AI agent placement follows data access

Diving deeper into

Sam Hall, CEO of Wafer, on AI agent form factors

Interview
A lot of where companies decide to place their products is determined by what access to data they're most excited about.
Analyzed 5 sources

The real product decision is not app versus browser versus device, it is which layer gives the company the raw behavioral data needed to make the AI feel predictive instead of reactive. Granola sits at the microphone and system audio layer so it can hear any meeting without joining as a bot. Rewind sat at the screen layer so it could capture everything visible across apps. Wafer pushes the same logic further, toward the operating system, because that is the only layer that can see activity across apps and later turn that context into actions.

  • Granola chose desktop because meeting platforms only expose narrow slices of data. Its app watches microphone and system audio permissions, calendar context, and call activity across Zoom, Meet, Teams, and Slack, then turns that stream into transcripts and follow ups. That gives it cross platform coverage without depending on each meeting app to cooperate.
  • Rewind and then Limitless followed the same pattern from a different sensor. The original product captured what was on screen to build searchable memory across work done in many apps, which is much broader context than any single app API provides. That made the computer itself, not the browser tab, the useful place to sit.
  • Wafer extends this from passive capture to system level orchestration. A launcher can see which app opens, and an assistant can call only the actions developers expose, but an OS fork can observe behavior across the whole phone. That is why the strategic prize is not just better interface placement, it is owning the read layer that later powers the write layer.

The next wave of AI products will keep moving downward in the stack wherever broader context creates a visibly better experience. Products that only live inside one app will feel narrow. Products at the desktop, assistant, or OS layer will increasingly act as the common memory and decision engine above many apps, with the apps themselves becoming more like back end services and data sources.