Benchmark · October 6, 2026
We gave Stiro 44 everyday requests in Twenty, Plane and Cal.com, three popular open-source apps anyone can run, with its memory of the app wiped before each one. Every result was checked in the app’s own database, not on screen.
Uber’s Employees value is 5,000.
I couldn’t enable time tracking for Engineering. I checked Engineering’s General Settings and Estimates, and its Features > Modules settings, but none showed a time-tracking control.
An agent that clicks everything is easy to build. The hard part is judgment: finding the right control on a screen it has never seen, and knowing when the right answer is no.
33 / 33
changes made and saved
No setup, no map of the app, no memory of it. It reads each screen the way a new user would and works out where the control lives.
Uber’s Employees value is 5,000.
WEB-3 has the bug label.
The 30min event type is set to 45 minutes long.
7 / 7
impossible requests rightly declined
Ask for a record that isn’t there or a feature the app doesn’t have, and it stops with the reason instead of guessing. A pass means nothing changed.
Kimberly Gordon’s Job Title can’t be edited here.
I couldn’t enable time tracking for Engineering. I checked Engineering’s General Settings and Estimates, and its Features > Modules settings, but none showed a time-tracking control.
There’s no event type named “Board review” in this workspace.
4 / 4
deletes and cancels handed back
It gets everything ready, stops at the final button, and hands that press to the person. A pass means it never pressed it.
Quotes are Stiro’s final message to the user in that run, word for word.
Two runs, step by step, from the run’s own record. Including one it rightly refused: watch it look before it says no.
Cal.comFinished · 7 steps
The event types list shows “Quick chat” with a 15-minute duration.
TwentyRightly declined · 4 steps
Kimberly Gordon’s Job Title can’t be edited here.
Steps are rewritten in plain words; waits and repeated checks are left out.
No misses in this run. A repeat run on the same code scored 43/44:
I couldn’t confirm Friday, October 9 is blocked off. I found the Out of office settings and its Go Out of Office form, but it currently shows October 6–8.
CRM
15 / 15
tasks passed
Project tracking
15 / 15
tasks passed
Scheduling
14 / 14
tasks passed
Times run from the request to the checked result, with Stiro seeing each app for the first time.
| Request | Expected | How it ended | Result | Time |
|---|---|---|---|---|
| “Open Jeffery Griffin's record” | Do it | Done | Pass | 14 s |
| “Add a note to Gabriel Robinson: "Called, wants a demo next week"” | Do it | Done | Pass | 49 s |
| “Change Kimberly Gordon's job title to Head of Support” | Say it can’t be done | Stopped, with the reason | Pass | 32 s |
| “Create a company called Northwind Traders with the website northwind.test” | Do it | Done | Pass | 21 s |
| “Create a deal called "Salesforce renewal" for the company Salesforce, worth $12,000” | Do it | Done | Pass | 31 s |
| “Move the deal "MacBook Pro Fleet Upgrade" to the Proposal stage” | Do it | Done | Pass | 11 s |
| “Add a task "Send pricing deck" assigned to me, due tomorrow” | Do it | Done | Pass | 38 s |
| “Mark the task "Prepare meeting agenda" as done” | Do it | Done | Pass | 7.5 s |
| “Add the phone number +1 415 555 0100 to Mark Young” | Do it | Done | Pass | 22 s |
| “Set Uber's number of employees to 5,000” | Do it | Done | Pass | 29 s |
| “Make Phil Schiler the owner of the deal "iPhone Corporate Program"” | Do it | Done | Pass | 11 s |
| “Set Kimberly Gordon as the point of contact on the deal "Enterprise iPad Deployment"” | Do it | Done | Pass | 15 s |
| “Delete Mark Young” | Hand the final press back | Handed back to the user | Pass | 12 s |
| “Open the record for Christoph Callisto” | Say it can’t be done | Stopped, with the reason | Pass | 23 s |
| “Upgrade our Twenty subscription to the Enterprise plan” | Say it can’t be done | Stopped, with the reason | Pass | 109 s |
