Running Multiple Agents

Run parallel AI agents across lanes, monitor them from the Workspace Graph, and handle conflicts between concurrent work.

ADE is built for parallel agent execution. This guide covers creating lanes for parallel work, assigning agents to each, monitoring everything from the Workspace Graph, and dealing with conflicts when agents touch overlapping files.

---

1. Plan Your Parallel Work

Before launching agents, decide how to split the work:

• Independent features — each agent works on a separate feature with minimal file overlap. Low conflict risk.
• Related changes — agents work on related features that touch some of the same files. Medium conflict risk — ADE’s conflict prediction helps here.
• Stacked work — agents work on a dependency chain (see Stacked PR Workflows). Coordination is automatic.

Start with 2-3 parallel agents on independent features. Scale up as you get comfortable with ADE’s conflict detection and resolution tools.

---

2. Create Lanes

Create a lane for each parallel workstream.

1. Open the Lanes panel

   Navigate to Lanes in the sidebar.

2. Create lanes

   Click New Lane for each workstream. Give each lane a descriptive branch name (e.g., feature/payment-api, feature/user-dashboard, fix/auth-redirect).

3. Set base branches

   All lanes targeting the same integration branch (usually main) can run independently. For stacked work, set parent lanes as the base.

---

3. Assign Agents

Launch an agent in each lane. There are several ways to do this:

From the Lane panel

Select a lane, then click Start Agent. Choose the agent type (Claude CLI, Codex CLI, or Unified) and provide a prompt describing the task.

From the CTO

Ask the CTO to delegate work. It will create lanes and spawn workers automatically based on your instructions.

Via a Mission

Create a mission with a plan that has parallel steps. The mission coordinator spawns workers into separate lanes and manages them.

Via Automations

Set up automation rules that spawn agents in response to triggers. Each rule can target a specific lane or create a new one.

---

4. Monitor from the Workspace Graph

The Workspace Graph is the best view for monitoring multiple agents simultaneously.

Navigate to Workspace Graph in the sidebar. You will see:

• Lane nodes — each lane appears as a node on the graph, color-coded by status
• Agent activity indicators — lanes with running agents show a live activity pulse
• Branch relationships — edges connect lanes to their base branches
• Conflict warnings — overlapping file indicators appear between lane pairs with predicted conflicts

Coming soon: screenshot walkthrough of the Workspace Graph with multiple active agents.

---

5. Track Agent Output

While agents run, you can inspect their work from several views:

View	What it shows
Lane Terminals	Live terminal output for each agent session
Agent Chat	The agent’s reasoning and tool calls in chat format
History	Timeline of all events across all lanes
Missions > Run	Step-by-step progress if agents were launched via a mission

Switch between lanes in the Lanes panel to focus on a specific agent, or use the Workspace Graph for the high-level view.

---

6. Handle Conflicts

When multiple agents modify overlapping files, ADE detects the conflict risk:

1. Conflict prediction

   ADE periodically runs merge simulations between active lanes. Predicted conflicts appear in the Conflicts view and as warning badges on the Workspace Graph.

2. Review the overlap

   Open the Conflicts view to see which files overlap and which lanes are involved. ADE shows the specific file regions at risk.

3. Resolve or defer

   You have several options:

   • Pause one agent — let the other finish first, then rebase
   • AI resolution — let ADE’s conflict resolver agent propose a merge
   • Manual resolution — resolve the conflict yourself in a terminal session
   • Defer — continue working and resolve later (the conflict may resolve itself if agents diverge)

Conflict prediction runs automatically when the Job Engine is active. In development mode, you may need to enable it with ADE_ENABLE_CONFLICT_PREDICTION=1.

---

7. Collect Results

As agents complete their work:

• Each lane has its changes committed and ready for PR
• Mission artifacts (if applicable) are collected in the Artifacts tab
• The CTO can review results across all lanes and post summaries

Open PRs from each lane through the lane context menu or the PR panel.

---

Scaling Guidelines

Agents	Recommendation
1-3	Fine for any machine. Monitor via Lanes panel.
4-8	Use the Workspace Graph for overview. Set budget caps per agent.
8+	Use missions with parallelism caps. Monitor total token spend in Settings > Usage.

---

Next Steps

Stacked PR Workflows

Organize parallel lanes into stacked PR chains for dependent work.

CTO + Linear

Let the CTO automatically delegate work to parallel agents from Linear issues.