Field notes
from real delivery.▍

Vibe coding only took off last year, and the halo barely lasted twelve months. Over the past six months the market's mood has shifted from weekend-app wonder to a harder reality: demos everywhere, systems that survive go-live almost nowhere. This piece breaks down the eight walls between demo and production, why code that "looks right" is the most dangerous kind, and how professional teams using the same AI ship systems that run daily operations — including how we replaced Swingvy in four weeks, full video included.

AI makes "just build it ourselves" look trivial: two internal engineers, Cursor plus Claude Code, a working demo by week 4. But enterprises don't need demos — they need systems that employees still want to use 18 months later, that audits clear, that don't blow up at the next compliance check. This essay walks the timeline of the DIY-with-AI path — what it looks like at week 4, month 6, month 12, and month 18 — and why the gap between expert and non-expert AI use is the 5–10× output multiplier that decides which path you actually walk.

We haven't shipped Agentforce for a client yet — but we've spent 18 months tracking it. This post compiles failure modes from Western early adopters, Salesforce's platform evolution from Agent Builder to Testing Center to Agentforce Script, and a decision framework with code samples for enterprises preparing to launch in 2026.

We turned the knife on ourselves — replacing the external SaaS we had been using with our own EKel Finance Cloud, rebuilt via Agentic Coding. A traditional estimate would have been 4–6 months; we shipped Web, iOS, and Android in four weeks. This piece breaks down how humans and AI divide labor at every engineering stage, with the pitfalls we hit and a workflow you can take home.

Our financial-services delivery experience comes from Australia — our CTO led FSC implementations at two Australian Tier 1 banks and one mid-sized bank. This article maps that experience onto Taiwan's regulations, core systems, and budget structures, giving decision-makers about to kick off a project a frank, vendor-spin-free basis for judgement.

EKel's difference is not that we use one more development tool. We run Salesforce implementations through our self-developed AI Agent platform: from initial requirements and solution design to configuration, development, testing, UAT, and go-live. AI Agents assist — and at selected steps lead — the work, increasing efficiency and quality while lowering cost, so clients receive more competitive pricing.

McKinsey and BCG both put enterprise digital-transformation failure at around 70%. The problem isn't the technology — it's organisational self-deception. This post distils five lessons our CTO learned the hard way across multiple industries in Australia, plus a 5-question maturity self-check that tells you in 30 seconds how far you are from succeeding.