Deep work and agents

A pair of April 2026 posts about work habits and artificial intelligence agents landed on the same problem: people keep bouncing between Slack, Jira, and other tools, while software is increasingly being built to stay on one task until it is done. (x.com, x.com) Context switching is the act of leaving one task before finishing it and loading a different one into your head. Asana says the average worker now switches across nine apps a day, and cites notifications and fragmented tools as a core cause. (asana.com) That fragmentation is not small. A 2022 Harvard Business Review analysis of 137 users across 20 teams at three Fortune 500 companies found workers toggled roughly 1,200 times a day and spent just under four hours a week reorienting after those switches. (hbr.org) The three-hour “deep work” block recommendation in one of the posts is a scheduling response to that pattern: reserve a long stretch for one cognitively heavy task, and push messages and ticket churn outside it. Asana’s guidance uses the same logic, describing context switching as attention moving from one app, project, or task before the first one is finished. (x.com, asana.com) The agent side of the argument starts with a different capability. OpenAI says its Operator system, now folded into ChatGPT agent mode, can use a browser to click, type, scroll, and complete multi-step tasks with user handoff only when needed for things like logins, payments, or CAPTCHAs. (openai.com) Microsoft is making a similar pitch to office buyers. In an April 13, 2026 post, the company said Microsoft 365 Copilot agents can bring third-party business apps into Copilot chat so teams can update records, work on documents, and complete actions “without ever leaving Copilot chat.” (microsoft.com) That puts the current debate less on whether workers need more focus and more on where the switching happens. Humans are being told to carve out uninterrupted blocks, while vendors are trying to move the tab-switching, form-filling, and status-updating into software that keeps the task context loaded. (x.com, openai.com, microsoft.com) Employers are already measuring the strain of always-on work. Slack’s 2023 State of Work survey of more than 18,000 desk workers in nine countries found 53% felt pressure to respond quickly to messages even after working hours, and 63% said they tried to keep their status active online even when they were not working. (slack.com) The posts do not prove agents are better workers than people, and companies still have to decide where automation is safe, who approves actions, and which work needs a human in the loop. But they do capture the same shift: teams testing agents are also being pushed to redesign calendars, notifications, and tool sprawl around fewer interruptions. (x.com, microsoft.com, slack.com)