AI Is Changing Who Builds Software
If you run a Malaysian SME, artificial intelligence may soon help create internal tools, automate reports, connect business systems and improve customer service without requiring every task to begin with a developer writing code from scratch. That sounds like a major advantage—and it can be—but the most important lesson from a recent VentureBeat analysis is that generating software is becoming easier while controlling its business impact remains difficult.
AI coding agents can inspect files, suggest changes, create tests and work through technical problems. The bigger question is whether the instructions, data and approval rules around them are strong enough to stop a confident-looking mistake. For your business, that distinction matters. A system that produces an incorrect quotation, duplicates an order, exposes customer information or reports the wrong sales figure can create operational problems even when its code appears technically sound.
What Happened
The source article describes a shift in software engineering. Tools such as Cursor, Claude Code and other agentic workflows can now help produce an initial implementation for complex integrations and data pipelines, rather than leaving engineers to write every line manually. AI agents can navigate repositories, inspect errors, propose refactors and respond to test results, according to the article.
However, an agent does not automatically understand the history and meaning behind a business process. It may follow an outdated assumption, fix a visible symptom instead of the underlying issue, or use a field that appears suitable but represents something completely different. The article calls this buildup of stale assumptions and unresolved dependencies “operational entropy”, explaining that reliable feedback—such as tests, precise contracts and deterministic tools—helps keep an agent on course.
A simple example shows the risk. An agent asked to add a customer_tier field might use an existing status field because the names look related. The result may pass technical checks while remaining wrong for the business. A proper data contract would specify how the tier is calculated, who owns the definition and which source is allowed. In that situation, the engineer’s most valuable contribution is not typing the transformation. It is creating the boundary that makes an incorrect transformation fail visibly.
Why This Matters for Malaysian SMEs
Many Malaysian SMEs are now connecting more systems than before: accounting software, point-of-sale platforms, online marketplaces, payment gateways, customer relationship tools, inventory applications and messaging channels. A restaurant group may link orders to kitchen operations and stock records. A wholesaler may connect sales orders to warehouse quantities and delivery updates. A service company may combine enquiries from WhatsApp, Facebook and its website into one follow-up process.
AI can help configure these connections, generate scripts and summarise operational data. Yet the same customer or product may be represented differently across systems. A customer marked “active” in one platform may simply mean that the account exists, while another system may define “active” as having purchased within a certain period. If an AI agent joins these records using assumptions rather than agreed definitions, your dashboard may look polished while giving you the wrong answer.
Malaysian business conditions also make boundaries particularly useful. Your processes may involve multilingual customer messages, sales tax settings, local delivery zones, bank reconciliation, supplier terms and different approval levels across branches. Rules may be spread across spreadsheets, staff habits and old software settings. An agent cannot safely infer all of this from a short instruction. You need to make the important rules explicit before automation is allowed to act.
| Business area | Useful AI role | Boundary you should define |
|---|---|---|
| Sales enquiries | Draft replies and identify follow-up tasks | Human approval before unusual discounts or commitments |
| Inventory | Flag low-stock items and suggest purchase orders | Do not change stock balances without a recorded transaction |
| Accounting data | Classify documents and detect missing information | Restrict posting rights and require review for exceptions |
| Customer records | Find duplicates and summarise interactions | Use approved fields and protect personal information |
Start With Business Contracts, Not Prompts
A prompt tells an AI agent what you want it to attempt. A business contract tells it what must be true before the result is accepted. For example, instead of saying “update the customer category”, define the approved categories, the source of truth, the person responsible for the definition, the conditions for changing it and the action to take when information is missing.
You do not need a large technical department to begin. Choose one recurring workflow and document its boundaries in plain language. Identify the input, the expected output, the systems involved, the decisions that require approval and the records that must never be overwritten. This can cover a lead assignment process, supplier invoice capture, appointment reminders or stock replenishment.
AI becomes more dependable when your business makes its definitions, permissions and failure conditions clear.
Next, separate suggestions from actions. An AI assistant can recommend a reorder quantity, draft a customer response or highlight an unusual transaction. It should not automatically approve every recommendation. Use approval gates for actions that affect customer commitments, inventory, accounting records, staff access or confidential information.
Build Feedback Loops That People Can Understand
An AI workflow needs more than a successful completion message. It should show what data it used, which rule it applied and why it made a recommendation. When something goes wrong, the person reviewing it should be able to identify the cause without searching through an unclear chain of automated decisions.
For a Malaysian SME, useful feedback may be simple: a daily exception list, a duplicate-customer report, a failed integration alert or a review queue for messages that contain uncertain delivery dates. Keep a record of corrections. If staff repeatedly change the same type of AI output, that pattern shows where the workflow or business rule needs improvement.
It is also important to test with realistic cases rather than only clean examples. Include cancelled orders, partial payments, returned goods, duplicate names, incomplete addresses, late supplier updates and products with similar descriptions. These situations reveal whether the automation understands your operating boundaries or merely performs well in ideal conditions.
The Bigger Picture
The article’s wider argument is that software engineering is moving from writing every piece of local logic towards designing systems in which generated logic can be trusted. That does not make human expertise less important. It makes system design, process knowledge and clear decision-making more important.
For you as an SME owner, this means automation should not be measured only by how quickly it produces an output. The better questions are: Can you verify the result? Can you reverse an incorrect action? Is responsibility clear? Does the workflow protect customer and business data? Can a staff member understand what happened when an exception occurs?
The practical path is to start small. Choose a process with a clear outcome, create a written definition of success, limit the agent’s access, add approval points and review the results regularly. Once the workflow is stable, you can expand its role. The aim is not to let AI operate without limits. The aim is to give it a well-defined operating area where it can move quickly without crossing boundaries that matter to your business.
AI may increasingly write the implementation. Your business still needs to decide what is acceptable, what requires human judgement and what must never happen automatically. Those boundaries are the foundation that turns AI from an impressive demo into dependable business automation.
Ready to Streamline Your Operations?
Technology moves fast. Your operations should keep up. AutoRunBiz builds AI systems that run your daily workflows — from WhatsApp order capture to accounting. Book a free 15-min ops audit →