Buoy vs Argent: one AI, four bugs, two tool sets

Argent is a free toolkit from Software Mansion. It lets an AI agent tap, swipe and type in iOS and Android apps. It can also take screenshots and record the screen. It can check speed and run code in the app. It does not need an SDK in your app. See the Argent README.

Buoy MCP connects your AI to the Buoy tools inside your app. You add Buoy packages to the app first. Then the AI can read and change the app's web calls, storage and state while it runs. Buoy MCP needs a Pro plan. See the MCP docs.

We used our own food ordering test app. It runs on iOS Simulators, and Buoy is in it. The AI was Claude Sonnet 5.5 in Claude Code. We used Argent 0.27.0 and the Buoy MCP build from October 3, 2026.

• We put four bugs in the app. We took out any notes in the code that gave a hint.
• Each run started from a clean copy of the app. The AI could read and edit the code. It could not use a shell.
• Each tool got each bug three times, with a 7 minute limit. That is 24 runs.
• After each run, a script opened the app again. It checked the fix like a tester would.

This is one app, and we wrote the tasks. We kept the log of every run. Your app and your tasks may give different results.

Both tools fixed all twelve of their runs. The times below are the middle of three runs.

Argent was 2 seconds faster on the offers bug. One Argent run on that bug took 5 minutes 52 seconds.

The biggest gaps came when the AI had to set up the bug before it could check the fix.

One customer sees NaN. Argent has no tool to act as another user. In all three runs, the AI found Buoy's own Impersonate tool in the app's menu. Then it tapped through it, 12 to 17 taps each time. With Buoy MCP, the AI did the same thing in one call.

Offers server down. Argent has no tool to make a web call fail. In one run, the AI broke the app's code on purpose. It added a line that throws an error. Then it checked the screen and took the line out. With Buoy MCP, the AI made the call fail with one mock rule.

Wrong cart total. With Argent, the AI tapped through the menu to build a cart, 8 to 11 taps each run. With Buoy MCP, it set the cart in the app's store in one call.

With both tools, the AI mostly read the code and fixed it first. Then it used the app to check the fix.

Before the bug test, we ran ten short tasks, three times each. Some were: read a value, fill the cart, tap a tab, and save a test. We made Buoy better using these same tasks. So give this test less weight.

Argent missed the task where you save a test and run it again. All three runs ran out of time at 150 seconds. On a checkout task, the AI turned on Argent's network log after it placed the order. So the log was empty. Buoy had already saved the failed call. So the AI could read the 402 error and why it failed.

Argent is built to drive the device. It sends real taps and swipes. It was faster on our tab bar task. Its tools reference also lists native views, speed checks with Instruments and Perfetto, screen recording and screenshot diffs. It works with TVs and Electron apps too.

Argent needs no SDK. So it can work with an app you can't change. It can also drive a real iPhone over USB. Its physical device page says JavaScript debugging, speed checks and screen recording only work on a simulator.

We checked this list against Argent's tools reference on October 4, 2026. Argent can run any code in the app. So an AI can sometimes build its own way around a missing tool. That is what it did on the offers bug.

• ✓Show web calls from the moment the app starts
• ✓Send a fake response for any API call
• ✓Make the server fail on purpose
• ✓Make the network slow or offline
• ✓Read and change app storage (AsyncStorage, MMKV, SecureStore)
• ✓Read and change Zustand, Redux, Jotai and React Query
• ✓See the app as a real customer (Impersonate)
• ✓Save the whole app state and put it back later (Time Machine)
• ✓Move the clock ahead
• ✓Fake the GPS location
• ✓Load a test setup with one call (Scenarios)
• ✓Read Sentry events before they leave the app
• ✓Check images and assets

Buoy's tools also work for people. Testers and support staff can use the same tools from the app's menu and from Buoy Desktop. See the docs overview for what each plan includes.

With Buoy MCP, the AI hit 33 tool errors in the bug test. With Argent, it hit 8. Most Buoy errors came from bad inputs. Some came from a missing device id when several apps were open. Some came right after the app reloaded. The AI got past each one, but they cost time. We are fixing them.

Buoy's real taps were slower on the tab bar task. Buoy also needs its packages in your app and a Pro plan for MCP.