12 · AI Across IT: DevOps, QA, Data, Security and More

Coding is only part of IT work. AI helps across the whole software lifecycle: requirements, architecture, testing, operations, security, data, documentation and communication. This page is a tour of high-value uses in each area, with starter prompts and the cautions that matter.

Level 145 minTrack 17 · AI across IT

By the end of this page you can

AI across the software lifecycle

Planuser stories, specs,estimates, risks Designarchitecture options,ADRs, diagrams Buildcode, refactor,migrations Testunit/property/e2e,test data, coverage Review & secureAI code review,vuln triage DeployCI/CD, IaC,release notes Operatelogs, incidents,runbooks, postmortems Learnanalytics, feedback,docs, knowledge AI assists at every stage

Requirements, planning & project management

Here are rough notes from a product meeting. Turn them into user stories in the format
"As a <role>, I want <goal>, so that <benefit>" with Given/When/Then acceptance criteria.
Then list ambiguities I must clarify with the product owner.
<notes>
CautionEstimates from AI are guesses. Use them to structure discussion, not to commit dates.

Architecture & design

Act as a principal engineer. Here is my proposed design for an event-driven order pipeline
(Kafka + Scala services). Run a pre-mortem: list the 10 most likely ways this fails in production
(ordering, duplicates, poison messages, back-pressure, schema evolution...), and for each the
mitigation. Then write an ADR draft for the final design.

QA & testing

Given this function and its existing tests, list behaviours that are NOT tested, ranked by risk.
Then write the 5 most valuable missing tests.

DevOps, CI/CD & infrastructure as code

Write a GitHub Actions workflow for an sbt multi-module Scala 3 project: JDK 21, cache Coursier
and sbt, run scalafmtCheckAll, compile, test in parallel per module, and upload test reports.
Explain every step in comments.
High-risk area IaC and pipeline changes can delete infrastructure or leak secrets. Always run terraform plan / dry-runs and review them yourself. Never give agents production credentials. Check that AI-written pipelines don't print secrets to logs.

SRE, operations & incident response

Here is a JVM thread dump from a Scala service that stopped responding. Identify deadlocks,
thread-pool starvation (esp. blocking on a compute pool), and the threads most likely responsible.
Explain in plain language and suggest the fix.
<thread dump>
CautionLogs often contain personal data, tokens and internal hostnames. Redact them, or use an approved internal or local model.

Data engineering & analytics

Schema below (Postgres). Write a SQL query for monthly active customers per region for the last
12 months, where "active" = at least one paid order. Explain the query, then list assumptions
I should verify (time zones, refunds, test accounts).
<schema>
Always validate generated SQLCheck results on a small sample where you know the answer. Text-to-SQL is often almost right: a missing filter, a wrong join type, double counting.

Security (DevSecOps)

Create a STRIDE threat model for this feature: users upload CSV files that are parsed by a
Scala service and stored in S3, with results shown in a web UI. Output a table:
threat, STRIDE category, likelihood, impact, mitigation.
CautionAI review is an extra layer and does not replace security tooling (SAST/DAST/dependency scanning) or expert review for sensitive systems.

Documentation & communication

Frontend, UI & accessibility

Mobile & Apple platforms

IT operations, support & sysadmin

Explain this command piece by piece, then tell me if it is safe to run:
find / -name "*.log" -mtime +7 -exec rm {} \;
Never run commands you don't understand Especially commands involving rm, chmod -R, dd, curl | sh, firewall rules or production databases. Ask the AI to explain and to propose a dry-run first.

Your career & soft skills

Risk matrix: how much to trust AI per task

RiskExamplesApproach
LowExplanations, drafts, brainstorming, throwaway scripts, test ideasUse freely, skim-check
MediumProduction code, SQL for reports, docs, CI changesReview fully + automated checks
HighSecurity, IaC, data migrations, financial calculations, anything touching productionExpert review, dry-runs, staged rollout, two-person rule
Not appropriateUnreviewed decisions about people (hiring, performance), legal/medical advice, autonomous production changesHuman decision, with AI only as a helper
Try it (15 min) Pick the area of IT you touch least, for example CI/CD or incident response. Use one starter prompt from this page on a real or realistic example. Write down what you learned about that area, as well as what the AI produced.

Key takeaways