When AI notifications become the new meeting tax

Organisations spent a decade fighting meeting overload, and largely lost. The interruption problem never left; it simply changed shape. Microsoft's 2025 telemetry shows knowledge workers receiving a ping roughly every two minutes during core hours, about 275 a day, each costing some twenty-three minutes to fully recover from.¹ That is the baseline, and a new layer is now forming on top of it. AI tools have become, in one industry description, the newest and least-audited source of alert fatigue: smart nudges, real-time meeting summaries and automated check-ins that most notification policies were written before and never anticipated.² What makes them distinct is that they are always on and carry no social cost to the sender, which makes them harder to batch than a colleague's message.

For leaders focused on AI adoption, this is the part of the ledger that rarely gets totted up. The attention cost of the tools is real, it is accumulating, and almost nobody is measuring it.

The accountability tax

The volume of alerts is the smaller half of the story. A BCG study of 1,488 US workers found that AI widens what the researchers call the 'sphere of accountability'. Employees feel responsible for producing more work, monitoring more outputs and managing more information in the same hours, and the result is an acute cognitive fatigue that workers themselves have begun calling brain fry.³ This is not a soft wellbeing point. It comes with a number attached: BCG found that productivity gains flatten and then reverse after roughly three simultaneous AI tools, with brain fry raising error rates and intent to quit.³

For a CIO weighing the next tool to roll out, that is the single most decision-relevant finding available. It reframes AI adoption from a question of capability to a question of load. Three well-chosen tools that each remove work may help; a fourth and a fifth, each generating its own stream of nudges and summaries to be checked, can quietly tip a team past the point where the technology pays for itself.

A design problem, not a discipline one

The reason this accumulates unchecked is structural. Like other technologies built for maximum connectivity, AI tools ship without natural stopping points, the digital equivalent of a bowl that refills as you eat from it.⁴ Individual willpower is outmatched by design, which is why the familiar advice to turn off non-essential alerts only goes so far. The organisational picture confirms the pattern. Asana's survey of more than 9,000 knowledge workers finds 84% digitally exhausted, up nine points year on year, and 77% reporting unmanageable workloads, while only about 29% of organisations actually redesign work around AI. The rest bolt it onto existing workflows and automate the chaos.⁵

None of this is an argument against AI, and the evidence is explicitly two-sided. The same BCG study found that using AI to remove routine, repetitive tasks cut burnout by around 15% and lifted engagement; the harm came specifically from oversight-heavy, poorly-configured, stacked deployments.³ Context-aware assistants can triage email and batch low-value alerts, so a well-configured tool actually cuts interruptions. The honest stance is to manage the notification layer as deliberately as the deployment itself: configured well, audited, and measured for what it does to attention.

That points to a practical response with three parts. Set async-first communication norms, so that not every AI output demands an immediate human reaction. Adopt organisation-wide focus-block policies the tools are required to respect. And run a notification-hygiene audit across the AI stack, setting a per-role limit on how many tools any one person is expected to monitor.² The levers that change the structural conditions sit with leadership; the individual reaching for do-not-disturb cannot pull them. That is why this belongs on a CIO and CHRO agenda instead of a personal-productivity guide.

The missing ingredient is measurement. The interruption cost of an AI rollout can be read in work-pattern data, connecting a tool's notifications to the focus blocks they fragment, before it hardens into burnout or attrition. Most organisations can tell you how many AI licences they have bought. The harder question, and the one worth asking before the next deployment, is how much of their people's attention those licences are quietly spending?

Footnotes

1. Speakwise. (2026). Focus time statistics 2026 / Notification overload statistics (Microsoft 2025 Work Trend Index, via Speakwise). https://speakwiseapp.com/blog/focus-time-statistics ↩

2. Wellhub. (2025). Stop the ping tax: How HR can protect deep work and employee wellbeing. https://wellhub.com/en-us/blog/wellness-and-benefits-programs/alert-fatigue-deep-work-hr ↩ ↩²

3. Bedard, J., Kropp, M., Hsu, M., Karaman, O. T., Hawes, J., & Kellerman, G. R. (2026, March 5). When using AI leads to 'brain fry.' Harvard Business Review / Boston Consulting Group. https://www.bcg.com/news/5march2026-when-using-ai-leads-brain-fry ↩ ↩² ↩³

4. GMU College of Public Health. (2026, March). AI and the rise of cognitive overload. George Mason University. https://publichealth.gmu.edu/news/2026-03/ai-and-rise-cognitive-overload ↩

5. Asana. (2025). State of AI at work report (survey of 9,000+ knowledge workers; reported via LinkedIn summary). https://www.linkedin.com/posts/tbowman131_ai-wont-fix-broken-work-you-cant-just-activity-7378575089640042496-vnjN ↩