Overview
Granola co-founder and CEO Chris Pedregal and Every CEO Dan Shipper explore what work may look like when AI agents become persistent collaborators rather than isolated chat tools. Pedregal argues that meeting notes are only an entry point into a much larger opportunity: capturing rich organizational context, interpreting it accurately, and making it available wherever people work. Shipper describes work splitting between public, asynchronous delegation in tools such as Slack and focused collaboration inside agent workspaces such as Codex. Their discussion surfaces several unresolved design problems, including how users resume delayed agent tasks, how agents and humans manipulate the same live interface, and how software should represent context spanning many meetings rather than one transcript or chat thread. Granola’s strategy is consequently twofold: become dramatically better than general-purpose agents at meeting-adjacent jobs, while exposing its context through APIs and MCP so other agents can build on it. Pedregal also describes a staged product-development process, proactive generation of time-sensitive meeting briefs, and experimentation with team roles as AI changes how products are built. Both founders conclude that the winning interfaces, organizational structures, and economics remain uncertain, but specialized context, shared state, and productized versions of expert workflows are emerging as durable foundations.
Sections
Emerging Patterns
Higher-order implications derived from the founders’ discussion of agent interfaces, context, and product strategy.
- The durable layer in AI software may be shared domain state rather than a proprietary general-purpose agent. Applications that structure context, preserve provenance, and expose reversible actions can remain valuable even as users switch among increasingly capable agents.
- AI agents introduce temporal interface design as a first-class discipline: products must preserve not only task state, but also the user’s original intention and the reasoning needed to review work after attention has moved elsewhere.
- The most valuable meeting product may not resemble a notebook. As context expands across meetings, messages, relationships, tone, and organizational history, the winning interface may center decisions, actions, and situational awareness rather than documents.
- Usage frequency is an incomplete metric for ambient AI. A rarely opened artifact can still be highly valuable if it is immediately available during a consequential moment and dependable enough to change the user’s behavior.
Forecasts
Future developments explicitly anticipated by the speakers.
- Current AI products and meeting-note competition will appear minor compared with the coming transformation of computing and knowledge work.
- Workflows currently assembled by technically advanced users will be productized and adopted by ordinary users within a few years.
- Granola’s API and MCP capabilities will improve substantially as external use of its meeting context becomes a first-class company goal.
- New canonical interfaces will emerge for working across many contexts and concurrent agent tasks because chat threads are insufficient for that form of work.
- Agentic product features will eventually become expensive enough to challenge Granola’s current unit economics, even if model costs decline.
Operating Lessons
Lessons drawn from building, scaling, and using AI-native products.
- Treat emerging AI product categories as complex systems: probe with real products, observe behavior, and accumulate reliable insights instead of pretending a complete theory can be derived in advance.
- Explore many possible solution shapes before investing heavily, validate that a small group genuinely chooses the product, and only then make it reliable and scalable.
- Preserve product coherence while scaling by assigning explicit ownership of user experience, implementation, and strategy, even if traditional job titles no longer map cleanly onto the work.
- Observe advanced API and agent users as product researchers, then productize only those workflows where the company can create a materially better experience for many people.
- Track AI expenditure before it becomes a constraint, but avoid treating token volume as a proxy for employee productivity or product value.
Memorable Quotes
Statements that capture the interview’s central arguments and metaphors.
- What has come thus far will pale in comparison to what will come soon.
- Startups are like knife fights.
- The pirate's job is to just build as fast as possible to find something valuable.
- We kind of want to get out of the way until you really need us.
- AI is like the the ligaments and the muscles, and then software has to be the bones.