Research figures drawn from peer-reviewed and primary sources cited on each page; verified April 2026. Your mileage will vary by team and context.
"It takes 23 minutes and 15 seconds to fully return to the task."
"Eliminate interruptions" is neither achievable nor desirable. An engineer who receives zero interruptions, zero meetings, and zero messages is also an engineer who is disconnected from team coordination, unaware of priority changes, and unable to collaborate. The goal is not to eliminate interruptions but to manage them by class.
A severity-aware interruption policy operates on a simple principle: compress Class 3 and Class 4 into Class 1 and Class 2. Reduce the frequency and impact of unplanned interruptions. Increase the share of batched and scheduled communication. Budget the scheduled interruptions carefully. This is achievable. A team that successfully moves 70% of its Class 3 traffic to Class 1 and eliminates most Class 4 interruptions outside of on-call has reduced its effective context-switching cost by roughly half, without eliminating any genuinely necessary communication.
Examples: Email, pull-request notification digests, RSS feeds, newsletters, Linear/Jira status digests
The defining characteristic of Class 1 is that the reader controls when to process. An email sitting in an inbox is not an interruption; it becomes one only when push notifications convert it to Class 3. The knowledge-work organisation's highest-priority task is expanding this class, not contracting it. Policies that move more communication into batched async formats (daily digest, end-of-day summary, weekly status post) directly reduce Class 3 and Class 4 costs without any loss of information.
Counterpoint: The Class 1 failure mode is volume: if an engineer receives 200 Jira digest emails per day, the processing time itself becomes a cost. The intervention is not to re-interrupt (push notification), but to reduce the signal-to-noise ratio of the digest.
Examples: Planned meetings, scheduled pairing sessions, daily standups, weekly 1:1s
Class 2 interruptions are known in advance, which makes them budgetable. An engineer can plan their morning around a 2pm meeting; they cannot plan around a 2pm tap on the shoulder. The cost of a Class 2 interruption is the meeting duration plus the 20% refocus tax on the hour that follows (approximately: for a one-hour meeting, 12 additional minutes of reduced productivity).
Counterpoint: Class 2 becomes toxic when meetings are scheduled back-to-back throughout the day, eliminating the margin between scheduled interruptions that makes recovery possible. The Paul Graham anti-pattern (see /ic-vs-manager-maker-schedule) - a single 11am meeting destroying the maker's morning - is the canonical Class 2 failure mode. The meeting itself is legitimate; its placement is the problem.
Examples: Slack DM with a red notification dot, @-mention in a team channel, email marked urgent, 'quick call?' message
Class 3 is where most organisational attention damage happens. The interruption is asynchronous in form (a message, not a physical presence) but synchronous in psychological effect (the notification demands attention immediately, by design). The tool's business model is the root cause: Slack, Teams, and most email clients are optimised for engagement metrics (daily active users, message open rates), not user focus time. The urgency signal is the product, not a side effect.
Counterpoint: The correct fix is not to remove Slack. It is to change the default from 'all-on' to 'all-off, opt-in for genuinely urgent'. A team agreement that DMs are for urgent-only questions (minutes-matter scale), combined with Slack channel notification muting, converts most Class 3 to Class 1 without losing any real information value.
Examples: Tap on shoulder, desk drive-by, phone call, someone physically walking over, 'do you have a minute?'
Class 4 is the most costly interruption and the most common in open-plan offices. The physical presence requires immediate social response; there is no way to 'batch process' someone standing at your desk. Gloria Mark's 23-minute refocus finding applies in full, and often more: a Class 4 interruption in the middle of a deep-work block typically terminates the block entirely, because the remaining time after the conversation is often too short to justify the mental cost of re-entering the work. A single Class 4 interruption per morning can halve an engineer's productive output for the day.
Counterpoint: On-call engineers are legitimately exposed to Class 4 interruptions as part of their role, and this is compensated accordingly. Pager duty is Class 4 by design. The argument is not to eliminate Class 4 for on-call engineers, but to eliminate accidental Class 4 for engineers who are not on-call.
Four actionable interventions, in order of leverage:
Compress Class 3 into Class 1. The team agreement that DMs are urgent-only, combined with channel muting, converts most Slack notifications to batched async. This single change, applied team-wide, reduces Class 3 frequency by 50-80% in most organisations without any loss of real communication value. Link to tools that hurt for notification settings.
Compress Class 4 into Class 2. Office hours on the calendar - a published two-hour window when a team member is explicitly available for walk-up questions - convert most desk drive-bys into scheduled or batched interruptions. Pair programming windows booked on the calendar turn ad hoc "can you look at this?" into a planned session.
Budget Class 2 carefully. No-meeting days, caps on total meeting hours per week (recommended: under 8 hours for engineers), maximum attendees for standing meetings, and explicit agenda requirements. See deep work blocks for the calendar design.
Expand Class 1 deliberately. Invest in async communication infrastructure: weekly written status posts, project decision docs, architecture decision records, retrospective pre-reads. Each async artefact replaced one synchronous meeting or several Slack notification pings.