# From Autopilot to Team in 2build: Touch Only Brief and Handover

> Micro-managing coding agents prompt by prompt wastes engineering time. The Team model in 2build needs you only at Brief and Handover; in between, agents write code, review, and QA inside isolated Orca worktrees.

- URL: https://2found.dev/blog/from-autopilot-to-team-in-2build-en/
- Published: 2026-11-11T09:06:00.144Z
- Updated: 2026-11-11T09:06:00.144Z
- Language: en

---

## The problem: working with coding agents can feel like micro-management

As software projects grow, feeding prompts one by one, waiting on single-turn replies, and copying diffs back and forth quickly turns you into a micro-manager. You end up spending more time supervising the agent than solving product problems.

[2build](https://2found.dev/tools/2build/) started with **Autopilot** — taking a single ticket from requirement to a local verified commit. But when building a multi-phase feature, you cannot babysit every terminal command. That is why the **Team** model was created (previously Foreman): you only need to show up at two critical moments — **Brief** and **Handover**. Everything in between runs autonomously across coordinated agents.

---

## Boundaries: Autopilot vs. Team

Both tools serve product construction in 2build, but their scope and operating rhythm differ substantially:

| Characteristic | Autopilot | Team |
|---|---|---|
| **Scope** | Single ticket (feature or bug fix) | Entire project / multi-phase release |
| **Speed & Cost** | Fast, lightweight, lowest cost per change | Multi-task orchestration, longer horizon |
| **Stop point** | QA-verified commit on a local branch | Complete Handover ready for release |
| **Human involvement** | Monitored step-by-step | Only at Brief and Handover |

If you need to fix a single bug or add an API endpoint, Autopilot is the cleanest tool. When you need to build a complete flow from database migrations to UI screens, that is the cue to switch to Team.

---

## The only two human touchpoints

Under the Team model, you do not review individual git diffs between intermediate steps. Human involvement is intentionally restricted to two checkpoints:

### 1. Brief: Finalize the specification and prototype
Here, you define the project scope with the agent:
- Confirm requirements specifications (spec) and user journeys.
- Agree on acceptance criteria.
- Review and approve the design prototype before code is written.

Once you sign off on the Brief, the agent team enters autonomous execution.

### 2. Handover: Verification and decision trail
Coding, cross-reviews, and test suites run in background Orca worktrees, keeping your working branch clean. When execution completes, the agent presents results at Handover:
- Test reports and feature validation verified against the acceptance criteria agreed in the Brief.
- A **decision trail**: clear rationale explaining why specific architectural and implementation decisions were made.
- A final release decision ready for your confirmation.

---

## Five strict checkpoints requiring human intervention

To maintain reliability without nagging you, 2build Team enforces **5 strict checkpoints**. The agent team pauses and requests human review only when encountering one of these blockers:

1. **Budget / Cost:** Expenses exceed the configured token or compute threshold for the task.
2. **Credentials & Secrets:** Access requires an API key, certificate, or credential that the agent is not permitted to provision automatically.
3. **Release fails twice:** Automated testing or deployment fails two consecutive times.
4. **Git base branch conflict:** Merge or rebase conflicts with the base branch that automatic algorithms cannot resolve cleanly.
5. **Sensitive action approval:** Any operation modifying sensitive data structures or access control policies.

Outside of these five scenarios, syntax errors, failed test assertions, and refactoring are resolved autonomously by agents in isolated worktrees.

---

## Token optimization via automated Model Routing

Running a multi-agent team across a complex project can consume large token volumes if expensive models handle every subtask. 2build Team routes models dynamically based on task complexity:

- **Lite / Flash:** File discovery, search, context summarization, linting, and mock data generation.
- **Pro:** Primary code implementation, unit test authoring, cross-review, and QA validation.
- **Max:** Difficult domain logic, architecture design, authentication configurations, payment transactions, and critical security decisions.

This tiered approach directs high-tier reasoning where needed while keeping token consumption predictable.

---

## Where to start

There is no complex configuration syntax to learn. In 2build, start with a clear Brief: define the problem, specify acceptance criteria, and let Team handle implementation through to Handover.

---

Source documentation: [2build Team](https://2found.dev/docs/2build/team/)
