours.
  “estimated 21,000 developer hours saved through automated test generation.”summary, stated

Alternatives considered (roads not taken)

Let every team build isolated agent workflows

Rejected because: Uber needed standardized tooling as agentic patterns spread.

Keep product components product-specific

Rejected because: The strategy explicitly extracts reusable components from initial products.

Patterns

multi-agent-orchestration (agent platform)

LangGraph workflows coordinate specialized agents for validation, generation, execution, and analysis.

tool-use-orchestration (developer workflow)

The framework integrates AI agents with IDE, build, coverage, test, and internal systems.

durable-execution (graph workflows)

Agent workflows are modeled as reusable graphs rather than ad hoc prompt calls.

Tradeoffs

Build an internal wrapper around LangGraph and LangChain.

Chose Lang Effect over Using off-the-shelf orchestration directly in each product.

Rationale: Uber needed internal-system integration and standardized patterns across teams.

Pursue intentional tech transfer.

Chose Extract reusable primitives from products over Let product-specific implementations remain isolated.

Rationale: This accelerates future AI tools and avoids duplicated agent infrastructure.

Gotchas (what bit them)

Agentic tooling can fragment without shared primitives.

Root cause: Multiple teams independently build similar agent workflows.

Resolution: Wrap LangGraph/LangChain in an internal framework and extract reusable components.

Developer adoption depends on workflow integration.

Root cause: Tools that force context switching interrupt engineers.

Resolution: Integrate Validator and AutoCover directly into IDE workflows.

Evaluation

Uber evaluated the platform through product adoption and impact from tools built on it, including Validator interactions and AutoCover coverage/hours metrics.

Metrics: 5,000 developers. · Hundreds of millions of lines of code. · 33 million trips daily. · 15,000 cities. · 21,000 developer hours saved.

Outcomes

• Uber created Lang Effect as a shared internal agent framework.
• Validator and AutoCover demonstrated reusable graph composition.
• The platform helped deliver thousands of daily fixes and major test-generation savings.

The platform supports a 5,000-developer organization and hundreds-of-millions-line codebase.