The best operations person I ever worked with spent most of her time reading a book at her desk. Deploys — automated releases of new code to live servers — went out on time. She resolved incidents before anyone noticed. New hire onboarding ran like clockwork.

Her manager almost fired her for "not being busy enough."

She wasn't idle. She was finished.

The gap nobody talks about

Work culture rewards visible effort. The developer typing furiously looks productive. The one staring at the ceiling for twenty minutes — thinking through architecture — looks lazy. The person answering emails at 11 PM earns "dedicated." The one who leaves at five earns "not committed."

A 2022 study from researchers at Columbia, Georgetown, and Harvard confirmed what ops people already knew: managers consistently rated "busy-looking" employees as more competent, even when their actual output fell below that of calmer colleagues. We reward the appearance of work, not work itself.

On March 12, PagerDuty unveiled its SRE Agent as a virtual responder — software that detects outages, runs diagnostics, and follows fix-it procedures without a human touching a keyboard. Four days later, at GTC on March 16, NVIDIA announced the Agent Toolkit with OpenShell — infrastructure for running autonomous operations agents safely in production. On March 24, at YC Demo Day, startups like IncidentFox pitched autonomous incident response as their entire product. The signal from the market: if a task follows a predictable pattern, a human shouldn't do it by hand.

Which raises a question that ops teams everywhere now face: if AI agents — programs that act on their own, making decisions and executing steps without constant human supervision — handle the visible firefighting, what does an ops person do all day?

The answer hasn't changed. But the pressure to understand it has.

In operations, the old incentive structure creates a perverse dynamic — a system that rewards exactly the wrong behavior. If your company values you for fighting fires, you have zero motivation to prevent them. If your manager measures your worth by how many urgent Slack messages you handle, building systems that eliminate those messages makes you look dispensable. And now AI agents fight fires too. Faster. Without sleep. Without complaining.

The paradox at the heart of ops

The better you are at operations, the less you appear to do. A firefighter who prevents fires looks unemployed. An ops person whose systems never break looks like they're slacking. The visible work disappears precisely because someone did the invisible work right.

I've watched this pattern repeat for years. The ops person who automates their job gets questioned: "What do you do all day?" The one who manually handles every incident, working twelve-hour days, gets promoted for "going above and beyond."

One built a system. The other built a dependency on themselves. Ask yourself which one the company actually needs — and which one an AI agent replaces first.

What good ops looks like in practice

Good operations work happens in two phases.

Phase 1: Build the systems. This part is visible and time-limited. Writing runbooks — step-by-step guides for handling specific situations. Setting up monitoring — automated checks that catch problems before users notice. Creating automation for repeating tasks. Documenting processes so anyone can follow them. This phase runs hot: typically two to six months of focused work.

Phase 2: Maintain the systems. This is where the confusion starts. The systems run. Alerts fire and runbooks handle them — increasingly, AI agents execute those runbooks without human intervention. New hires onboard themselves through documented processes. Deploys flow through CI/CD pipelines — automated sequences that move code from a developer's laptop to production servers without manual steps.

The ops person's job in Phase 2: watch for patterns that suggest a system is degrading. Run post-mortems — structured reviews of what went wrong and why. Plan for future capacity. Decide which new processes to hand off to agents and which still require human judgment. Read. Learn. Think.

That last part looks like "doing nothing." But an ops person who isn't studying new tools, modeling failure scenarios, evaluating which agent frameworks fit their infrastructure, and planning for situations that haven't happened yet will get caught flat-footed when they do. Google's Site Reliability Engineering handbook puts it plainly: the job is to engineer reliability, not to heroically recover from its absence.

Busy is a bug report

I'll say the quiet part loud: constant busyness signals broken systems, not dedication.

Always firefighting? Your prevention systems failed. Always context-switching — jumping between unrelated tasks every few minutes? Your prioritization failed. Always in meetings? Your communication systems failed. Always training new hires by hand? Your onboarding failed.

"Busy" is not a state to aspire to. "Busy" is a bug report.

The goal of operations — and honestly, most knowledge work — is to reach a state where systems handle the predictable ninety percent and you have bandwidth for the unpredictable ten. That unpredictable slice is where human judgment matters. Everything else should run itself. In March 2026, "run itself" increasingly means an AI agent runs it — and the ops person who built the system decides what the agent should and shouldn't touch.

The path out

If you're drowning in operational work right now, here's a practical sequence.

Weeks 1–2: Track everything. Every task, every interrupt, every recurring problem. Don't fix anything yet. Just observe.

Weeks 3–4: Categorize. What repeats? What follows a pattern? What could a script — a small program that automates a manual step — a checklist, or an AI agent handle? Typically sixty to seventy percent of operational work falls into "predictable and automatable."

Weeks 5–8: Automate or document the top ten time-consumers. One per week. Start with whatever interrupts you most. For incident response — the process of detecting and fixing outages — consider agent-driven triage: tools like PagerDuty's SRE Agent or open-source alternatives handle pattern-matched incidents and escalate novel ones to you.

Month 3: You now have forty to fifty percent more bandwidth. Invest it in the next tier of problems.

Month 6: You're reading a book at your desk. Your systems run. Your agents handle the predictable. You look idle. You're not idle. You're finished.

A note for managers

If your best ops person looks bored, congratulations. Your systems work. Do not assign busywork to justify their salary. Do not make them perform productivity theater — that performative scramble of looking occupied to satisfy someone's expectation of what "working hard" should look like.

Instead, ask them: "What would you build if you had three months of uninterrupted time?" Then give them those three months. What they build will save more than any amount of visible busyness ever could.

The most productive person in your company might be the one who appears to do the least. That's not a paradox. That's what finished looks like.