Beyond Text: Building Multimodal Workflows
Multimodal is most powerful when each input feeds the next step. Drop in a PDF, ask for the gist, turn it into a list, then dictate the follow-up, all in one thread.
TLDR
Most people meet multimodal AI as a list of capabilities. The real shift is that a single conversation can flow between formats without losing the thread.
One conversation, many inputs
Photograph a whiteboard from Friday's planning session, drop it in, and ask for a structured task list. From the same thread, attach last quarter's report and ask which goals carried over. No re-explaining, no copy-paste between tools.
Design for the hand-off
Think in hand-offs, each input becoming raw material for the next step:
- Capture with an image or document instead of retyping.
- Distil into a summary, list, or table you can act on.
- Transform that output into the thing you actually need.
- Dictate the follow-up by voice when your hands are busy.
The formats change. Your judgment is the constant that ties them together.
Keep the human at the seams
The value comes from staying in the loop at the joins: confirming the task list before it becomes a project, glancing at the comparison before it goes in the deck.
