Engineering & Architecture: Build Decisions This Week
2026‑10‑15
The past week’s headlines have reshaped the priority list for CTOs who serve both South Africa and the UK/European markets. Two converging forces—rapid infrastructure expansion in SA and a new insider‑threat risk profile—have nudged engineering leaders toward a trio of concrete build decisions that will determine resilience, compliance, and productivity over the next 2–3 years.
---
Fibertime’s latest rollout shows it has 604 k homes connected as of September 2026, up 245 % from September 2025 — a growth that translates into tens of thousands of concurrent active users each month (Source: “Company that wants to be the Capitec of fibre closes in on 2 million users” – MyBroadband). That traffic volume will saturate a single cloud region if the application is still tightly coupled to centralised services.
Trade‑off:
Given the engineer’s warning that South Africa will need 10 × more engineers to stabilize its grid (Source: “Acclaimed engineer who lost Eskom top job…has a warning about South Africa's grid” – MyBroadband), building local compute nodes or leveraging low‑latency edge functions can decouple service availability from national power fluctuations. For the UK/Europe, where power grids are stable but data residency rules tighten, a hybrid approach keeps the core in regulated regions while serving edge caches close to end users.
---
Uber’s internal engineering narrative shows how an organisation can “build internal developer tools, integrate AI agents into engineering workflows, and measure their impact on productivity” (Source: “Uber’s AI infra, from the inside” – Pragmatic Engineer). Their model is two‑fold:
Adopting a similar AI‑centric workflow can dramatically cut review cycles for new features, which is crucial when scaling to 2 M users on Fibertime’s network. The cost of training and hosting internal models must be weighed against the reduction in manual QA effort and faster time‑to‑market. Additionally, any AI model that ingests source code or logs will need to comply with POPIA (SA), UK GDPR, and EU AI Act data protection requirements—an overhead that should be accounted for in the tooling budget.
---
The arrest of a Saudi engineer who allegedly planted “a secret recording device behind a TV” at OpenAI underscores how insider threat now blends physical and digital vectors (Source: “Toilet paper and secret recorders…” – BBC Business). For high‑value IP teams—particularly those working on proprietary AI models—a single compromised meeting room can erode years of investment.
Build Decision #1: Deploy tamper‑resistant hardware (e.g., wired microphones, camera blocking solutions) in all executive and R&D rooms.
Build Decision #2: Implement a Zero‑Trust Network Access (ZTNA) model that ties device health to authentication tokens, ensuring that only vetted, encrypted endpoints can reach internal services.
Build Decision #3: Institute an employee vetting program that includes background checks for all roles with access to sensitive AI assets.
These steps will mitigate the dual threat of physical surveillance and data exfiltration but come at a tangible cost: hardware procurement, policy enforcement, and increased audit overhead. In South Africa, where budgets may be tighter, a phased rollout—starting with key IP custodians—offers a balanced approach.
---
With 37 k new homes connected in September alone (Source: “Company that wants to be the Capitec of fibre closes…” – MyBroadband), traffic spikes are expected. Auto‑scaling groups tuned to per‑region usage metrics, coupled with content‑delivery networks (CDNs) that cache static assets at the edge, will smooth load on origin servers. However, each additional layer introduces latency and potential points of failure; careful monitoring is essential.
---
---
Review Note:
---
**
---
Sources: