ClickOps, Chaos, and AI Misfires: Why Solid IaC Still Wins
Artificial intelligence is accelerating across the industry, promising automation, efficiency, and intelligent operations. But while AI is powerful in many domains, infrastructure management is not one of them.
Recent incidents have shown that AI-driven automation, when granted control over infrastructure, can cause catastrophic, irreversible failures. From wiping out storage volumes to erasing production databases without backups, AI has demonstrated that it lacks the deterministic behavior required for safe infrastructure operations.
This article breaks down the risks, the real-world failures, the parallels with click-ops anti-patterns, and the role of well-structured Infrastructure-as-Code (IaC) and configuration management. We’ll also cover the limited areas where AI genuinely benefits infrastructure teams.
1. AI Lacks Predictability, and Infrastructure Demands It
Infrastructure management is fundamentally deterministic:
- A change must produce the same result every time (and across resources).
- The outcome must be predictable before execution.
- Every modification must be visible, tracked, and reversible.
AI violates all three principles.
Large language models are non-deterministic by nature - responding differently based on context, temperature, prior prompts, or internal heuristics. What works today may behave differently tomorrow.
And in infrastructure, that unpredictability is not just inconvenient - it's dangerous.
2. Real-World AI Failures Highlight the Risk
Recent, highly publicized examples include a cache clearing operation that removed an entire drive's contents, and the infamous production database deletion.
These are billed as vibe coding problems, but notice that both are actually infrastructure issues. When vibe coding goes wrong you just ctrl-z and move on. When your database is deleted you're in for a much rougher ride.
At Absolute Ops we regularly test the state of the art in AI driven infrastructure development. We have found many examples of very scary behavior including a lot of subtle security problems and unnecessary rebuilds which could lead to outages and data loss.
3. AI-Driven Infra Automation Feels a Lot Like Click-Ops
Click-ops is considered an anti-pattern for good reason:
- Changes are invisible
- They’re not repeatable
- They’re not version-controlled
- They’re not tested
- They’re not peer-reviewed
AI-generated infrastructure changes suffer from the same problems:
- AI may generate different code each time
- It may structure resources inconsistently across environments
- It may apply “creative” solutions that diverge from standards
- It may skip critical steps if not explicitly told to include them
- It may hallucinate configuration values or defaults
This inconsistency erodes reliability and supportability - the exact problems IaC was invented to fix.
4. Infrastructure Requires Guardrails - IaC Provides Them
Deterministic automation tools like:
- Terraform / OpenTofu
- Ansible / Chef / Puppet / Salt
- Kubernetes manifests / Helm
…exist because infrastructure demands:
- strict version control
- repeatable builds
- reproducible state
- predictable diffs
- validated plans
- enforcement through policy
- promotion workflows through environments
- traceability for audits and compliance
AI cannot replace these practices.
It can accelerate parts of them - but it should never be the execution engine and it should never be trusted.
5. Where AI Does Provide Real Value in Infrastructure Work
AI is dangerous at execution time, but beneficial in supporting workflows.
Here are areas where AI genuinely helps:
A. Code generation (but not execution)
AI can draft:
- Terraform modules
- Bash scripts
- YAML templates
- Policy-as-Code examples
- Infrastructure diagrams
- Documentation
These drafts still require human review, adjustment and testing before they’re run.
B. Troubleshooting and root cause identification
AI excels at:
- explaining error messages
- identifying misconfigurations
- summarizing logs
- generating hypothesis trees
It is an excellent assistant for debugging, not deploying.
C. Knowledge retrieval & search
AI helps teams:
- find cloud documentation
- interpret pricing models
- locate best practices
- explore architectural patterns
This accelerates engineering without introducing execution-time risk.
D. Recommendations, not actions
AI is powerful when used for:
- cost anomaly alerts
- rightsizing suggestions
- policy recommendation drafts
- security misconfiguration reports
But an experienced human with strong domain knowledge must validate the final action.
6. The Path Forward: AI + IaC, Not AI Instead of IaC
The safe future of infrastructure automation isn’t model-driven execution - it’s AI-assisted engineering with deterministic delivery pipelines.
The workflow of the future looks like this:
- AI drafts or recommends changes
- Engineers review, adjust, and enforce standards
- IaC stores the canonical source of truth
- Policies validate correctness and security
- A controlled orchestrator executes approved changes
- Monitoring and drift detection verify results
AI becomes an accelerant, not a controller.
This maintains the guarantees that infrastructure demands:
- Safety
- Repeatability
- Predictability
- Compliance
- Auditability
- Operational uptime
AI alone cannot provide these guarantees - but IaC and configuration management can.
Final Thoughts: AI Is a Tool, Not an Operator
AI is one of the most powerful tools ever created, but infrastructure operates on principles that AI cannot satisfy:
- determinism
- predictability
- strict safety boundaries
- explicit state management
- reliable change control
Using AI as an autonomous infra operator is dangerous - as real-world failures have already shown.
But using AI as an assistant for code generation, analysis, documentation, and troubleshooting?
That’s where AI truly shines.
The future isn’t AI replacing infrastructure engineers.
The future is infrastructure engineers who use AI - safely, carefully, and with the right guardrails.
And a Final Note of Caution
The danger of engineering complacency is real with AI. Several recent studies, like the one from MIT reported on here show that AI dulls our critical thinking skills.
There are many reasons for that, but a big one is that it's an almost perfect SEP Field (Somebody Else's Problem). Once we hand responsibility for infrastructure management to AI, we can easily check off the task as no longer relevant to us.
Lack of code review depth on an application can lead to increased test failures and development cycles. In IaC it can lead to disaster.
