AI summary by TwinMind
AI-powered workflow automation with Zapier co-founder Mike Knoop
Sequoia Capital
10m 37s
Introduction and Context
- Mike (co-founder of Zapier) presenting at conference
- Follow-up to V1 AI actions launch from previous year
- Update: executed over 500,000 AI actions on behalf of users in past year
- Total: over 50 million AI tasks run on Zapier platform to date
- Task = completely automated, hands-off process
Problem with Current Zapier Classic
- Canvas view interface requires high learning curve
- Lots of structure, configuration, nitty-gritty details needed
- Creates barrier preventing successful adoption by many users
- Puts "huge depression" on number of people who can use Zapier
New Product: AI Bots in Zapier Central
- Ground-up reimagining of AI automation
- Goal: help more people discover, use, and adopt automation using AI
- Product name: Zapier Central
- Features AI Bots that can be customized through "behaviors"
Live Demo Walkthrough
Initial Setup
- Started with blank slate bot (standard chatty interface)
- Demonstrated customization through behaviors
- Focus on background automation rather than conversational interaction
Behavior Creation Process
- Natural language prompt: "Schedule every morning get the list of today's meetings on my gcal summarize find short bullets emojis send the summary to me and slack add a fun automation inspired quote at the end"
- Bot suggested schedule trigger for 5 AM daily
- Required equipping bot with specific actions:
- Google Calendar find event (specific calendar selected)
- Slack direct message (to user)
Testing and Execution
- Introduced "threading" concept for grouping context in single task
- Bot retrieved calendar events, provided preview
- Multiple confirmation steps during execution
- Generated summary with emojis and automation quote: "automation is like a good cup of coffee wakes you up with the possibilities of the day"
Feedback and Learning Capabilities
- User provided feedback about slack message formatting
- Bot used "updating instructions" tool to incorporate feedback
- Feedback gets integrated into future autonomous runs
- Self-healing capabilities observed in production
Technical Capabilities and Advantages
Zero-Shot Inference
- Bots can guess and infer actions without explicit configuration
- Leverages AI actions technology from past year
- Some users comfortable with fully autonomous decisions
- Others need deterministic control for broader use cases
Self-Healing Properties
- Handles scenarios that would break traditional zaps
- Example: column name changes in Google spreadsheets
- Natural language systems can debug automatically and loop to fix issues
- Eliminates need for manual debugging and learning structured canvas view
User Feedback Integration
- Users can give thumbs up/down, natural language feedback
- Strong user perception of training individual bot instances
- Feedback incorporation planned for next few weeks on roadmap
- Focus on consistency and reliability for AI automation
Product Details and Availability
Scale and Integration
- Demo showed 2 apps but works with all 7,000+ Zapier integrations
- Supports any pairwise combination of apps
- Bots can perform multiple actions at once
- Available immediately at central.zapier.com
Naming Convention
- "Bots" name chosen based on user data analysis
- Over 1 million users have named their zaps with "bot" in the name over 10 years
- Seemed natural and social way to refer to these systems
Access
- No wait list
- Available today with existing Zapier account or new free account
- Immediate access to try the platform
