Upgrade & Secure Your Future with DevOps, SRE, DevSecOps, MLOps!
We spend hours scrolling social media and waste money on things we forget, but won’t spend 30 minutes a day earning certifications that can change our lives.
Master in DevOps, SRE, DevSecOps & MLOps by DevOps School!
Learn from Guru Rajesh Kumar and double your salary in just one year.
The platform team at a mid-sized freight-booking company had a problem common among teams adopting AIOps tooling: the tooling worked, but nobody on the team really understood why it did what it did. Remediation scripts fired overnight, a model flagged odd latency curves, and every morning someone said “huh, that was clever” without being able to explain it. This is the story of how ten minutes a day turned that shrug into real fluency.
The morning the team admitted the gap
It started after a minor incident. An automated restart had cleared a stuck queue worker at three in the morning, and the on-call engineer, still half asleep, could not say which signal had triggered it. The fix was right. The understanding was missing. In the retro, the team lead asked a plain question: how many of us could replay that decision from the logs, cold, without help?
Two hands went up out of seven. Nobody wanted to block out a whole day for a training session, and the roadmap was already crowded. So the team made a smaller promise. Each person would spend ten minutes a day on one automated decision from the previous night, and that was the whole commitment. No slides, no workshop, no certificate at the end. The agreement lived in a single line on the team’s whiteboard, and the line was easy to remember because it was so short.
What a ten-minute rep looked like in practice
The rule was boring on purpose. Open the decision log for exactly one automated action. Find the input that triggered it. Say, out loud or in a chat thread, why the system chose that response instead of another. Then stop. If the ten minutes ran out halfway through the reasoning, that was fine; the unfinished piece became tomorrow’s rep.
The first week was clumsy. People spent most of their time just locating the right dashboard panel and the right log stream. By the second week the hunting got faster, because the same few places kept coming up. One engineer started pasting the three most useful queries into a shared note, and the others added theirs. Nobody planned a knowledge base; it accumulated.
The reps that taught the most
- Replaying a scale-up decision and noticing the model had been watching queue depth, not CPU.
- Reading a suppressed-action record and discovering a guardrail nobody remembered configuring.
- Comparing two nights where the same symptom produced different responses, then tracing the difference to a changed threshold.
None of these discoveries needed a long meeting. Each one fit inside a single sitting, and each one changed how the team talked about the system. “The model thinks” slowly turned into “the model weighted this signal because of that rule”, which is a much more useful sentence at three in the morning, when the person reading it has no patience for guesswork and nobody else is awake to ask.
Why small reps stuck when big blocks did not
The team had tried a half-day deep dive the previous spring. Everyone enjoyed it, wrote a few notes, and within a month the details were gone. The ten-minute habit survived for a different reason: it was small enough to do on the worst days. During a release week, when a half-day was unthinkable, ten minutes still fit between stand-up and the first code review. Missing a day cost almost nothing, and picking the thread back up the next morning took no ceremony.
There was a quieter benefit too. Short sessions also made it socially safe to say “I do not understand this one.” In a long workshop, admitting confusion feels like holding everyone up. In a ten-minute rep, it is just the next question. The team began posting their unsolved puzzles in a thread, and answers tended to arrive from whoever happened to glance at it.
A mid-story detour into outside material
About a month in, one engineer wondered whether there was a gentler way to see which parts of their own operations they understood least. She found an engineering-operations site offering a short quiz plus a library of planning guides, worksheets and starter documents, and used the quiz to open a conversation rather than to hand down a verdict. Its own description frames the output as support for deciding, and the team took it in that spirit.
The team still did their reps from their own logs. The outside material only helped them decide which decision to open next.
How the habit changed the way they adopted AIOps
By the third month, the effect showed up in design discussions. When someone proposed turning on a new automated remediation, the first question was no longer “does the vendor demo look good?” but “could we replay this decision in ten minutes if it surprised us?” If the answer was no, the team asked for clearer logging before enabling the feature. That single question prevented at least two awkward rollouts, according to the team lead, who admitted that the habit had been more effective than any of the dashboards they had bought.
If you want to see how that thinking connects to the wider conversation, TheAIOps has a piece on how AI can support the decisions IT teams make every day, which pairs naturally with the replay habit described here. And for anyone whose first reaction to automation is an avalanche of notifications, the write-up on calming event noise with smarter tooling sets the stage for why replaying a single decision matters in the first place.
What the team would tell you to copy
When people ask what to steal from the story, the team gives a short list. Keep the rep tiny enough that skipping it never feels necessary. Pick exactly one decision per day, not a category. Write down the one thing you learned in a sentence, because the sentence is the only part you will keep. Share the confusing ones, not just the tidy ones. And resist the urge to expand the session when it goes well; the day you extend it to thirty minutes is the day it starts getting skipped.
The freight company’s platform team still runs the habit today. New hires get a short version of the rule on their first Monday, along with a pointer to the shared note of queries. Their automation has not become simpler, but their relationship with it has. Instead of treating the system as a clever stranger, they treat it as a colleague whose reasoning they can read, one small decision at a time.
Starting your own version tomorrow
You do not need approval or a budget to try this. Tomorrow morning, pick one thing your tooling did overnight, open the record for it, and give yourself ten minutes to explain it to an imaginary teammate. If you find the answer, write it down. If you do not, write down the question. Either way you have done a rep, and a pile of reps is how a team goes from watching its automation to genuinely understanding it.