I met a fascinating character at an AI Tinkerers meetup in Hong Kong last week. The managing director of a 70-person firm that develops and manufactures consumer electronics products. He’s replacing his entire SaaS stack (incl. ERP, HRMS, CRM, helpdesk, and soon e-commerce) using agentic AI and a system built on OpenClaw, three Claude Max subscriptions, and one Codex subscription. His system consumes 250 million tokens per day. He never reads the generated code.

He’s cutting tens of thousands of dollars a month by eliminating SaaS licensing and transaction fees – a welcome boost to his company bottom line. But the real payoff is speed. In a hardware market where rivals can reverse‑engineer a new design within weeks, time is everything. Every system he builds mirrors how his company works, instead of catering to thousands of customers the way SaaS platforms must. As a result, his 70‑person team moves through these custom tools with far less friction and far more speed than they ever could in off‑the‑shelf software.

Here’s how he built his HRMS. He asked his HR manager to sit for a two‑hour interview with Claude and simply describe how her world works: attendance, leave, onboarding, reporting… Claude absorbed the entire workflow, distilled it into structured requirements, and broke those into build tasks organized on a Kanban board for Codex. Forty‑eight hours later, he had a working HRMS. That’s faster than most firms can assemble a vendor shortlist, never mind get a system deployed.

When Code Gets Cheap, Judgment Gets Exposed

The common conclusion from this is that software will be fully democratized, enterprises will build their own tools, and SaaS faces structural disruption. I think that conclusion confuses two very different capabilities: building software and conceiving software worth building.

Building software is getting cheaper at a shocking rate. That’s a given. Knowing what to build, where to standardize, where to customize, and how to change work around it is still difficult. The managing director I met in Hong Kong is unusual for a reason. He can see his company as a set of workflows, decisions, handoffs, and failure points that can be redesigned quickly. I call this ‘business-as-software.’ It is systems thinking applied directly to how a company operates. To be blunt, few organizations have a clean enough view of how work gets done today to do the same.

I think three patterns matter here:

  1. Business process design and software delivery collapse into one loop. You can already see it in smaller firms and AI-native organizations, where speed matters more than elegance and teams are willing to work through operational ambiguity. They capture the process, generate the application, test it in the flow of work, and refine as more data comes back. The speed is real, but so is the debt: code that no one wrote by hand is code that no one fully understands, and maintaining what you don’t understand is a cost that compounds silently. This is a different operating model: one in which a process owner plays a direct role in shaping systems, data, and workflows. Scaling it demands the right mental model for agentic AI: treating each agent as both a skill and a product, with the governance foundations to match.
  2. Larger enterprises will struggle to implement this operating model. They lack workflow clarity, governance, and leadership confidence. A bank in Singapore, or a public sector agency in Australia may have access to the same models and tools, but in most cases, the organization is still trying to turn fragmented process knowledge into usable software without creating control gaps, security exposure, or another layer of technical debt. Process intelligence platforms such as Celonis and Signavio address part of this problem by giving firms a clearer view of how work actually flows across the organization. But the governance architecture for AI-generated applications entering production, including who owns it, who controls it, who is accountable when it fails, is a gap most CIOs have not yet closed.
  3. SaaS vendors face uneven pressure. Systems that mainly package straightforward workflow and light configuration are most exposed. If someone who understands the problem well enough to describe it clearly can replicate 80% of the value in a weekend, the SaaS subscription is hard to justify. Systems encoding deep domain judgment – the compliance logic refined across thousands of implementations and shaped by regulatory frameworks – still have room to defend their business. Workday’s value isn’t the interface. It’s the HR compliance logic that satisfies auditors across 30 jurisdictions. That expertise cannot be replicated in a weekend, regardless of how cheap code becomes. But features alone will not hold. That game is getting cheaper by the quarter.

What it means for CIOs

The Hong Kong MD is an early adopter. When the economics shift this fast, edge cases are important signals that reveal where the market is heading long before most enterprise leaders are willing to acknowledge it. Software economics are evolving far faster than enterprise governance models can keep up. The gap between what can be built and what organizations can safely absorb is where CIOs should focus now:

  • Audit your application portfolio. Flag every application where the primary value is workflow mirroring and the differentiating value is thin: little domain compliance logic, low switching cost, no network effect. These are the candidates for AI-assisted build, extension, or replacement. Forrester’s REAP framework can help structure this assessment.
  • Close the governance gap before the build wave arrives. Who in your company can actually articulate a process well enough to rebuild it? Who can decide what must be standardized and where variation makes sense? Who owns the control points as AI generated applications begin entering production? This is where the gap will widen between firms that use these tools effectively and firms that accumulate pilots with no operational impact.
  • Change how you evaluate vendors. Probe for domain expertise, regulatory coverage, ecosystem value. If the answer you get sounds like another feature roadmap, pay attention – you are paying for convenience that is about to get very cheap.