Working as Team Lead

In Exponential, you are the team lead (팀장). Agents are your teammates (팀원). This guide covers the patterns and practices for managing them effectively.

The team lead mindset

Your job is not to do the work — it's to:

1. Define what needs to happen — clear task descriptions with context
2. Choose who does it — match agents to tasks based on expertise
3. Review the approach — approve or redirect plans before execution
4. Evaluate the output — review deliveries and give feedback

This is management, not delegation. You're actively involved at every decision point.

Running multiple agents

Exponential supports running multiple agents in parallel. Each agent works on its own task, and you supervise all of them from the board.

Setting up parallel work

1. Create multiple tasks — break your project into independent units of work.
2. Assign different agents — use specialized agents for specialized tasks (e.g., a frontend agent for UI work, a backend agent for API work).
3. Stagger assignments — you don't have to assign everything at once. Start with the highest-priority tasks and add more as you process approvals.

Managing the board

Your kanban board shows all tasks by status:

The key skill is scanning. Check the board regularly:

• Are any tasks stuck in Planning? → Review the plan
• Are any agents asking questions in Working? → Answer them
• Are any deliveries waiting in Review? → Evaluate the output

Reviewing plans effectively

The plan review is your most important touchpoint. A few minutes here saves hours of rework.

What to look for

• Does the approach make sense? — Is this how you'd solve the problem?
• Are the steps in the right order? — Dependencies should be handled first.
• Is anything missing? — Tests? Error handling? Edge cases?
• Is it over-engineered? — Simpler is usually better.

How to give feedback

Instead of rejecting a plan outright, use comments:

"The overall approach looks good, but I'd skip step 3 — we don't need that abstraction. Also, add error handling for the API call in step 5."

This tells the agent exactly what to adjust. It's faster than a full re-plan.

When to redirect

Redirect (reject + provide new direction) when:

• The agent fundamentally misunderstood the task
• The approach won't work for reasons the agent can't see (e.g., upcoming changes, business constraints)
• You want a completely different strategy

Reviewing deliveries

When an agent submits a delivery, you see:

• Summary — what was accomplished
• Artifacts — files created or modified
• Activity timeline — the full history of what the agent did

Review checklist

1. Read the summary — does it match what you asked for?
2. Check the artifacts — open the files and review the actual changes.
3. Look at the timeline — did the agent encounter issues? Did it make decisions you disagree with?
4. Test if applicable — for code tasks, run the tests or try the feature.

Giving feedback

If the delivery needs changes:

1. Post a comment describing what needs to change.
2. Be specific — "The error handling in handleSubmit should show a toast notification, not a console.log" is better than "Fix the error handling."
3. The agent will re-enter Working to address your feedback.

Communication patterns

When to comment

• Answering agent questions — agents sometimes ask for clarification. Answer promptly to unblock them.
• Adding context mid-task — if you realize the agent needs information you forgot to include, add it as a comment.
• Redirecting — if you see the agent going in the wrong direction during Working, interrupt with a comment.

When NOT to comment

• Routine progress — let agents work. You don't need to acknowledge every status update.
• Obvious corrections — if the agent will clearly catch it in the next step, wait.

Patterns for complex projects

Sequential tasks

For work that has dependencies:

1. Create Task A (the dependency).
2. Wait for Task A to be done.
3. Create Task B, referencing Task A's output.

Parallel + merge

For work that can be done independently and then combined:

1. Create Task A (frontend) and Task B (backend) simultaneously.
2. Assign different agents.
3. When both are done, create Task C to integrate and test.

Iterative refinement

For work that needs multiple passes:

1. Create Task 1 with the initial scope.
2. Review the delivery.
3. Create Task 2 with refinements based on what you learned.
4. Repeat until satisfied.

Tips

1. Match agents to tasks — a frontend agent on a backend task will produce mediocre results. Use the right specialist.
2. Keep tasks focused — one clear objective per task. "Build the entire auth system" is too big. "Create the login form component" is better.
3. Front-load context — the more context in the task description, the better the plan. Include links, examples, and screenshots.
4. Use memory — if you find yourself repeating the same context in every task, add it to space memory instead.
5. Trust the process — the plan-approval gate is your safety net. Read plans carefully, and you'll rarely be surprised by the output.