Skip to main content
pre.dev combines an AI architecture agent with autonomous build agents running in secure sandboxes. Here’s what happens under the hood when you create a project. 
> Build a marketplace for freelance designers...

  [+] [Deep Spec] [Pro]          [Start building →]

​
Phase 1: Architecture

When you describe your idea, the architecture agent takes over — researching, designing, and producing a complete roadmap: 
# pre.dev — architecture-agent

[done] User Flows
[done] Tech Stack
[done] User Stories
[ 63%] System Architecture ████████........
[    ] Subtasks
[    ] Coding Agent
1

Research

Understands the domain, identifies edge cases, considers scalability
2

Stack Selection

Chooses framework, database, auth, and hosting based on your requirements
3

System Design

Produces component hierarchy, database schema, API contracts, and data flows
4

Specification

Outputs structured milestones, user stories, and subtasks with complexity estimates
The output is a living document that serves as the blueprint for the entire build.

​
Phase 2: Building

Once you approve the spec (or hit Build on Autopilot), the build agents execute:
  1. Task isolation — Each subtask is assigned to a focused agent with only the relevant context
  2. Code generation — The agent writes implementation code, tests, and configuration
  3. Acceptance verification — Automated checks validate the work:
    • TypeScript/type compilation passes
    • Linting rules pass
    • Unit tests succeed
    • Visual browser verification (for UI tasks)
  4. PR creation — A feature branch is created and a pull request opened
Agents cannot mark a task as complete until acceptance criteria pass. If verification fails, the agent iterates until it’s resolved.
Agents self-correct. If type checks fail or tests break, the agent automatically fixes the issues before opening a PR. You only see verified, working code.

​
Phase 3: Review & Ship

You stay in control:
  • Review PRs — Every task produces a reviewable pull request
  • Merge on your terms — The AI never pushes to main
  • Track progress — The roadmap view shows real-time status across all milestones

​
The Sandbox Environment

Build agents run in isolated, secure containers:
  • Sandboxed execution environment — zero risk to production
  • Full development toolchain available (Node.js, Python, system tools)
  • Access to your Knowledge Base API keys for real integrations
  • Long-running — agents can work for hours on complex tasks without timing out

​
Parallel Execution

When you click Build All, pre.dev doesn’t build sequentially. Tasks within the same milestone that have no dependencies run in parallel, maximizing throughput. 
Milestone 1
├── Task 1.1 ──→ Agent A (building...)
├── Task 1.2 ──→ Agent B (building...)  ← parallel
└── Task 1.3 ──→ (waiting on 1.1)       ← sequential dependency

Milestone 2
└── (waiting on Milestone 1 completion)

​
Context & Knowledge

Agents have access to:
SourceDescription
SpecificationThe full architectural blueprint for the project
CodebaseAll code written so far in the repo
Knowledge BaseYour stored API keys and documentation
External DocsScraped documentation from URLs you provide
This means agents don’t hallucinate integration details — they reference real documentation and use real API keys.

​
The Result

Every project follows the same pipeline: Idea → Architecture Agent → Specification → Build Agents → Verified Code → PRs → You Merge → Deployed App
No prompt engineering. No babysitting. No “looks good but is broken.” Just verified, reviewable code shipping autonomously.