We wanted to find out: how does Superconductor actually stack up against the competition? So we took six cloud agent platforms — Claude Code Web, Codex Web, Cursor Agents, Devin, Jules, and Superconductor — and ran each through the exact same task, start to finish.
An important note: most platforms lock you into a specific coding agent (Claude Code Web only runs Claude Code, Devin only runs Devin, etc.), while Superconductor lets you choose from any agent, including running different agents in parallel on the same ticket. For this test, we ran Claude Code (Opus 4.6) on Superconductor.
The Task
We tested each platform on a personal project built by one of our team members: an animal finder app — a Python Flask app where you type in any animal name and it shows you zoos where you can find that animal, places to see it in the wild, and more.
For each platform, we created an account, went through onboarding, set up the dev environment (if required), and ran agents on two simple tickets:
- Update the habitat maps to use a different map library so that all labels display in English
- Reorder the "zoos" and "tours" tabs
For both tickets, we also asked agents in the ticket prompt to run tests and verify the change in the browser if possible.
Ticket 1 prompt:
Ticket 2 prompt:
The Results
Here's how every platform performed. All times are standalone durations (mm:ss):
| Platform | Agent | Onboarding | Ticket 1 | Ticket 2 | Total |
|---|---|---|---|---|---|
| Claude Code Web | Opus 4.6 | 1:36 | 13:27 | 3:37 | 18:40 |
| Codex Web | Codex 4.5 | 2:20 | 12:12 | 3:30 | 18:02 |
| Cursor Agents | Opus 4.6 | 10:15 | 13:10 | 5:48 | 29:13 |
| Devin | Devin | 1:03 | 17:29 | 4:26 | 22:58 |
| Jules | Gemini 3.1 Pro | 1:53 | N/A* | 8:00 | 52:46** |
| Superconductor | Opus 4.6 | 4:07 | 8:31 | 1:49 | 14:27 |
*Could not complete due to breaking bug with PR creation
**Only completed second ticket - extra time spent trying to get Ticket 1 PR creation to work
Metric guide
- Onboarding: Time from account creation to first ticket started. A purely platform-dependent metric.
- Ticket 1: End-to-end time from ticket creation to tested and PR merged, including agent work and human review.
- Ticket 2: End-to-end time from ticket creation to completion.
- Total: Sum of the three time columns above.
Superconductor completed both tickets in 14:27 — the fastest total time of any platform tested. Ticket 2 took just 1:49, nearly half the time of the next-fastest competitor.
Here's the breakdown. For detailed comparisons of each platform against Superconductor, check out our comparison docs.
- Claude Code Web — Smooth and easy to onboard. Fast dev env setup. Agent got stuck restarting Playwright for browser testing for ~10 minutes but eventually succeeded on its own. Adding secrets was not intuitive.
- Codex Web — Fast agent execution, easy to read work logs. However, the times shown in the table are from the 8th successful attempt — the previous seven failed with "cannot clone repository" errors during various stages of implementing the first ticket. Browser testing never worked, and adding secrets was not intuitive.
- Cursor Agents — Slow overall. Dev env setup happens in a separate interface with a separate agent and took ~10 minutes. Agent execution was also slower than most other tools. In-app preview was nice, though a bit hard to find.
- Devin — Best UX of the competitors. Some unique features: no separate dev env setup step, secrets requested in-chat of first ticket, and you can ask questions mid-run without stopping the agent. Guided code review and in-app previews built in. Clean run, zero issues. Price point was high compared to other tools.
- Jules — Most problematic by far. Extremely slow, no work log during planning mode, and a breaking bug prevented completing ticket 1 entirely. Could not create a PR for Ticket 1. Some unique ideas (task scheduling, notifications) but unusable due to bugs.
- Superconductor — Fastest overall. Ticket 2 done in 1:49. Dev env setup took slightly longer (4:07) due to spinning up a full cloud environment with live preview — but that investment paid off: human-testing the change in a live preview while the agent ran its own tests saved significant time on review. Both PRs merged cleanly.
What This Means
No platform was perfect, but the differences were significant. Some were fast but unreliable (Codex Web). Some had great UX but were slower and more expensive (Devin). One was essentially broken, at least on our attempt (Jules). Superconductor delivered the fastest end-to-end time with the smoothest path from ticket to merged PR — and the live preview made it easy to verify changes with confidence before merging.
Because Superconductor is agent-agnostic, you're never locked in. As better models and agents ship, you can switch immediately — no migration, no new platform to learn.
If you're evaluating cloud agent platforms for your team, the best test is always your own codebase with your own tasks. Sign up for Superconductor and see the difference for yourself.
We're excited to keep testing more platforms as the space evolves. If there's a cloud agent platform you'd like us to compare against, let us know!