Dropbox engineering blog

Dropbox explains how Reclaim evolved into an AI-native calendar assistant without replacing its existing scheduling foundation. The design combines an agent loop, controlled tools and context, shared Schedule Actions, MCP integration, and a Redis-backed Preview Mode so users can review calendar changes before applying them.

Dropbox CTO Ali Dasdan and engineering leader Uma Namasivayam explain how the company evaluates AI productivity beyond code volume, using workflow outcomes, quality, throughput, and business impact. They share lessons on addressing bottlenecks, measuring ROI, deploying coding agents, and preserving human judgment as AI adoption scales.

Dropbox announces a modernized API documentation experience with interactive endpoint testing, embedded AI assistance, improved search and navigation, and complete request, response, schema, and error details. Developers can also connect AI tools through the provided MCP server.

Dropbox describes an in-house cookie auditor that uses Playwright and isolated browser sessions to validate consent behavior across more than 200 web surfaces. The system tests regional and GPC experiences, tracks cookies before and after choices, discovers URLs from traffic data, and produces weekly reports for Privacy and Engineering teams.

Dropbox describes a system-level approach to infrastructure efficiency as AI-driven demand grows, covering capacity planning, fleet balancing, power-saving Deep Sleep, storage density, and hardware lifecycle management. It also details a rack power redesign that supported higher-density seventh-generation servers without rebuilding the data center.

Riviera is Dropbox’s universal content processing platform, and this post explains how it has evolved over nearly a decade to improve content transformation across products. The article focuses on how the platform adapted for AI-related workloads while continuing to support broader infrastructure needs.

We used DSPy to improve LLM judges and optimize our chat experience, creating an evaluation-driven feedback loop that produced better outputs.

Using an agentic AI system to surface threat models during code review and spot gaps between security requirements and implementation.

How Dropbox is moving from AI tools that assist engineers to agentic systems that can execute scoped tasks, and how we’re building platforms to support those workflows.

Nova lets engineers run multiple coding sessions in parallel and lets internal systems use AI agents as part of automated workflows.

By turning compaction into a layered, adaptive pipeline and strengthening our monitoring and controls, we made Magic Pocket more resilient to workload changes. The post describes the pipeline changes and operational improvements that increased storage efficiency and robustness under varying workloads.

This post describes strategies Dropbox used to reduce monorepo size to improve developer velocity. It covers approaches such as removing unnecessary files, optimizing build artifacts, and adjusting dependency management to speed up checkouts and CI workflows.

Dropbox used DSPy to turn prompt engineering for their relevance judge into a measurable, automated optimization loop. This improved task performance, reduced costs, and increased reliability in production.

Describes how Dropbox combines human labeling with LLM-assisted labeling to improve search ranking for Dash. The post explains the labeling workflow, quality-control measures, and the impact of mixed labeling on search relevance and model performance.

Explores techniques for low-bit inference that make AI models more efficient and cost-effective for products like Dropbox Dash. The article covers approaches to reduce resource use, trade-offs in precision, and practical implications for deploying models at scale.

Recaps an executive roundtable at Dropbox about AI tooling and its effects on engineering productivity. The piece summarizes perspectives on AI coding tools, early results from adoption, and open questions about best practices and impact.

A deep dive with Dropbox Engineering VP Josh Clemm into how knowledge graphs, indexes, MCP, and DSPy are used to improve Dash. The article discusses how these systems interact to support retrieval, ranking, and prompt optimization.

Details the feature store that powers real-time AI functionality in Dropbox Dash and how it supports ranking and retrieval. The post covers architecture choices, data pipelines, and how features are served to models in production.

Highlights the experiences and projects from Dropbox’s 2025 summer intern class, focusing on growth, innovation, and community building. The post showcases notable intern contributions and program design elements that support development and connection.

Explains how context engineering helps Dash focus models on the most relevant information to produce better outcomes. The article outlines techniques for constructing context, prompt strategies, and how these practices improve agentic AI behavior.