Personalized AI assistants and workflow automation
Town
Town builds personalized AI assistants called Townies for individuals and teams. They draft emails, prepare meeting briefings, work through to-do lists, and run recurring routines such as client summaries and inbox triage. Users can choose ready-made routines or describe their own. Town Decks creates editable presentations using connected business data, including Stripe figures, with reusable templates for reports and pitches.
Co-founders Jean-Denis Greze, CEO, and Tony Vincent, CPO, previously worked together at Dropbox. Greze was CTO of Plaid; Vincent led Applied AI Product at Google and co-founded Aspen and Sold, acquired by Google and Dropbox respectively. Town’s technical approach centers on persistent context about a user’s voice, relationships, priorities, and judgment. Its Town Wiki records that understanding, updates nightly, and lets users edit it directly.
Town connects assistants to workplace tools including Google Workspace, Outlook, Slack, and CRM systems; teams can share routines and integrations. Its privacy policy for AI training makes training on customer data opt-in and bars its model providers from training on customer inputs and outputs. In 2026, Town raised a $55 million Series A led by Andreessen Horowitz, with participation from Forerunner Ventures, Alt Capital, and Conviction.
Explore the recordings
Town’s supplied archive contains one recording: Jean-Denis Greze’s exploration of agent-to-agent coordination. Use it as a guide to information access, trust boundaries, and shared knowledge—not as evidence of Town’s current products or capabilities. The arguments and examples below describe what Greze said in the recording.
Start with the coordination problem
In Agents’ next frontier: agent-to-agent and network effects, Greze frames coordination as a search and context problem: getting the right information to the consequential model call. Follow his comparison of five approaches, from shared trust boundaries and custom tools to shared knowledge spaces, human-mediated requests, and broad search with selective owner approval. This is a useful entry point for comparing ways agents might work across private information silos.
Follow the shared-knowledge and safety thread
Return to the same recording for Greze’s near-term preference for automatically maintained wikis and databases. His personal wiki retaining the old agent name Apex after a rename to Ivy illustrates stale knowledge. Pair that example with his warnings about prompt injection, persistent incorrect memory, and unintended disclosure, and his proposed safeguards: approvals, logs, reversibility, audits, and human review for more sensitive information. These are recorded recommendations, not verified descriptions of Town’s implementation.
Explore the network-effects hypothesis
The closing discussion in Greze’s talk connects cross-company coordination to potential network effects. Treat this as a hypothesis to examine alongside the earlier privacy constraints: his finance example is unnamed, and he expresses uncertainty about trusting agents to make every privacy decision. The supplied catalog does not establish present-day adoption, outcomes, or product status.
1 talk
Newest first
1 speaker at AIE
Affiliations reflect their AIE appearances, not necessarily current employment.