The suspect nobody books
A founder tells me their best engineer "isn't shipping like she used to." The ops lead "takes forever on reports now." The team is good, the tools are good, and somehow everything moves slower every quarter. Then we pull one ordinary day apart, hour by hour, and find the engineer's calendar: a 9:30 standup, a 10:15 "quick sync," Slack expectations measured in minutes, four channels with @here privileges, and a "focus block" at 14:00 with a dentist's-office quality to it — technically reserved, constantly violated.
She isn't slower. She's interrupted — and the difference matters because the two problems have opposite fixes. You manage a motivation problem with people skills. You manage an interruption problem with architecture. Teams everywhere are applying the first fix to the second problem, then escalating it when it fails.
The science: residue and the 23 minutes
Two findings carry this whole topic, and they're worth knowing by name.
Attention residue — Sophie Leroy. When you switch from Task A to Task B, part of your attention stays with A — measurably. Leroy's experiments showed performance on B degrades while the mind keeps processing the unfinished A, and the residue is thickest when A was interrupted before completion. The implication inverts how people schedule: it's not the minutes the interruption takes that cost you — it's the contaminated quality of the minutes after. You return to the report, but the Slack thread came with you.
The refocus tax — Gloria Mark. Her workplace observation studies found knowledge workers switch activities every few minutes, and that returning to full engagement with an interrupted task takes, famously, around 23 minutes. Not 23 minutes of absence — 23 minutes of being back but shallow: rereading the paragraph, rebuilding the mental model, finding the thread. Her studies also caught the compensation move: interrupted people work faster to catch up, at the price of measurably higher stress and frustration. The team feels busier and slower simultaneously — which is exactly how your office feels, and now you know why.
The interruption takes thirty seconds. The recovery takes twenty minutes. Your calendar bills you for the first and hides the second — all day, every day, for everyone.
The bill, itemized
Conservative numbers, one knowledge worker: 15 meaningful interruptions a day (observational studies typically find more), each costing even 10 minutes of degraded refocus — call it 2.5 hours daily in the recovery zone. That's over 30% of the workday, paid at full salary. For a 10-person team at modest knowledge-work salaries, fragmentation runs comfortably into six figures annually — a sum that would trigger an emergency meeting if it appeared on any invoice, and triggers nothing because it appears on none.
And the spreadsheet still misses the worst line: category loss. Some work doesn't degrade gracefully under fragmentation — it simply fails to occur. The strategy rethink, the real document, the gnarly debugging session: these need 60–90 minutes of unbroken runway (flow has entry conditions, and a runway is the first one). A day of twenty-minute fragments contains zero deep-work slots regardless of total hours — which is why the important project is perpetually "next sprint" while everyone works flat out. The hours were all spent; they were just spent in denominations too small to buy anything that matters.
The multitasking autopsy
The defense everyone offers — "I'm good at multitasking" — has been thoroughly autopsied. For cognitive tasks, the brain doesn't parallel-process; it switches rapidly, and the task-switching literature (Rubinstein, Meyer & Evans) measured the toll: every switch costs time and accuracy, scaling with task complexity. The Stanford heavy-media-multitasker studies added the uncomfortable coda: people who multitask most — and rate themselves best at it — test worse on the underlying cognitive controls, not better. The skill people think they've built is, measured, the damage.
The one legitimate exception: pairing thinking with the genuinely automatic — walking while talking, folding laundry while listening. Two thinking tasks always queue behind the same bottleneck. The bottleneck is you, and it has never once been dual-core.
The architecture that buys it back
- Team quiet hours — formal, defended, daily. A 2–3 hour window where chat expectations are officially suspended: no @here, replies not expected, emergencies have a separate channel (which, once defined, turns out to carry two messages a month). The formality matters — individuals can't unilaterally disarm in an armed room; the policy disarms everyone at once.
- Async by default. Replace "online = responsive" with an explicit reply expectation — four hours, say. The always-available tax is the single biggest residue generator, because it keeps a listening thread open in every head all day. Written-first culture also forces clearer asks, which kills a second class of interruptions: the clarification ping about the first ping.
- Cluster the meetings. Two meetings with a 40-minute gap don't cost two meeting-lengths — they cost the morning, because 40-minute gaps are too short for deep work and too long to waste. Stack meetings back-to-back in defined bands (early afternoon protects the morning peak — see energy curves), and the same meeting load suddenly coexists with real focus time.
- Batch the shallow work. Email at 11:30 and 16:00 beats email-as-ambient-condition. Same throughput, a fraction of the switches. The inbox is other people's priorities arriving in random order; batching converts it from a continuous interrupt into a scheduled task.
- Fix the upstream causes of "quick questions." Most recurring interruptions are documentation failures wearing social clothing: the same five questions, asked serially, forever. Every repeated question is a missing one-page SOP announcing itself.
Stop treating interruptions as a personal discipline problem — "people should protect their focus" — and start treating them as traffic design. Nobody asks pedestrians to personally negotiate with cars; you build crosswalks. Quiet hours, async defaults, and meeting bands are crosswalks. Build them once; everyone's attention crosses safely forever.
Where the pings should go instead
Audit a week of your team's interruptions and a pattern appears: a large share were never conversations at all. Status requests ("where's the order?"), routine lookups ("what's our policy on…?"), schedule changes, data shuttling between tools. These are queries and transactions wearing the costume of messages — and they interrupt a human only because no system exists to receive them.
That's the quiet second payoff of the agent era: an AI agent that answers status questions, reschedules appointments, chases the invoice, and updates the CRM doesn't just save the minutes of the task — it deletes the interruption that the task used to arrive inside, along with its 23-minute shadow. When we run audits, fragmentation is often the largest hidden line: not the hours of machine-work itself, but the focus its random arrival was destroying around it. Recover both and the team you already employ gets measurably larger — without a single hire.
How fragmented is your team's day?
The audit maps where the interruptions come from, what they cost in recovered-hours terms, and which ones an agent can absorb entirely — before anything is built.
Book a Free Audit →Frequently asked questions
What is context switching and why is it costly?
Moving attention between tasks needing different mental models. Residue from the previous task degrades the next (Leroy), and full refocus after interruption takes ~23 minutes (Mark). The switch feels free; the recovery is the cost.
How much productivity is lost to context switching?
Conservatively 2.5 hours per knowledge worker per day — 30%+ of the workday in the recovery zone — plus the category loss: deep work that never happens in fragments at all. Team-level, it's six figures that appear on no invoice.
How do I reduce context switching for my team?
Architecture: formal quiet hours, async-by-default reply expectations, clustered meeting bands, batched shallow work, SOPs for repeated questions — and agents absorbing the status-request traffic entirely.
Is multitasking actually a myth?
For cognitive work, yes — it's rapid switching with a measurable toll, and self-rated multitaskers test worse. The only free pairing is thinking plus something genuinely automatic.