How I Completed a 2-Week Sprint in 48 Hours

A concrete walkthrough — not a hype piece — of what parallel agents actually delivered.

The Setup

I had a 2-week sprint with 8 tickets: 3 new features, 2 bug fixes, 2 refactoring tasks, and a test coverage requirement (bring coverage from 45% to 70%). Normally this is a full sprint's worth of work for one developer. I ran it over a weekend with an amux agent fleet.

This isn't a story about AI magic. It's a story about task decomposition, good specifications, and letting agents run overnight while I slept.

Friday Evening: Setup (2 hours)

I spent 2 hours not coding, but writing. Task descriptions for each of the 8 tickets, with explicit file paths, interface requirements, and success criteria. I registered 6 amux sessions, assigned tasks via the board, and launched them at 10pm. Then I went to sleep.

# The board at 10pm Friday
FEAT-01: User profile API endpoints [agent: api-1]
FEAT-02: Email notification system [agent: api-2]
FEAT-03: Dashboard analytics component [agent: frontend-1]
BUG-01: Fix pagination off-by-one [agent: api-1] (queued)
BUG-02: Fix mobile menu z-index [agent: frontend-1] (queued)
REFAC-01: Extract auth middleware [agent: api-2] (queued)
REFAC-02: Consolidate CSS variables [agent: frontend-2]
TEST: Coverage blitz — models + controllers [agents: test-1, test-2]

Saturday Morning: Review (3 hours)

I woke up to 5 completed tasks and 3 still in progress. I reviewed the completed work — merged 4 items cleanly, sent one back with corrections (the notification system had used the wrong email template structure). The test agents had brought coverage to 68%.

By noon Saturday all 8 tickets were done. I spent the afternoon doing integration testing and review — finding 3 issues the agents had missed (one security, two UX problems) and fixing them myself. These were genuinely complex issues that required judgment. The agents handled everything they could handle.

What Actually Took Time

The 48 hours included: 2 hours of task specification, 3 hours of code review, 2 hours of integration testing, and 1 hour of fixing what agents missed. Total hands-on time: 8 hours. An agent fleet ran for 36 hours of that window handling the rest.

What the Agents Couldn't Do

• Catch a subtle race condition in the notification queue (human caught it in review)
• Decide that the analytics component needed a different data model than specified
• Notice that the email notification design didn't match the company's style guide

These are judgment calls. They require context, taste, and product knowledge. The agents delivered implementations — the developer delivered decisions.

git clone https://github.com/mixpeek/amux && cd amux && ./install.sh
amux register myproject --dir ~/Dev/myproject --yolo
amux start myproject
amux serve  # → https://localhost:8822