We get asked roughly weekly: “where’s the AI in Vloud?” The answer is “everywhere it helps.” A more interesting list, though, is the places we deliberately don’t have AI, and the reasons.
Where AI shows up in Vloud (the easy list)
- AI Doctor explains deployment failures
- The AI Code Assistant summarizes long log windows
- Smart triage groups likely-related security events
- The dashboard’s “what changed today” panel uses an LLM to write a short narrative from the day’s audit log
- Setup wizard suggests a config based on what’s running on your host
These are all places where AI does what AI is good at — synthesis, explanation, fuzzy matching — under operator supervision.
Where AI deliberately doesn’t show up
License verification. Cryptographic signatures only. No “the model thinks this license looks valid.” A license is valid or it isn’t.
Threat detection — at the decision boundary. Detection itself can use ML/AI heuristics; the decision to act (quarantine, block, alert) is rule-based and explainable. Every triggered alert has a cited reason, not a model confidence score.
Backup verification. Hash comparisons only. There’s no interpretation. Either the bytes match or they don’t.
Audit log. Every entry is structured, indexed, and queryable. We do not summarize the audit log into a narrative for compliance purposes — auditors get the raw events, with the published Merkle root. The model can describe trends; the canonical record is the chain.
Self-healing remediations. The decision to remediate is rule-based. The action set is whitelisted. The trigger conditions are explicit, not learned. We don’t ship “the AI noticed something weird and decided to restart the database.”
Pricing tiers. No usage-based “AI predicted you’d want the business plan.” We tell you what each tier costs. You pick.
Why the list matters
AI is a great answer to “what’s the most-likely cause of this failure?” It’s a terrible answer to “should we actually take this destructive action?” The first is exploration. The second is decision-making with consequences.
We use AI for the first and refuse it for the second. The rule isn’t “AI bad” or “AI good.” The rule is AI assists, deterministic systems decide. Every place we use AI is a place where being wrong costs nothing more than the operator scrolling past the recommendation.
Every place we don’t use it is a place where being wrong costs the customer real harm.
If you can articulate that line in your product, you can ship AI features without the liability tail. If you can’t, you ship a product that occasionally sets your customers on fire and calls that “innovative.”
We picked the line. It’s working.