The Hook
“Hey, I'm Savannah, and I'm excited to introduce Agentbox: a new interface for managing dozens of agents inspired by the email inbox.”
Problem Agitation
“Today, the number of agents most of us run is limited by how many we can reasonably keep track of at once in separate windows. So we got rid of the chat window model entirely and gave your agents an inbox instead.”
Product Reveal
“One that's free, open source, and makes managing over 20 agents at once trivially easy. Let me show you how it works.”
Feature Teaser
“Suppose this is the Linear board we use for tracking product issues and features to build. Normally I'd work through just a couple of these at a time with agents running in separate terminals. Instead, I'm going to send all 12 over to Agentbox at once. What happens next is that each issue gets worked on by its own agent in a dedicated thread. And at this point there's nothing for me to do, so I'm just going to go get lunch.”
Feature Teaser
“All right, I'm back, and my inbox is filled up. This is where it gets interesting. Each thread gives me a concise summary of what the agent did and what it needs from me so I can decide what happens next. If it fixed the bug and the tests pass, I press 1, and it gets merged. I do that continuously, and I can get through these exceptionally quickly.”
Feature Teaser
“Now, this agent came back with a couple design options for me to choose from rather than a finished change. I like this version, so I pick it, and the agent goes off to build that one. Sometimes none of the versions are quite right, and in that case I just type my feedback and send it back.”
Feature Teaser
“This one is a good example of why you still want a person in the loop. It got the failing test to pass by changing the test rather than fixing the actual bug. So I tell it to fix the code instead, and it gets back to work. That's really what my job turns into with this setup. I'm not watching agents work anymore; I'm reviewing what they made and deciding whether it's good enough.”
Feature Teaser
“Once those agents come back, I merge those in, too, and then I add one last thread asking an agent to write the release notes and ship it. And that's it: all 12 Linear issues are built and shipped. I didn't open a terminal once, and at no point was I sitting around waiting for an agent to finish.”
Feature Teaser
“That's the whole idea behind giving your agents an inbox instead of a chat or terminal window: you can run far more of them than you could ever watch, and you only interact with an agent when it actually needs you. Agentbox is built around the keyboard with shortcuts for most actions and a command-K menu for almost anything else. And it works with the Claude Code or Codex plan you already have, so there's no extra subscription.”
Call to Action
“And it's available to use for all Mac users as of today. Thanks so much for watching, and let me know...”
Synthesizable in Remotion · key components of the developer-inbox-launch-orchestrator template.
Multi-window terminal grids dynamically shrink, strip padding, and reorient into vertical inbox list items.
Map grid item coordinates using interpolate and spring on frame count, interpolating width, height, and border-radius into compact list rows.
Cards stacked in z-space pop off sequentially to the top-right upon pressing keyboard number 1.
Use an array of cards indexed by current review item. On keypress frame, trigger spring-based translateY(-120px) and opacity fade.
Hundreds of micro-agent task tiles pulse status indicators, blurring non-urgent cards when filtering down.
Render CSS grid of 100+ compact cards. Apply CSS filter blur and opacity driven by interpolate when filter boolean activates.
Individual mechanical keyboard keys float in dark 3D perspective and snap sequentially into an ortho keyboard footprint.
Wrap keycap elements in a 3D perspective container; interpolate rotateX, translateY, and translateZ per key with staggered delay.
Purpose. Identify multi-terminal agent overload and introduce inbox metaphor.
Execution. Split view between founder webcam and graphic title card with technical grid background.
Purpose. Demonstrate effortless 12-task parallel assignment and autonomous execution.
Execution. Command execution screen transitioning into clock fast-forward while user steps away.
Purpose. Showcase keyboard triage, design selection, and catching rogue agent test fixes.
Execution. Quick card-stack keyboard merges, 3-way layout picking, and annotated inline code diff.
Purpose. Prove horizontal scalability to 100+ agents and integration into existing models.
Execution. Macro-zoom out to 117-agent canvas, floating keycap navigation, and plug-in wire diagrams.
Purpose. Deliver immediate Mac desktop availability and free open-source proposition.
Execution. High-contrast hero typography card accompanied by corner founder live badge.