SMS Group
Engineering2026 · 06· 1 min read

What "AI systems" actually are (and what they're not)

An AI system isn't a model. It's the pipeline, the data, the guardrails, and the operations around it. Here's how we think about building them at SMS Group.

There's a recurring conversation we have with founders and operators across East Africa and the Gulf. It starts with: "We want AI." It usually ends with us asking: "What do you actually want to happen, that isn't happening today?"

Interactive · System Anatomy

Click any layer to see what it does in a production system.

An AI system is not a model. The model is the smallest, easiest part. The system is the pipeline that feeds the model, the data that grounds it, the guardrails that constrain it, the integrations that make it useful, the monitoring that keeps it honest, and the team that operates it.

The Model

What everyone talks about

  • An LLM API call
  • System prompt engineering
  • Temperature / top-p tuning
  • Model selection
Effort share~10%

The System

What actually gets built

  • Data ingestion + cleaning
  • Processing pipeline + orchestration
  • RAG / vector search layer
  • Guardrails + output validation
  • Third-party integrations
  • Cost tracking + rate-limit handling
  • Structured logging + alerting
  • Deployment, versioning, rollback plan
  • Operational runbook
Effort share~90%

When we build an AI system at SMS Group, we start from the operation — the real work happening today, the bottleneck the team is fighting, the decision that takes too long. Then we engineer the smallest, simplest system that removes that bottleneck. Sometimes the answer involves a large language model. Often it involves three lines of code, a queue, and a webhook.

A vibe doesn't run in production at 2am. A system does.

Interactive · Request Journey

Trace a single WhatsApp message through a production AI system.

Press "Trace request" to watch a customer message travel through the system, or click any step directly.

The mistake we see most often is treating AI as a vibe. A vibe doesn't run in production at 2am. A system does.

If you're trying to figure out whether your problem is actually an AI problem, that's exactly the conversation our strategy call is for.

Interactive · AI Problem Diagnostic

Three questions. Find out if your problem is actually an AI problem.

1 / 3

Does your team repeat the same task or answer the same question more than 10 times a week?

Ready to build something real?

A 30-minute call is the fastest way to know if we're the right team for what you're building.