Use Cases — Microservices

Cross-service context, compiled ahead of time, for every repo, through one connection.

“What changed wasn’t the model. It was the context.”
Matt Nassr — Head of Global Data Engineering and AI Transformation, Optiver

The challenge

Microservices break the one thing an agent depends on: seeing the code.

A coding agent is only as good as what’s in its field of view. In an enterprise software estate with hundreds of repos, the code that explains a change lives in services the agent never sees.

01

Co-locality

Agents reason only about code they can see. Nobody has 100+ repos cloned at once, so when the order service calls payment, the logic that matters was never in context.

02

Implicit dependencies

The links between services live in config strings, shared tables, and queues, not in code. The dependencies that matter are invisible to an agent reasoning over one repo.

03

Context that doesn’t scale

Per-repo docs go stale. Internal context tools become their own product to maintain. Neither keeps pace with how fast the system changes.

order-service f(createOrder) payment-service f(charge) fulfillment-svc consumer orders table POST /charge — config URL string queue: order.created → order.fulfilled — every edge above is invisible to an agent scoped to one repository —
service / store invisible cross-service edge
“The map of how our services fit together used to live in people’s heads. Now it lives where the agents can reach it.”
Jacob Radkiewicz — CTO, ShipBob

Why the usual fixes fall short

Driver compiles complete context ahead of time. Other approaches fall short.

Per-repo docs

Accurate the day they’re written, drifting by the next merge. Across hundreds of services the staleness compounds faster than any team can keep up with.

Homegrown tools

The instinct is to build internal context plumbing. A team at ShipBob started building one, then realized within weeks it would become a year-long product of its own, and stopped.

RAG / embeddings

Retrieval samples chunks by similarity. It returns the top-K chunks that look relevant while discarding the call graph, the shared schema, the producer/consumer link, the structure that actually defines the system.

How Driver solves it

Compile every codebase. Query them through one connection.

Driver compiles exhaustive, symbol-level context for every codebase ahead of time, then synthesizes across services at the moment a task needs it. The agent reasons over the system, not a single repo.

Without Driver

agent scoped: 1 repo order-service payment ? inventory ? fulfillment ?

Reasons only about the repo in view. Everything else is a guess.

With Driver

agent one connection Driver compiled context payment inventory fulfillment

One connection resolves to compiled context across every service.

unseen / guessed Driver connection

One connection, all codebases

A single MCP integration. No co-locality requirement, nothing to clone or pre-index. The agent reaches every service through the same connection.

Cross-codebase context synthesis

Describe the task. Driver’s gather_task_context reads pre-computed context across the relevant services and returns one dense, high-signal answer — the capability that matters most when the work spans repos.

Always current

Incremental re-analysis runs on every push. Context stays accurate as the codebase moves, and nobody on the team is assigned to maintain it.

# gather_task_context(
  "why do bookings complete
   without a payment record?"
)
→ context across 5 services

On access: cross-repo context spans only the codebases a user is already authorized for. Access is enforced at the codebase level, so synthesis reads within those bounds, never around them. Deployment isolation (single-tenant SaaS or custom VPC) keeps IP contained.

Trust Center →

In practice

Three places it changes the day-to-day.

Cross-service debugging

A bug that spans five services

Bookings completing without a payment record. The defect threaded through four or five services and had been missed several times. With cross-service context, it was traced and resolved in minutes.

Automated cross-repo tasking

One Jira story, split by repo

A 200+ microservices team built a skill that breaks a Jira story into per-repo subtasks using parallel calls to Driver’s gather_task_context. The “you’ve got to know which repos to touch” constraint simply disappears.

Onboarding

A service you’ve never opened

After a reorg, engineers get an architecture overview and cross-service context for a service they’ve never touched, instead of losing weeks rebuilding the map a departed expert carried in their head.

In an unfamiliar domain, context-gathering dropped from 30–40 minutes to about 10.
Kashyap Mukkamala — Principal Engineer, ShipBob · ShipBob case study

Case study

Measured in production, across 450+ repositories and 20M+ lines of code.

ShipBob — microservices architecture

450+ repos · 20M+ LOC

3x output across the team
200+ engineers
450+ repositories
20M+ lines of code
<2 wks to deploy

Before Driver, a domain expert changing teams meant 6–8 weeks of lost cross-service knowledge. Now that context is compiled and queryable on day one.

Read the full case study →

Under the hood

Real infrastructure, not retrieval.

A symbol-complete compiler frontend reads each codebase exhaustively.

01

Compile

A symbol-complete compiler frontend builds exhaustive, structural understanding of every codebase.

02

Document

Deep Context Documents (architecture, onboarding, changelog), plus navigation primitives: code map, file/folder docs, raw source, history.

03

Synthesize

Driver’s gather_task_context runs as a headless sub-agent in a sandbox, using those same primitives across services.

04

Update

Incremental re-analysis automatically keeps the compiled context current as code changes.

Time to value

One connection. The tools you already use.

Driver connects through a single MCP integration and shows up inside the agents your team already works in — no migration, no new surface to learn.

Claude CodeClaude DesktopCursorGitHub Copilot
Production in < 2 weeks

Get started

Agentic development fails without context. We provide it.

Customer Story

How ShipBob drove a 3x increase in output across 200+ engineers with Driver