Signal
AI assistants have started acting before anyone asks. OpenAI's Pulse researched overnight and opened each morning with a set of cards it had put together on its own, and Google's Spark keeps working in the background after the laptop is shut. The counter-move arrived inside the same launches: Spark is designed to ask first before spending money or sending emails, and Google's shopping agent promises to always ask for permission before it buys. Even Pulse ended up on the asking side, retired within nine months in favour of daily briefings the user sets up. There is a limit to how much unasked help people want. In a controlled experiment, programmers liked an assistant that offered a suggestion and then waited, but when it kept interrupting with new suggestions, fewer than half preferred it to a plain chat assistant.
So what?
Acting before you ask is becoming free. Asking first is becoming the feature.
Evidence
OpenAI, 25 September 2025 (launch): OpenAI released a preview of ChatGPT Pulse to Pro users on mobile. It researches overnight from a person's chats, feedback and connected apps such as the calendar and opens each day with a set of visual cards, under the line that ChatGPT can now start the conversation. On 17 June 2026 the release notes retired Pulse, moving proactive updates into tasks the user schedules.
Google, 19 May 2026 (launch): Google announced Gemini Spark, a cloud-based personal agent that keeps working in the background when the laptop is closed or the phone locked. Google says it is designed to ask first before high-stakes actions such as spending money or sending emails, with trusted testers first and a beta for US Google AI Ultra subscribers a week later.
arXiv, 28 February 2025 (study): in a CHI 2025 paper, a randomised experiment with 65 programmers compared a chat assistant that only answered when asked with three versions that offered help on their own. Nine in ten participants preferred the version that made a suggestion and then waited, while only 47 percent preferred the version that kept suggesting, which they called distracting and annoying.
Worth asking
Where in your work would you rather be asked than anticipated, even when the guess would be right?
