Brief in progress — review the script and demo before recording.
Angie, this video draws directly on your experience building Goose. Most students will not build a system like it, but many will need those lessons when adding AI features to their own platforms. Even those who mainly use existing agents will benefit from a general understanding of where to look and what to consider when building these systems.
- Daily Work: agents still work with humans, but increasingly they also interact with one another and bring only the most important tasks to human attention. Prompts, attachments, and inline context may come from either humans or agents. Agents need an easy way to communicate through both text and file references, as well as visibility into one another’s existence and progress.
- Workers: subagents and workers (separate instances) can work individually or form teams and other structures around a shared task. This raises challenges around context management, synchronizing changes, and shared resources. Agents may need worktrees, their own sandboxes, servers, or even dedicated machines. They also need to work alongside humans using shared specifications and issue trackers, so we should show how to organize that collaboration.
- Management: working with more than a few agents quickly becomes difficult for a human to manage. We need structures, tools, and processes that let agents manage themselves while staying within scope. A CLI or UI that lets agents communicate with and spawn other agents is important here. These setups also need to support agents running both locally and in the cloud.
- Ambient: as agents take on longer-running tasks, we need to explore moving them to the cloud and making them proactive through triggers, hooks, events, heartbeats, and schedules. We should show a few examples of how to think about agents that work without supervision or with minimal oversight.
We may want to showcase scenarios such as:
- A dive into the architectural concepts behind Goose. It offers a useful foundation for introducing custom agentic systems, while the remaining details can be covered in the written lesson.