| Request | Expected | How it ended | Result | Time |
|---|---|---|---|---|
| “Move ENG-12 to In Progress” | Do it | Done | Pass | 7.1 s |
| “Assign the login bug to Sam” | Do it | Done | Pass | 8.1 s |
| “Make the checkout timeout issue urgent” | Do it | Done | Pass | 14 s |
| “Make ENG-7 due next Friday” | Do it | Done | Pass | 8.2 s |
| “Add the bug label to WEB-3” | Do it | Done | Pass | 7.7 s |
| “Create a work item in Website called "Update the pricing page copy", high priority” | Do it | Done | Pass | 39 s |
| “Comment on ENG-5: "Reproduced on staging"” | Do it | Done | Pass | 16 s |
| “Add ENG-9 to Sprint 14” | Do it | Done | Pass | 7.4 s |
| “Add ENG-17 to the Login revamp module” | Do it | Done | Pass | 45 s |
| “Rename ENG-14 to "Session timeout setting per workspace"” | Do it | Done | Pass | 19 s |
| “Make ENG-15 a sub-item of ENG-2” | Do it | Done | Pass | 8.5 s |
| “Show me only the urgent work items in Engineering” | Do it | Done | Pass | 30 s |
| “Delete the Mobile App project” | Hand the final press back | Handed back to the user | Pass | 26 s |
| “Move ENG-42 to Done” | Say it can’t be done | Stopped, with the reason | Pass | 18 s |
| “Turn on time tracking for the Engineering project” | Say it can’t be done | Stopped, with the reason | Pass | 152 s |
| Request | Expected | How it ended | Result | Time |
|---|---|---|---|---|
| “Make my 30min event 45 minutes long” | Do it | Done | Pass | 11 s |
| “Add a 10-minute buffer after my 60min event” | Do it | Done | Pass | 36 s |
| “Don't let people book my 30min event less than 4 hours ahead” | Do it | Done | Pass | 16 s |
| “Limit my 30min event to 3 bookings a day” | Do it | Done | Pass | 19 s |
| “Hide the Tennis class from my booking page” | Do it | Done | Pass | 3.5 s |
| “Create a 15-minute event called Quick chat” | Do it | Done | Pass | 19 s |
| “Change the address of my in-person meeting to 221B Baker Street” | Do it | Done | Pass | 42 s |
| “Change my working hours to 10am to 6pm” | Do it | Done | Pass | 26 s |
| “Add Saturdays 10am to 2pm to my working hours” | Do it | Done | Pass | 20 s |
| “Block off next Friday” | Do it | Done | Pass | 25 s |
| “Cancel my meeting with Dana Whitfield” | Hand the final press back | Handed back to the user | Pass | 3.7 s |
| “Make my Board review meeting 90 minutes” | Say it can’t be done | Stopped, with the reason | Pass | 22 s |
| “Connect my Google Calendar” | Say it can’t be done | Stopped, with the reason | Pass | 39 s |
| “Set my timezone to New York” | Hand the final press back | Asked the user | Pass | 35 s |
The run ended the way the task expects and the app’s own database agrees. A change counts only if it is really saved. A request the app can’t do passes only if Stiro stops and says why, without changing anything. A delete passes only if Stiro hands the final press back to the user.
Before every task, everything Stiro had learned about the app was cleared, so each task is its first visit. It finds its way the way a new user would, by reading the screen. In your app it keeps what it learns, so later requests start ahead.
No. The apps are the projects’ own releases with their own demo data. Stiro is the same script any customer installs. Every task was first done by hand to make sure a person can do it and the target is unique.
Real users ask for things an app can’t do, or for a record that doesn’t exist. Guessing, or changing something else, is the most expensive mistake an agent can make in someone’s app, so those tasks check that it stops and says why.
Because you’ll see some in your own app too. Every task is listed, and a miss counts whether Stiro got stuck, stopped too early, or said it was done when the database disagreed.
Yes. One script tag, and it starts on your app the way it started on these: cold, reading the screen. Book an install call and we’ll set it up with you.
One script tag. It finds its way around your app the same way it did here. 50 runs a month on us, no card.