Overview

Exponential is built for teams where humans and AI agents work side by side. The core idea: humans define what to build, agents handle the how.

This isn't about replacing engineering judgment — it's about amplifying it. You bring the product thinking, the architectural decisions, the tradeoff calls. Agents bring tireless execution against precise specs.

The workflow

A typical human-agent workflow in Exponential follows four steps:

1. Describe — capture a rough idea as an epic. It can be a sentence or a paragraph. You don't need to have all the answers yet.
2. Decompose — work with your agent to break the epic into estimated, dependency-mapped stories with clear specs. The agent drafts, you review and refine. See Writing Specs for what makes a spec the agent can execute well.
3. Drive — run xpo drive and the agent picks up the next story, implements it, runs your tests, and merges the result. See Agentic Drive for how the loop works.
4. Review — check the walkthrough, review the diff, leave feedback if needed. Ship when you're satisfied.

You stay in control of the what and the when. The agent handles the how and the how fast.

Three ways agents connect

Agents interact with your backlog through the same tools your team uses:

MCP Server — structured tool calls for listing, creating, updating, and linking issues. This is how agents in Claude Code, Cursor, and other MCP-compatible editors read and write your backlog natively. See MCP Server.

CLI — the same xpo commands you use. Agents can shell out to xpo list, xpo start, xpo done. Works everywhere, no setup required.

Autonomous drive — xpo drive orchestrates the full loop: pick a story, plan, implement, test, review, merge. This is the hands-off mode where the agent runs independently. See Agentic Drive.

Setting up

1. Install Exponential and run xpo init in your project
2. Configure the MCP server so your agent can access the backlog
3. Set up agent identity so agent actions are attributed correctly
4. Start a conversation — ask your agent to read the backlog and help you plan

What to read next

• Writing Specs — how to write specs that agents can execute reliably
• Agentic Drive — the autonomous execution loop
• AI Agents — configure CLAUDE.md, agent identity, and drive settings