Engineering insight
Practical thinking on software engineering, digital transformation, AI, data, cloud, integration, automation, and complex operations.
Published thinking
Teams sometimes choose a familiar integration pattern before agreeing on response time, ownership and failure handling. The result can be fast data delivery that the operation did not need, or slow data where speed matters.
Read insightAn event named after a database update says little about what happened in the business. Consumers then guess at meaning and timing.
Read insightA practical way to map system ownership, handoffs and failure routes before choosing an integration interface.
Read insightA focused 90-day transformation should clarify one workflow, prove one change and leave an accountable operating capability behind.
Read insightPredictive models cannot compensate for missing asset histories, inconsistent failure codes and maintenance work that is not recorded reliably.
Read insightThe value of supply-chain software appears when plans break. Make exceptions visible, prioritised and owned across the workflow.
Read insightTreat important datasets as products with users, owners, quality commitments and change practices when operations depend on them.
Read insightA recovery plan becomes credible only when teams restore real services, dependencies and decisions within agreed business objectives.
Read insightHardware trust depends on standards, provenance, deployment controls and end-of-life decisions across the full technology lifecycle.
Read insightNIST's 2026 guide frames ICT supplier due diligence around ownership, provenance, resilience, cyber practices and supply-chain tiers.
Read insightBefore replacing algorithms, organisations need to know where cryptography protects data, identities, software and long-lived business records.
Read insightShort architecture decision records preserve context, alternatives and consequences so future teams can change systems without guessing.
Read insightTechnical debt decisions improve when teams connect each constraint to operational impact, change frequency and risk instead of debating code quality in isolation.
Read insightEngineering metrics become useful when teams connect delivery speed and reliability to customer, operational and financial outcomes.
Read insightA useful golden path accelerates common delivery while making constraints visible and legitimate exceptions governable.
Read insightAI workloads add new platform consumers, cost patterns and governance needs. Extend proven platform principles instead of building a separate island.
Read insightInterfaces do not solve unclear ownership, inconsistent meaning or weak recovery. Reliable integration is an operating model as well as code.
Read insightEvent-driven architecture works when events represent durable business facts with clear ownership, ordering and recovery behaviour.
Read insightReliable customer, supplier and product data requires named business ownership, decision rights and correction workflows across systems.
Read insightData contracts make ownership, meaning, quality and change explicit between producing and consuming teams before failures reach operations.
Read insightA vendor API is only useful when its coverage, controls, limits and change practices match the workflows your organisation must operate.
Read insightChoose build, buy or integration by testing differentiation, workflow fit, ownership and change, not by comparing licence price with development cost.
Read insightClosing South Africa's digital skills gap requires stronger transitions from learning to accountable workplace delivery, not training volume alone.
Read insightAn AI inventory gives organisations visibility into tools, data and decisions. It is a stronger starting point than a policy teams quietly work around.
Read insightHuman-in-the-loop controls fail when reviewers lack time, context or authority. Design oversight as an operational role with clear decision rights.
Read insightAI workload cost should be visible at design time. Connect consumption, unit economics and business value before usage becomes difficult to control.
Read insightEnterprise AI evaluation should measure the quality of the operational decision, workflow controls and exception handling, not only model accuracy.
Read insightModel Context Protocol connections can simplify AI integration, but every exposed tool still needs production-grade identity, authorisation and monitoring.
Read insightUseful agent audit trails connect intent, evidence, tool calls, approvals and outcomes so teams can explain and reverse important actions.
Read insightAI coding tools can accelerate strong teams and expose weak delivery systems. The difference lies in review, testing, architecture and feedback.
Read insightA legacy system does not need one dramatic answer. Assess each workload and choose the least risky path that creates measurable business value.
Read insightInternal platforms create value when they remove proven delivery friction for real users, not when they merely assemble fashionable infrastructure tools.
Read insightObservability is useful when it helps teams answer business and technical questions quickly, not when it simply creates more telemetry to store.
Read insightAn SBOM is useful inventory, but dependable software supply chains also require protected builds, provenance, verification and response.
Read insightCloud region is only one part of a cross-border data decision. Map the processing chain, transfer basis, safeguards and operating evidence.
Read insightSecurity claims are difficult to compare. A structured procurement conversation can reveal how a supplier designs, operates and supports its software.
Read insightBackups are only one part of resilience. Recovery objectives, dependencies, tested restoration and clear decisions determine whether operations return.
Read insightA tender can be disqualified before anyone reads the proposal. A practical checklist for the CSD, SARS, and B-BBEE documents that most often cause it, and the habits that keep businesses ahead of expiry dates.
Read insightA practical way to govern AI decisions, data, vendors and accountability before experiments become business-critical systems.
Read insightCloud cost becomes manageable when engineering, finance and product teams connect spend to workload design, ownership and business value.
Read insightRunning five disconnected systems means someone becomes the manual bridge between them. What that actually costs a growing business, and what real system integration looks like in practice.
Read insightMost operations are data-rich and insight-poor. Useful analytics starts from the decision a team needs to make this week, not from what's easiest for a dashboard to display.
Read insightA manual stock count is accurate for a moment, and that moment has usually passed by the time anyone acts on it. Why periodic counting keeps drifting from reality, and what closes the gap for good.
Read insightEvery vendor claims to offer visibility. Real-time visibility means answering where an order or shipment is right now without phoning someone or checking three systems, and it starts with connected, current data, not a dashboard.
Read insightEvery vendor spreadsheet starts small and reasonable, then quietly becomes a supplier database nobody fully trusts. What breaks as the vendor list grows, and what good supplier management actually requires.
Read insightCloud migration is easy to keep postponing until a server failure forces the decision under pressure. What's actually driving businesses to move, and what to think through before migrating.
Read insightCustomer-facing products get real design attention. Internal tools employees use every working day routinely get none. That double standard is a genuine, recurring cost most businesses never measure.
Read insightLast-mile delivery carries challenges specific to South Africa, from address inconsistency to load-shedding-driven traffic unpredictability. A look at where the real bottlenecks form and what actually fixes them.
Read insightOff-the-shelf supply chain software fits fine at first. The real comparison isn't custom vs. a perfect off-the-shelf fit, it's custom vs. the accumulated cost of workarounds around a system that's close but not quite right.
Read insightA framework for deciding which operational tasks to automate, which decisions should remain human, and what a production workflow needs around the automation.
Read insightA grounded way to assess AI for procurement and supply-chain work, starting with the task, data, risk, controls, and human decision that surround it.
Read insightShare the question or operational constraint. It may inform a future article or a focused project conversation.
