Practitioner's Corner
Practitioner's Corner
Ananya Gubbi Mohanbabu and the Control You Can Get Back To
Ananya Gubbi Mohanbabu, a UC Berkeley PhD student, builds tools for blind and low-vision people who use screen readers, software that speaks the screen aloud. One of her systems hides page elements that seem irrelevant to the user's stated goal. In a 12-person study, its screen reader users finished tasks in about half the time they took without it. It can also hide the button someone needed. Her newest work tests AI agents that click and type for you, and asks what a person who cannot watch the screen needs in order to follow one, stop it, and take over.
Ananya Gubbi Mohanbabu and the Control You Can Get Back To
Ananya Gubbi Mohanbabu, a UC Berkeley PhD student, builds tools for blind and low-vision people who use screen readers, software that speaks the screen aloud. One of her systems hides page elements that seem irrelevant to the user's stated goal. In a 12-person study, its screen reader users finished tasks in about half the time they took without it. It can also hide the button someone needed. Her newest work tests AI agents that click and type for you, and asks what a person who cannot watch the screen needs in order to follow one, stop it, and take over.
The Questions Completion Rates Never See
Boston gave residents seven categories for reporting city problems, plus "other." When a researcher read a sample of reports, nearly half sat in "other," and trash, among the most common complaints, had no box of its own. The city could see that gap only because people still reached for "other."
Agents that finish tasks for us may teach people to stop reaching for it. If a vague request flounders and a narrow one succeeds, users learn to ask narrowly. Completion rates climb, and nothing in them shows which questions disappeared.
The Questions Completion Rates Never See
Boston gave residents seven categories for reporting city problems, plus "other." When a researcher read a sample of reports, nearly half sat in "other," and trash, among the most common complaints, had no box of its own. The city could see that gap only because people still reached for "other."
Agents that finish tasks for us may teach people to stop reaching for it. If a vague request flounders and a narrow one succeeds, users learn to ask narrowly. Completion rates climb, and nothing in them shows which questions disappeared.
Your Ear Is Not a Lock, and Other Things a Pharmacy Technician Told Us Over the Phone
CONTINUE READINGShort Explainer
Say a program asks a server to make a payment, and nothing comes back. No error, no receipt. The connection hangs for a while, then gives up. On the server, the payment might have gone through, might be half done, or might never have started, and the program can't tell which.
A person can muddle through this. They check the bank app, dig through email, and call support. An automated program (an "agent") mostly gets two options, and both are guesses. If it retries, maybe the card gets charged twice. If it reports a failure, maybe the order already exists and some poor human gets to redo it by hand.
The web's basic rules (HTTP) won't bail you out here. The main HTTP spec flatly declines to define what happens inside the server. A timeout tells you nothing about what the server did.
So the agent has to keep its own records. Before sending, it attaches a unique ID (a "key") to the request and writes that key down. Then it can safely ask again, but only if the server knows what to do with the key. Without one, the agent is guessing with someone's credit card.
Further Reading
