CHAPTERS
- 0:01 – 0:31
Slack’s core problem: information overload and “being a great host”
Slack frames AI adoption around a longstanding challenge: information overload. The team ties this to a product principle—helping users distill noise into what matters most.
- •Slack users face chronic information overload
- •Product principle: “be a great host” by guiding attention
- •Early goal: help people identify what’s important amid noise
- 0:31 – 1:01
Defining the first AI use cases: search answers and high-volume summarization
When modern AI became viable, Slack focused on two clear, high-impact problems: improving search and summarizing large amounts of information. This set the direction for early experimentation.
- •AI efforts anchored to concrete problems, not novelty
- •Primary targets: search and summarization
- •Goal: distill high volumes of information into usable output
- 1:01 – 1:03
Early experiments with Claude and the first “holy cow” moment
Slack experimented specifically with Claude and quickly saw promising search-answer behavior. The team describes an immediate fit—an early signal that the approach could meaningfully improve the product.
- •Testing Claude specifically for search experiences
- •Strong early quality in returned answers
- •Immediate sense of product-model alignment (“mind meld”)
- 1:03 – 1:33
User impact: automated answers and summaries that save time daily
The conversation shifts to how AI features feel to users in practice: the system answers questions and summarizes content automatically. The benefit is framed as reclaiming minutes every day through reduced cognitive load.
- •Automated answering reduces manual digging
- •Summaries help users process threads and context faster
- •Time savings accrue in small daily increments
- 1:33 – 1:34
Measuring outcomes: query success rate and perceived noisiness
Slack evaluates the AI work through product metrics that reflect both effectiveness and user sentiment. They highlight improvements in search success and reductions in users reporting Slack as “noisy.”
- •Metric 1: query success rate
- •Metric 2: self-reported perception of Slack being noisy
- •Reported improvement across both measures
- 1:34 – 1:50
Beyond customer features: Claude Code for internal engineering acceleration
Slack also applies Claude internally via Claude Code to speed up engineering workflows like bug fixing. This changes how engineers spend time—more on architecture and planning, less on rote implementation.
- •Claude used for both customer-facing features and internal tooling
- •Internal use cases include bug fixing and team velocity
- •Engineering work shifts toward planning/architecture and deeper thinking
