Superintelligence at Work: How Vercel Built v in Slack
Key Takeaway
Embedding enterprise agents directly into Slack workflows alongside team members creates shared, compounding context and enables secure, multi-agent collaboration across internal business systems.
Highlights
Vercel built its internal company agent, V, using its open-source agent framework, EVE, and the Vercel AI gateway.
Slack integrates multiplayer agent collaboration by allowing coding agents to spin up dedicated channels, review code diffs, and auto-archive threads for searchable historical context.
Vercel's architecture avoids static tokens by using on-demand, short-lived tokens minted with fallback scopes based on user permissions and channel context.
Runtime model selection through a single AI gateway decreases token spend and routes requests to specialized models based on request classification.
Domain experts build specialized sub-agents like DZero for data queries and Penny for billing insights using natural language skills without requiring coding experience.
Timeline
Moving Agents Into Slack for Multiplayer Collaboration
Initial attempts at Vercel involved building separate agents for data, docs, and sales, but they lived in a disjointed web app that forced single-player interactions. Shifting these agents into Slack enables multiplayer agentic collaboration where work already happens. This architecture connects people, agents, and platforms without requiring employees to bridge separate silos manually.
- Proliferating standalone agents across separate internal web apps creates single-player silos.
- Placing agents directly into Slack channels meets employees within their existing workflow.
- Unifying people, agents, and platforms in a single interface eliminates productivity drains caused by disjointed systems.
Practical Workflow of an Agent in Slack
A practical example demonstrates a coding agent updating a web page by pulling in a marketing team member for missing links and a technical lead for bug fixes. Team members review changes through HTML artifacts and code diffs directly inside the Slack interface. Once completed, the channel auto-archives, allowing future employees to search past decisions and compound organizational context over time.
- Coding agents can spin up dedicated code channels while retaining context from previous threads.
- Team members collaborate directly around code diffs, HTML previews, and pull requests inside Slack.
- Auto-archived channels preserve decision histories that remain searchable years later.
Architecture and Security of Vercel's Internal Agent, V
Vercel developed V as a unified entry point to internal intelligence using the open-source agent framework EVE. Independent teams build and deploy specialized agents autonomously while contributing to the same codebase. The architecture relies on Vercel's AI gateway for runtime model classification, reducing token spend while maintaining a shared observability and governance layer. Security measures omit static tokens in favor of channel-scoped permissions and auditable interaction turns.
- V serves as the centralized entry point to all Vercel intelligence and sub-agents inside Slack.
- The open-source framework EVE enables independent teams to deploy specialized agents autonomously.
- Security is enforced through channel-scoped responses, full observability, and the elimination of static tokens.
Handling Complex Multi-Agent Requests and Key Learnings
A typical request concerning customer usage and billing routes from V to specialized sub-agents, including the data specialist DZero and the finance specialist Penny. Vercel mints short-lived, on-demand tokens to query external databases like Snowflake securely. Key lessons from development include recognizing that system connections do not equal business context, implementing strict trace policies for telemetry, and evaluating the entire multi-agent journey rather than isolated sub-agent outputs.
- Slack webhooks trigger requests routed through V to specialized sub-agents like DZero and Penny.
- Connections to raw databases do not inherently provide business context without domain-specific skills.
- Comprehensive telemetry and rigorous end-to-end evaluations are required for complex multi-agent interactions.
Developer Tools and Rapid Agent Creation
Slack offers a robust suite of developer tools, including the Slack Agent Kit, SDK, BlockKit Builder, and CLI to support custom integrations. Granular role-based access controls and MCP servers ensure secure operations within the workspace. Additionally, the Add Agent feature enables developers and non-coders alike to construct secure internal company agents within minutes using simple prompts.
- Slack provides developer kits, SDKs, BlockKit Builder, and CLI tools for custom agent integrations.
- The MCP server and granular role-based access controls secure workspace deployments.
- The Add Agent feature allows users to build internal company agents in minutes using a single prompt and natural language instructions.
Description
Vercel.com