Last week, I finally cracked. After months of trying to get my inbox in check, unsubscribing from mailing lists, deleting unneeded messages and filtering out the junk, I reached breaking point. More than 110,000 (no, the comma is not in the wrong place) unread emails were cluttering up my inbox, and about 60 per cent of those were automated mailing lists, marketing mailouts or completely irrelevant to me. <\/p>\n
If hell is other people, the 2026 version is other people who keep adding you to their mailing lists. <\/p>\n
I have tried multiple ways to get that number to zero. Unsubscribing from the genuine mailing lists that I no longer need, marking the obvious scammers as spam and sending anything that I hadn\u2019t explicitly signed up for straight to the junk folder made a slight dent, but the number was still disturbingly high. I even tried hiding the badge showing how out-of-control the problem had become. Out of sight, out of mind \u2013 until your email programme reminds you that your inbox is teetering towards capacity. Deleting messages en masse wasn\u2019t an option. <\/p>\n
Being overly enthusiastic with the delete button in the past has meant I missed more than a few important emails. Recently, I repeatedly binned what turned out to be reminders to register for an event I had signed up for. That mistake that only came to light when a gentle reminder from an actual person arrived via WhatsApp<\/a>, and was easily undone. Other mistakes may not be a quick fix. <\/p>\n It was becoming a full-time job. Every time I thought I was making progress, a few days off or a weekend would put me back a few hundred messages. While writing this column alone, 44 new emails arrived. It is utterly relentless. <\/p>\n But spare a thought for Meta<\/a> Superintelligence\u2019s head of safety Summer Yue, who recently saw the downside of artificial intelligence (AI)<\/a> at close quarters. Yue had been testing out agentic AI \u2013 the next generation of assistants that will carry out tasks independently on behalf of people \u2013 through a service called OpenClaw, running it on a test inbox to organise her email. <\/p>\n When she launched it on her main inbox, it began to delete everything and lost some of the instructions, specifically the ones to stop and ask for confirmation. It was a \u201crookie mistake\u201d, she said, comparing it to defusing a bomb. <\/p>\n Agentic AI is considered the next big thing, the pay-off for all the investment being poured into the AI money pit. When AI agents are mainstream, we will be able to sit back while the robots take on the tasks of online shopping, booking flights, organising emails and generally managing the boring bits of admin that we don\u2019t even want to think about. And they will do it around the clock. In theory, at least. <\/p>\n As Yue\u2019s experience shows, we are not quite there yet. If an experienced tech professional can go wrong with AI, imagine the mischief the rest of us could get into. <\/p>\n