Deep Work When You Get Paged: Attention for On-Call Developers

You cannot mindfulness-app your way out of a PagerDuty rotation. The generic deep work advice on the internet assumes you control your calendar, your notifications, and your nervous system. Ops engineers control none of those fully. That does not mean attention is doomed. It means the system has to be different.

Quick summary: This post gives on-call developers and SRE-adjacent builders a realistic attention budget: how to carve defensible focus windows, how to negotiate expectations with managers, how to price context switches honestly, and when interruption is not failure but duty. Pair it with judgment about information diet in Signal vs noise: a developer field guide to tech news. If labour-market anxiety drives you to overwork, separate hype from your actual stack exposure using /tools/will-ai-replace-me.

Why Most Deep Work Advice Fails for Ops

Calm morning blocks and phone-in-a-drawer rules assume failure mode is distraction. Your failure mode is duty. A page is not a distraction; it is the job. Advice that treats all interruptions as moral weakness will make you feel guilty for doing the work correctly.

The useful split is not "deep versus shallow." It is interruptible versus non-interruptible cognitive work. Code review can survive a pause. Designing a data model often cannot. Incident analysis can be sliced into timed boxes. Writing a complex migration plan cannot.

If you schedule "deep work" without naming what kind of cognition you are doing, you will always lose to reality.

Pair programming and mobbing can help when the work is interruptible but complex: one person drives while another handles the pager context switch. That is not deep work in the romantic sense; it is shared working memory, which is often the right tool when duty is unpredictable.

Attention Is a Budget, Not a Virtue

Think in units per day, not vibes. Sleep is the first line item. If you steal from sleep to "catch up," you borrow from tomorrow's attention with interest.

Meetings are not neutral; they are pre-paid context switches. A 30-minute meeting often costs 45 to 60 minutes of usable focus because of ramp down and ramp up. Track that honestly for a week. You will hate the result, which means it is useful.

Pager load has variance. Build slack into the budget on weeks you are primary. If your calendar pretends every week is identical, it is lying to you.

Caffeine can mask sleep debt but does not repay it. If you find yourself triple-shooting espresso just to read a config file, the fix is schedule and staffing, not a darker roast.

Tools and Notifications: Defaults That Steal Hours

Every tool ships with engagement defaults: email on every comment, mobile push for mentions, red dots that never clear. Those defaults optimise the vendor's retention curve, not your cognition. Spend one hour quarterly auditing notification settings across git host, chat, CI, and ticketing. Turn off anything that duplicates another channel.

Batch CI noise: failing main matters; every intermediate flake on a feature branch might not. Route alerts by severity and ownership so your brain learns red means act.

Phone posture: if you can, keep work chat off the phone except on-call weeks. The pocket device turns five-second checks into twenty-minute detours. If policy requires mobile install, use per-app timers as a guardrail.

AI assistants in the IDE can help or hijack: constant autocomplete churn feels productive while fragmenting intent. Try muting inline suggestions during deep blocks and turning them on for boilerplate passes.

Defensible Windows: How to Protect 90 Minutes

Pick one non-negotiable focus window on non-primary days, and one shorter window on primary days if possible. Ninety minutes is enough for hard thinking; two hours is a luxury. If your life includes caregiving or split time zones, defend 45 minutes twice in a day instead of one heroic block; the total minutes matter less than the guarantee that some minutes exist.

Defensibility requires social contracts, not personal willpower. Tell your team: "I am offline 9:30 to 11:00 except pages." Put it in calendar with a literal name people can respect. If your culture cannot tolerate that, the problem is cultural, not individual.

Use physical or profile signals: noise cancelling, status set to deep work, chat in DND. These are not affectations; they reduce ambient pulls.

Batch shallow work: tickets, email, Slack triage in a defined block after your deep window. Inverting the order trains you to start the day reactive.

The Context Switch Tax and How to Pay It Once

Every switch has a tax: reload mental state, rebuild invariants, re-run grep in your head. The tax is higher for junior engineers and for unfamiliar codebases. Name three practices that lower it for your team: good runbooks, consistent repo layout, quick local repro commands.

