AI Try or Cry? Practical Lessons for IT Teams Adopting AI Without Breaking Things

The pressure to “add AI” is everywhere. But not every problem needs a model, and in IT operations, the wrong automation can create risk fast. This episode of the Simply Resilient Conversations podcast features CyberFortress team member and longtime Veeam Vanguard Jonah May in a discussion with Jeff on what AI is genuinely useful for in IT, what introduces risk, and how teams can adopt it with practical guardrails.

What you will get from this recap

• Where AI delivers real time savings for IT teams
• Why “AI everywhere” can make support and products worse
• A clear set of safety rules for using AI at work
• How to approach agentic AI without letting it cause damage
• What AI changes for IT careers, and what still requires human judgment

The core idea: durable value beats hype

The most useful way to evaluate AI in IT is to stop asking “How do we add AI?” and start asking:

• What problem are we solving?
• What is the safest and most reliable way to solve it?

In many cases, a deterministic script or rules-based automation is the better answer. AI earns its place when it measurably reduces time, improves accuracy, or helps teams handle complexity without creating new failure modes.

Where AI already helps in day-to-day IT work

1) Rapid log triage and troubleshooting

Logs are large, noisy, and time-consuming to parse. AI can accelerate the first pass by summarizing what stands out, grouping related errors, and pointing to likely root causes.

Use it for: narrowing the search space and speeding up investigation
Do not use it for: blind decision-making without validation

2) Code scaffolding for integrations and internal tools

AI is strong at producing a “first draft” for common patterns (connectors, basic UI flows, boilerplate, API wrappers). This is especially useful when teams are inheriting codebases, moving fast, or building small utilities.

Best practice: generate the structure quickly, then test, refactor, and harden it like any other code.

3) Smarter code reviews and faster QA

AI-assisted review can catch simple but costly issues early: duplicated variables, missed edge cases, crash-prone logic, and basic security concerns. It is not a replacement for review, but it is a useful second set of eyes that never gets tired.

Where AI can make things worse

AI support gates that block humans

A common pain point is AI chatbots and phone trees that make it harder to reach a human. For IT pros, this is especially frustrating because many support interactions happen after basic troubleshooting has already been done.

If AI is used in support, it should reduce time-to-resolution. If it slows down escalation, it is doing the opposite of its job.

The safety rules that actually matter

Keep secrets and sensitive data out of cloud models

Even when providers say they do not train on your data, the safest stance is to avoid sharing sensitive inputs.

Do not share:

• passwords, private keys, API tokens
• customer data, incident details, internal financials
• proprietary code or architecture that would be damaging if exposed

Use least-privilege roles

If a model has access to systems, scope permissions to the minimum required. Treat it like any other automation account and assume mistakes will happen.

Log everything

If AI touches operational workflows, build an audit trail by default. If something goes wrong, you need to know what happened, when, and why.

Prefer deterministic automation when safety matters

If the remediation is predictable, rules-based automation is more reliable than AI. A script will do the same thing every time. A model may not.

Agentic AI: value is real, risk is realAgentic AI is where models can take actions using tools and integrations, not just provide advice. This can be powerful, but it increases the blast radius when the model is wrong.

A safe approach looks like this:

• Sandboxed rehearsals first: test in isolated environments
• Immutable audit trails: every action is recorded and attributable
• Human-in-the-loop approvals: required for destructive actions (delete, disable, purge, credential changes)
• Clear “recommend vs execute” separation: start with suggestions, move to execution only after controls exist

What this means for IT careers

AI will change the workflow, not eliminate the need for judgment. Senior roles still require context, prioritization, risk decisions, and accountability.

At the same time, juniors can use AI to learn faster and become productive sooner by treating it like a tutor and accelerant, not an authority.

The winners will be the professionals who:

• know when to trust tools and when to verify
• build guardrails before automation
• can translate technical output into business risk and outcomes

A practical adoption plan for IT teams

1. Start with low-risk productivity wins
   Drafts, summaries, non-sensitive documentation, internal knowledge capture.
2. Use AI for analysis, not action
   Log triage, incident summarization, troubleshooting hypotheses.
3. Automate predictable remediations with scripts
   Rules-based actions for known patterns.
4. Introduce agentic AI only with guardrails
   Sandbox, least privilege, audit logs, human approvals.

Listen to the podcast to hear the full conversation.