At the Compile workshop, I used Grok Bot to capture speech-to-text notes and save them in Notion. Other bots worked from that material on a presentation, a blog draft, a social post, and an editorial review. That let me stay engaged with the event while those pieces moved forward.
Jet Semrick's workshop on 16 September 2026 covered the ingredients behind that kind of work: context, connectors, recurring tasks, and outcomes. These notes cover the recorded portion of the session.
Cursor Compile London Series
Chats, Copilots, and Assigned Work
Jet described three stages:
- Chat: answer a question; the person turns the answer into action.
- Copilot: work alongside someone inside a particular domain or application.
- Assigned agent work: take a directive, use connected tools, and produce a unit of work, including recurring routines.
Jet described the limits of a small founding team: several kinds of work need doing, but the same few people are responsible for all of them. His examples extended to enterprise departments, where independent tasks also compete for people's attention.
Three Parts of a Bot
| Part | Session examples |
|---|---|
| Initial context | Name, purpose, and the job it is expected to do |
| Connectors | CRM, warehouses, contacts, email, and other working applications |
| Outcomes | Meetings arranged, investigations completed, or pull requests shipped |
The interface was presented as messaging: give one bot meeting-preparation work while another handles an engineering task. Feedback and shared memory were intended to retain preferences and recurring workflows.
The Cloud Environment
Jet described an individualised, persistent cloud environment with files, context, and tools. Tasks could continue without occupying the user's laptop or depending on the lid staying open.
Browser and computer-use tools extended the work to applications without suitable APIs. The examples included navigating interfaces, reproducing bugs, simulating users, and preparing test results.
These were capabilities described in the workshop, not availability or uptime measurements from my own testing.
Recurring Tasks
The suggested tasks included information retrieval, issue investigation, inbox triage, calendar work, and meeting preparation.
Strategic work remained with people: talking to users, setting creative direction, choosing priorities, and making trade-offs. An agent could prepare information for those decisions.
Jet reported 418,000 weekly users, including 224,000 using the product on at least four days a week. Those are workshop figures, not independently audited adoption data.
Taking Session Notes With Grok Bot
I used Grok Bot to capture speech-to-text notes and save them into Emre's Brain in Notion. Five bots had separate jobs:
- Session notes: capture speech-to-text notes.
- Presentation: turn the captured material into a deck.
- Social: draft event material for X.
- Blog: draft articles from the sessions.
- Editorial: review the writing.
The presentation, social post, and article used the same captured notes with different briefs. I still reviewed the final outputs. The extra capacity was useful because I could keep listening, then return to the drafts and decide what belonged in them.
The recording ends before the workshop finishes, so there are no notes here for the remaining exercises.