Super Data Science: ML & AI Podcast with Jon Krohn · Jon Krohn

951: Context Engineering, Multiplayer AI and Effective Search, with Dropbox’s Josh Clemm

·1 hr·4 clips

Josh says multiplayer AI must understand you, your team, and connected data across Slack, docs, and people.

1. Super Data Science: ML & AI Podcast with Jon Krohn centers this episode on Dropbox Dash, an AI search layer for work apps. 2. Jon Krohn hosts Josh Clemm, Dropbox Vice President of Engineering, whose team builds Dash and whose background includes leading 400 engineers at Uber. 3. The episode asks how knowledge workers can search across Slack, Google Drive, Jira, browser tabs, and Dropbox without juggling 12 to 20 search bars. 4. Josh says Dropbox Dash is broader than file storage because it ingests cloud content, images, docs, and work apps into one searchable system. 5. He argues that traditional enterprise search is broken because each app has its own search bar and scattered content fragments the working set. 6. Josh describes multiplayer AI as an AI teammate that understands the user, the team, the org, and connected data rather than a single-person chatbot. 7. He points to Dropbox Stacks as a collaborative place to combine URLs, slide content, PDFs, comments, and chat queries for a team. 8. Josh says browser extensions matter because many managers work in the browser, where Dash can use the current page context and browser history. 9. He also notes that browser-native AI brings prompt injection and data exfiltration risks when agents are given tool use. 10. Josh says work slop happens when companies adopt AI without clear goals, producing generic or hallucinated output. 11. He compares good AI use to writing an essay draft: first draft, revision, stronger wording, and added research. 12. Josh says Dash uses ingestion, normalization, and indexing, including both lexical and vector indexes. 13. He names BM25 as the keyword-search workhorse and says it matters for part numbers, while semantic search helps with requests like vintage cars. 14. Josh says hybrid retrieval is needed because many customers want both keyword precision and semantic matching. 15. He explains context engineering as a response to growing context windows that still degrade when too much information is loaded. 16. Josh points to the Nolema benchmark, multi-turn conversation failures, and Dash’s move from multiple tools to a single super tool to reduce accuracy drops. 17. He says the episode is practical and technical, with examples from Dropbox, Uber, fantasy football, and side projects like Yaddle.ai. 18. Jon keeps the conversation conversational and specific, pushing on product design, AI safety, and workplace usefulness. 19. Useful for product builders, search engineers, and AI ops leaders. 20. Less useful for listeners wanting light AI news or beginner-only explanations.

As heard by us

Context engineering is the real story behind AI search at work.

The episode frames a familiar work problem: knowledge gets scattered across Slack, Drive, email, project tools, and other corners, which makes the case for Dropbox DASH as an AI-powered search and knowledge layer across work apps.

Read the full review in PlayNext →

Why you'd press play

If work search feels scattered across too many apps, start here.

Read the full recommendation in PlayNext →

Listen to the show on

Apple Podcasts Spotify

Play clips from this episode More from Super Data Science: ML & AI Podcast with Jon Krohn