Day 2 · Recorded 7 May 2026
Harvey Spectre: collaborative cloud agents built around team alignment
If your coding agents still live on one laptop, every run becomes a private artifact: hard to share, hard to hand off, and easy to lose context from. Joey Wang shows how Harvey's Spectre moves that loop to the cloud, treating a Slack thread as the session where engineers, PMs, researchers, and legal experts can align around one agent run.
Joey Wang, Engineering Lead, Harvey
What's in this session
Local agents are useful for fast ideation, but they do not naturally become team infrastructure. Harvey's bet is that the agent run should be durable, permissioned, and collaborative: a shared object that carries human context, logs, memory, and product discussion forward.
In this session, Joey Wang walks through Spectre, Harvey's internal collaborative cloud agent platform. The session connects coding workflows to the way legal teams already work: a GitHub repo as a kind of matter, a Slack thread as the collaboration surface, and scheduled agent runs as a way to turn repeat work into dashboards and triage loops.
For platform and engineering leaders, the takeaway is a sharper framing of the real bottleneck. The hard part is not only getting an agent to write code; it is keeping everyone aligned around the context, permissions, and follow-through that make agent work usable by a team.
Inside the recording
- 00:00 Moving coding agents from laptops to the cloud
Why Harvey treats agent runs as durable, shareable sessions instead of local tabs. - 04:00 The moment Spectre became necessary
Local agents unlocked parallel ideation, then ran into context, permission, and handoff limits. - 12:00 A GitHub repo as a legal matter
How Harvey maps code workflows onto the collaborative artifacts lawyers already understand. - 16:00 A Slack thread as the agent workspace
Lawyers, PMs, engineers, logs, memory, and Spectre all contribute context to one session. - 24:00 Scheduled agents and repeat work
How recurring sessions turn one-off prompts into dashboards, triage loops, and team workflows.
More sessions on agent infrastructure
- Backgrounding the Toil: Inside Uber's Agent-Ready Developer Platform — Nikhil Ramakrishnan, Uber
- Enabling AI Tools Without Losing Control: Lessons from a Regulated Bank — Suhail Patel, Monzo
- What I learned building a software factory in public — Zacharias Malguitou, software-factory.dev