Skip to main content

engineering

SimDrive vs Maestro: when each one wins

Maurice Carrier ·

Maestro is a good tool. It is one of the best things to happen to mobile test automation in the last five years. Anyone evaluating SimDrive against it should start there, because most of the comparison content on the internet starts from a strawman and we’re not going to do that here.

What we will do: lay out three honest dimensions of comparison, say what each tool actually wins at, and recommend running both for most teams.

Three dimensions

1. Authoring model

Maestro tests are YAML flows authored by humans. The YAML describes a deterministic sequence of taps, swipes, asserts, and text input, scoped to selectors (accessibility id, text matchers, point coordinates). The DSL is small, well-documented, and easy to read in a PR.

SimDrive flows are recorded by an agent or captured from a session. The agent walks the app — observing with a vision model, deciding next step, acting — and the result is a JSON journey you can replay deterministically. The authoring loop is “describe the goal in English, let the agent record it” rather than “open the YAML editor and write the flow yourself.”

Neither is universally better. Human-authored YAML is more legible in code review. Agent-recorded JSON is faster to create and tolerates UI drift better, because the recording captures intent rather than fragile selectors.

2. Runtime cost

Maestro runs deterministically with no AI in the loop. Every run is cheap, fast, and predictable. You can run a thousand of them on every PR and nobody notices the bill.

SimDrive runs in two modes. Record mode uses a vision-language model for each ios_observe call, so it costs real money — roughly a few cents per recording on typical apps. Replay mode doesn’t call the VLM at all; it dispatches the captured touches and verifies frames against the recorded baseline with SSIM. Replay is effectively free.

The dual-mode economic is the part worth understanding. You pay AI cost once when you record. Every subsequent replay against new builds in CI is $0. If your CI runs the same journey a hundred times in a quarter, the per-run AI cost amortizes to a rounding error.

3. Agent shape

This is where the tools genuinely diverge.

Maestro was designed before LLM agents were practical. The DSL is shaped for human authors and CI runners. You can wrap an agent around it — many teams have — but the seams are visible. The agent has to translate its intent into YAML, manage the runtime, parse output, and recover from selector failures.

SimDrive was designed for agents from the first commit. Tools are MCP-discoverable, return JSON-schema’d output, accept natural-language targets, and split observe-from-act so the agent can reason between them. There’s no DSL to translate into. The agent calls the tools directly.

If the workload is “Claude reproduces a bug from a Linear ticket,” SimDrive fits. If the workload is “QA team authors and maintains 200 deterministic smoke tests,” Maestro fits.

Where Maestro wins

  • Human-authored CI smoke suites. Teams that have already invested in YAML flows and want them to keep running cheaply for years should stay on Maestro for that workload.
  • Cross-platform parity. Maestro supports iOS, Android, Flutter, React Native, and web from one tool. SimDrive is iOS-only by design — that’s not changing in v1.
  • Deterministic flows in environments without a Mac. Maestro Cloud and CLI run anywhere. SimDrive needs macOS for the simulator runtime.
  • Test code review. Reading a Maestro YAML in a PR is faster than reading a JSON replay file.
  • No AI dependency. If your org policy forbids AI in the test loop, Maestro is the right tool — full stop.

Where SimDrive wins

  • Bug reproduction from a ticket. Paste the Linear ticket into Claude, get back a reproducing replay in 60 seconds. Maestro requires you to hand-write the flow first.
  • Vision-first observation. SimDrive’s ios_observe describes what’s on screen via a vision model. It tolerates UI drift that breaks accessibility-id selectors. Maestro is selector-based by default.
  • Agent-driven test authoring. Onboarding a new flow is “tell the agent what to do” rather than “open the YAML editor.”
  • Dual-mode AI economics. Record with AI once, replay without AI forever. Maestro doesn’t have an AI mode to compare; SimDrive’s replay mode is comparably cheap.
  • MCP-native distribution. Drop SimDrive into Claude Code, Cursor, or Continue with one JSON block. Maestro doesn’t ship as an MCP server.
  • Performance baselines and visual regression. SimDrive’s ios_perf_capture + ios_perf_compare and SSIM-gated replay are first-class. Maestro covers some of this surface but it’s not the primary shape of the tool.

Use both

This is a genuinely valid pattern and it’s what we’d recommend for most iOS teams above ten engineers.

  • Maestro runs your stable, human-authored smoke tests on every PR. They’ve been working for two years. They cover the critical paths. Don’t rewrite them.
  • SimDrive runs the agent loop: reproducing bugs from tickets, exploring unfamiliar parts of the app, recording new journeys, capturing perf traces, validating fixes. New flows that come out of SimDrive recordings can graduate into Maestro YAML later if the team prefers — the touch sequence is captured either way.

The integration is loose by design. Maestro and SimDrive don’t share state. They occupy adjacent slots in the test-automation pipeline. Most teams that adopt both find them complementary within the first sprint.

What ships in SimDrive 1.0.0b1

For evaluation purposes, this is the actual surface:

  • 32 MCP tools across observe, act, record, replay, journey, device, perf, doctor
  • Simulator support across iPhone 15/16/17 and iOS 17/18/26
  • Real-device support via WDA (iPhone, iPad)
  • Vision-first observation, no selector maintenance required
  • Replay runner with SSIM-based parity gates
  • Trial CLI: 14 days, no card
Terminal window
pip install simdrive
simdrive trial start --email you@example.com

Pricing is at simdrive.dev/pricing. Pro is $29/mo for a single dev. Team is $99/seat/mo with a 3-seat minimum. Enterprise starts at $50K/yr.

Maestro’s documentation is at maestro.mobile.dev. Their team has earned the recommendation.