Engineering & Architecture: Build Decisions This Week

Engineering & Architecture: Build Decisions This Week – 2026‑09‑19

The tech ecosystem is moving at a frenetic pace, but the week’s headlines show that operational risk and AI‑driven delivery pipelines are still in the front line. The following takeaways distill what a CTO should focus on today, where trade‑offs loom, and three concrete build decisions to surface in your backlog.

---

1️⃣ Agentic Software Factories – Scale or Control?

OpenAI’s latest deep dive into its agentic software factory exposes a production model that treats the LLM as an orchestrator rather than a single‑purpose inference engine. The key architectural change is moving from batch micro‑services to a continuously‑running, token‑driven loop that can invoke external tools, maintain state, and spawn sub‑agents.

Benefits

| Benefit | Trade‑off |

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

| Rapid prototyping of internal bots (e.g., PR reviewers, code linting) | Persistent compute cost per token; higher latency for large models |

| Centralised logging via the factory’s tracing | Potential leakage of PII if tokens are not scrubbed before storage or transmission |

| Simplified CI‑pipeline integration – a single entry point triggers all downstream tool calls | Requires tight outbound‑traffic controls to satisfy data‑protection regimes |

For teams that have already hit the “build‑once, run everywhere” wall, this model is tempting. However, the cost of an unlimited token budget is non‑trivial in 2026, especially for SA developers who face higher per‑compute rates on public clouds. A pragmatic approach is to pilot the factory with a single high‑value internal workflow (e.g., automated code review) and measure both compute spend and compliance audit risk before expanding.

---

2️⃣ Power Resilience in South Africa – Avoid “Zero Security”

The MyBroadband report that Johannesburg’s City Power failed to contract any security patrols exposes a silent threat: physical infrastructure can become the single point of failure. For on‑prem or edge data centres, this translates into two hard truths:

• Power interruptions are not just downtime – they are a compliance risk. Many SA banks are still subject to POPIA Act 4 of 2013 and the LRA 66 of 1995, which mandate continuity plans for critical services.
• Cloud multi‑AZ replication does not obviate local outages. Even when services run on AWS or Azure, they rely on Johannesburg’s grid for network connectivity.

Actionable recommendation:

Build a hybrid redundancy layer that couples on‑prem edge nodes with an automated fail‑over to an international cloud region. Equip each node with UPS + diesel generators and schedule bi‑weekly security drills to mirror the “zero‑security” warning. The extra cost of two power feeds is outweighed by the ability to keep APIs online during a citywide outage, preserving customer trust and revenue streams.

---

3️⃣ MVNO Playbook – Capitec vs. FNB

TechCentral’s analysis that Capitec Connect and FNB Connect share a common MVNO strategy may appear academic, but it has concrete implications for any SA fintech looking to bundle mobile services.

Considerations

• Customer acquisition channels: Capitec targets retail‑banking customers; FNB leans on corporate accounts. If your product is B2C, the Capitec partnership offers a larger base.
• Regulatory oversight: Both banks operate under strict SA telecom licences, but their audit trails differ. Capitec’s lighter regulatory burden can speed time‑to‑market for new bundles.
• Technical integration: The MVNO stacks use identical API patterns (REST + WebSocket), but Capitec’s mobile core has lower latency in the Gauteng region.

Decision point:

If your goal is rapid, low‑risk market entry, a Capitec partnership may be preferable. If you anticipate high‑volume corporate contracts and can afford a longer ramp‑up, FNB Connect offers deeper integration potential.

---

4️⃣ What to Pay Attention To

• OpenAI agentic factories: Pilot in controlled environments; monitor token cost & compliance.
• Power infrastructure risks in SA: Implement hybrid power/geo‑redundancy now—don’t wait for the next outage.
• MVNO strategy alignment: Match your target market with the appropriate bank partner.

5️⃣ What to Ignore This Week

• Ryanair’s apology – a brand PR incident with minimal technical impact unless you run airline reservation systems.
• easyJet cabin crew strikes – operationally relevant only if you rely on flight data feeds; otherwise, schedule this for later review.

---

Review Note

The build decisions above presuppose that your team has access to cloud multi‑AZ capabilities and the financial bandwidth to invest in backup generators. I recommend confirming:

• The exact token pricing tier with your chosen LLM vendor for the agentic factory prototype.
• Regulatory guidance on data residency for SA edge nodes vs. international clouds, especially under POPIA.
• The technical readiness of your MVNO API adapters to switch between Capitec and FNB environments without code changes.

Your validation will shape the final prioritisation matrix.

---