Warm-up notes help: a sticky doc with "where I was" costs 60 seconds and saves 15 minutes. Treat it like a mini flight recorder.

Avoid false multitasking: reading Slack while compiling is not free; it fragments working memory. Either wait or step away. The middle state is the expensive one.

Measure one week honestly: count how many times you started a hard task and abandoned it within twenty minutes because of chat or email. If the number shocks you, that is data you can bring to retro, not shame you should carry alone.

Async Discipline Your Team Can Actually Keep

Deep work requires async hygiene at the team level, not only individual discipline. Defaults matter: threads instead of DMs when knowledge should spread; summaries instead of play-by-play; decisions written with links; meeting notes that answer "what changed" in five lines.

If your organisation treats slow Slack response as disloyalty, no personal technique fixes that. Push for explicit norms: expected response times by severity, after-hours rules, and recognition that knowledge work is not vending-machine latency.

Managers: What to Ask For Without Sounding Lazy

Good managers want throughput, not theater. Translate your ask into outcomes: "I need 90 minutes to finish the failure-mode analysis; without it we risk a Friday deploy blind spot." That is not laziness; it is risk communication.

If your manager cannot protect time, negotiate scope instead: drop a nice-to-have sprint item, shift a meeting, or pair on triage so pages cluster. The worst outcome is silent suffering that becomes silent resignation.

Document chronic interrupt patterns with dates. Anecdotes get dismissed; a month of data changes staffing conversations.

If you are a tech lead without people management authority, you can still publish a team working agreement doc and ask for explicit sign-off. Authority sometimes follows clarity.

On-Call Ethics: When You Should Drop Deep Work Instantly

Some work is not worth protecting at the wrong moment. Customer pain, safety, security, data loss: if your contract with users says you respond, you respond. Deep work systems that shame people for answering pages are unethical systems.

The fix is not guilt. The fix is roster design: enough people, sane hours, handoffs that do not require heroes. If the organisation relies on individual sacrifice forever, no focus hack fixes that.

If you lead, watch for moral licensing: praising someone for answering every page quickly, then wondering why design quality slipped. Reward sustainable systems, not chronic heroics. Rotate shadow shifts so junior engineers learn without burning out seniors.

Recovery Blocks and the Day After a Bad Night

After a rough page storm, schedule recovery like work: light tasks, documentation, small fixes. Pretending to do architecture on 4 hours of sleep produces architecture-shaped mistakes.

If you have flexibility, swap deep thinking to the afternoon when adrenaline crashes. If you do not, shorten the ambition of the task list without hiding the reason. "Today is recovery mode" is professional honesty.

Longer incidents also produce emotional residue: hypervigilance, irritability, second-guessing. Name it. A short walk, a debrief with coffee, or thirty minutes of low-stakes cleanup work is not slacking; it is resetting the nervous system so tomorrow's deep block is not sabotaged by yesterday's cortisol.

When to Revisit This System

Revisit quarterly. Teams change, services change, children get sick, managers rotate. A focus system that worked in a six-person startup will choke in a compliance-heavy org. Treat the playbook like config: version it, adjust it, delete superstitions that stopped matching reality.

Key Takeaways

• Ops reality: pages are duty, not distraction; design systems around interruptible versus non-interruptible work.
• Budget attention like money: include sleep, meetings as taxed time, and pager variance.
• 90-minute window: protect one defensible block on light weeks; shorten but do not zero it on primary.
• Social contract: calendar + team norms beat willpower; batch shallow work after deep blocks.
• Context tax: runbooks, warm-up notes, and honest single-tasking reduce reload cost.
• Managers: frame focus as risk reduction; use data if norms resist.
• Ethics: answer real emergencies; fix roster design instead of blaming individuals.
• Recovery: schedule light work after bad nights; shrink ambition transparently.
• Tool defaults: audit notifications quarterly; batch CI noise; keep pocket devices from owning your attention.

Attention is finite. Good systems admit that. Great teams build around it.