Back office ops · Production

Dropbox Dash uses context engineering to improve agentic AI accuracy and performance

The problem

Dash's original RAG pipeline handled basic retrieval well but was insufficient for agentic tasks; adding more tools caused analysis paralysis, context rot, and accuracy degradation on longer-running jobs.

First attempt

Using multiple separate retrieval APIs proved unreliable—the model often failed to call all required sources—and MCP tool definitions consumed excessive tokens, degrading performance.

Workflow diagram · grounded in source
1
User submits agentic request
trigger
“open the editor and write an executive summary of the projects that I own”
2
Planning agent determines next action
ai_action
“The main planning agent decides when a search is needed and delegates the actual query construction to a specialized agent with its own prompt”
3
Specialized search agent constructs query
ai_action
“Query construction turned out to be a difficult task on its own. It involves understanding user intent, mapping that intent to index fields, rewriting queries for better semantic matching, and handling edge cases such as typos, synonyms,…”
4
Unified Dash Search index retrieves results
integration
“replacing all of those retrieval options with a single, purpose-built tool backed by the Dash universal search index”
5
Knowledge graph ranks and filters results
ai_action
“layered a knowledge graph on top to connect people, activity, and content across those sources. (A knowledge graph maps relationships between these sources so the system can understand how different pieces of information are connected.) …”
6
Main agent generates final response
output
“the model only sees content our platform has already determined to be relevant, which makes every piece of context meaningful”
Reported outcome

Dropbox adopted three context engineering strategies—consolidating retrieval into a single unified index, filtering context via a knowledge graph, and delegating complex query tasks to a specialized search agent—resulting in faster, more accurate agentic performance with leaner context usage.

Reported metrics
Model decision making speed and accuracyslower, less accurate decision making
Dash accuracy on longer-running jobsdegraded
Model performance with leaner contextleaner contexts don't just save resources; they also make the model smarter
Reasoning clarity and plan efficiencymakes its reasoning clearer, its plans more efficient, and its context use more focused
Reported stack
DashModel Context ProtocolDash SearchConfluence, Google Docs, Jira
Source
https://dropbox.tech/machine-learning/how-dash-uses-context-engineering-for-smarter-ai
Read source ↗

Frequently asked questions

What did this team achieve with this AI workflow?

Dropbox adopted three context engineering strategies—consolidating retrieval into a single unified index, filtering context via a knowledge graph, and delegating complex query tasks to a specialized search agent—resul…

What tools did this team use?

Dash, Model Context Protocol, Dash Search, Confluence, Google Docs, Jira.

What results were reported?

Model decision making speed and accuracy: slower, less accurate decision making; Dash accuracy on longer-running jobs: degraded; Model performance with leaner context: leaner contexts don't just save resources; they also make the model smarter; Reasoning clarity and plan efficiency: makes its reasoning clearer, its plans more efficient, and its context use more focused (source-reported, not independently verified).

What failed first in this deployment?

Using multiple separate retrieval APIs proved unreliable—the model often failed to call all required sources—and MCP tool definitions consumed excessive tokens, degrading performance.

How is this back office ops AI workflow structured?

User submits agentic request → Planning agent determines next action → Specialized search agent constructs query → Unified Dash Search index retrieves results → Knowledge graph ranks and filters results → Main agent generates final response.