Engineering & Architecture: Build Decisions This Week

Engineering & Architecture: Build Decisions This Week

2026‑09‑11

---

1️⃣ Platform Landscape – The Quiet Push for Cloud Value Realisation

The South African Cloud Value Realisation Index 2026, announced jointly by MyBroadband and BusinessTech, signals that enterprises are still struggling to translate cloud spend into measurable outcomes. “The cloud is a critical part of running a modern business,” the survey notes, urging companies to identify why they’re not getting full value from their investments.

> “The Index will identify whether businesses are getting full use of their cloud investment, and identify the biggest obstacles that are preventing this from happening.” – MyBroadband

The most tangible obstacle highlighted by respondents is IaC drift—provisioned infrastructure diverging from declared state. A small, targeted sprint to audit Terraform (or equivalent) modules for drift, enforce version constraints on provider plugins, and embed automated drift‑checks in CI pipelines can prevent weeks of manual remediation or unplanned cost spikes.

| Build Decision | What to Do | Trade‑Off |

|----------------|------------|-----------|

| IaC Hygiene & Drift Detection | Audit existing IaC modules for drift, tag with version constraints, add drift checks to every pipeline run. | Extra time spent now vs. potential months of manual investigation and unplanned egress charges later. |

---

2️⃣ Architecture Patterns – Edge‑First for AI‑Driven Surveillance

The South African Police Service (SAPS) has issued a tender requiring body‑worn cameras that perform facial recognition on‑device, with no fixed quantity announced.

> “The body‑worn cameras the South African Police Service (SAPS) wants to strap to its officers must be able to conduct facial recognition.” – TechCentral

Deploying AI inference at the edge dramatically reduces latency and data egress costs, but it also introduces challenges: model versioning, local storage limits, and higher initial development effort. A hybrid strategy—running lightweight models on‑device for real‑time identification while sending aggregated metadata to a central analytics platform—balances responsiveness with maintainability.

| Build Decision | What to Do | Trade‑Off |

|----------------|------------|-----------|

| Edge AI vs Centralised Inference | Deploy lightweight CNNs or quantised transformers on bodycams; route anonymised embeddings to cloud for deeper analysis. | Development overhead and model accuracy trade‑off versus reduced bandwidth consumption and faster response times. |

---

3️⃣ Developer Tooling – Compliance as Code

The court’s recent ruling that BEE licensing in the property sector was “inconsistent and invalid” (Moneyweb) underscores a broader theme: regulatory compliance is becoming a moving target. SAPS’s biometric requirement introduces obligations under POPIA, UK GDPR, and EU AI Act for any downstream processing or storage of facial data.

> “Court strikes down ‘inconsistent and invalid’ BEE licensing in property sector.” – Moneyweb

Embedding compliance checks into the CI/CD pipeline—such as automated schema validation against POPIA’s consent requirements, audit‑logging policies, and AI model fairness tests—creates a Compliance-as-Code culture. This approach mitigates the risk of costly post‑deployment remediation.

| Build Decision | What to Do | Trade‑Off |

|----------------|------------|-----------|

| Compliance‑as‑Code for Biometric Data | Add static analysis of data flows, consent checks, and model bias tests in every merge pipeline. | Extra build time vs. reduced legal exposure and faster regulatory reviews. |

---

4️⃣ Scaling Decisions – Balancing Cost & Performance

Capitec’s latest earnings outlook shows an 18‑20 % rise in both EPS and HEPS despite high inflation and interest rates, a testament to robust scaling of digital banking services.

> “Capitec is expecting a notable rise in both its earnings per share and headline earnings per share (HEPS) in its next interim results.” – BusinessTech

This financial momentum encourages investment in elastic infrastructure. Leveraging serverless functions for burst‑heavy workloads, coupled with reserved instances for steady traffic, can keep cost curves flatter while delivering near‑zero latency for core services.

---

5️⃣ Security Posture – From Perimeter to Privacy‑First

With AI bodycams becoming mainstream and cloud investments still under scrutiny, the security lens must shift from perimeter hardening to privacy‑first design. Implement end‑to‑end encryption for data in transit and at rest, enforce strict role‑based access controls (RBAC), and audit all biometric storage against POPIA’s “Sensitive Personal Data” category.

---

6️⃣ What to Pay Attention To This Week

| Priority | Why It Matters |

|----------|----------------|

| IaC Drift Audits | Direct impact on cloud spend, immediate risk mitigation. |

| Edge AI Deployment Strategy | Determines SAPS tender viability and operational cost. |

| Compliance‑as‑Code Pipeline | Legal compliance for biometric data; mitigates future penalties. |

---

7️⃣ What to Ignore (for now)

• Feature‑parade on new cloud services that have not yet demonstrated clear ROI in the South African market.
• Unverified “AI‑security” hype from vendors lacking concrete regulatory alignment.

---

> “The expected improvement in earnings was supported by continued growth across the group’s….” – BusinessTech

---

Review Note

• IaC Tool Assumption: The recommendation assumes use of Terraform; verify if other IaC solutions (Pulumi, CDK) dominate your stack.
• Edge AI Model Choice: Specific model families (e.g., MobileNet, Tiny‑Yolov5) are not cited in sources; validate performance trade‑offs locally.
• Compliance Scope: POPIA clauses referenced by the SAPS tender are inferred; cross‑check with legal counsel on exact consent and retention requirements.

---