Session
Who Owns the Click? When Humans and Agents Share a Running App
A browser inspector, a VS Code extension, a CLI and a coding agent all connect to the same running app. Reading state is easy. The trouble starts when two of them decide to tap the button.
Working on the shared inspector behind .NET MAUI DevFlow, we kept hitting problems that belonged in a distributed systems talk rather than a UI tooling one: a screenshot from 400ms ago paired with bounds from now, two clients mutating at once, a client that disconnected without telling anybody, a webview we had no reason to trust, and a recording that had to keep working when it changed hands mid-session.
I'll go through what we ended up with and what it cost us - mainly an app-scoped lease, so only one client can mutate at a time, and immutable snapshots, so nobody acts on a view of the app that stopped being true two frames ago. Both are less convenient than they sound. I'll also reproduce a couple of the races live, including my favorite category of bug, where the click lands on precisely the wrong element and every layer of the stack cheerfully reports success.
Very little of this turned out to be MAUI-specific. Mostly it was about deciding who is allowed to touch the app, and being willing to say when we didn't know.
For developers building agentic tools, IDE integrations, test infrastructure or remote inspectors. Based on the public experimental .NET MAUI DevFlow architecture. Preferred duration: 60 minutes including Q&A.
Dominik Titl
Software Engineering Intern on the .NET MAUI team at Microsoft
Prague, Czechia
Links
Please note that Sessionize is not responsible for the accuracy or validity of the data provided by speakers. If you suspect this profile to be fake or spam, please let us know.
Jump to top