AI Agents Are Becoming More Practical for Your Business
If you run a Malaysian SME, you may already be using AI for writing, customer replies, research, or administrative work. The harder question is whether AI can handle a complete business task without requiring you to check every small step.
That is where “agentic” AI comes in. Instead of giving you one answer at a time, an AI agent can work through a sequence: read information, make a decision, update a system, prepare a response, and ask for your approval when something is unclear. Anthropic’s latest Claude Fable 5.1 announcement shows that this type of work is becoming more focused on practical business use.
Anthropic says Fable 5.1 is typically around 25 percent cheaper than Fable 5 and up to 45 percent cheaper for complex agentic tasks, partly because of lower pricing for cached data that has already been processed and stored. Source: The Verge
TL;DR
Claude Fable 5.1 is designed to improve performance, reduce repeated processing, and make AI agents more useful for multi-step work.
For your SME, the main lesson is not to rush into a new model. Start by identifying one repeatable workflow, set clear approval rules, and test whether the AI can complete it reliably.
What This Means
A normal chatbot responds to a prompt. You ask it to draft a customer reply, summarise a document, or create a product description. An AI agent goes further by carrying out a sequence of related actions.
For example, an agent supporting a service business could read an incoming enquiry, identify the customer’s request, check a knowledge base, prepare a response, create a follow-up task, and route unusual cases to a staff member. The value comes from reducing the number of manual handovers between people and software.
Anthropic describes Fable 5.1 as stronger than its predecessor, with improvements in speed and token efficiency. An early user quoted by The Verge said the model was “fast, token-efficient, and crucially actually speaks like a normal person.” Source: The Verge
The announcement also highlights lower pricing for cached data. In simple terms, if the system has already processed a large set of instructions or reference material, it may not need to treat the same information as completely new each time. This matters for agents that repeatedly use your company policies, product catalogue, service procedures, or internal documents.
The practical benefit is not “AI that knows everything”. It is AI that follows your business process consistently, while knowing when to ask a human.
How This Applies to Malaysian SMEs
1. Customer service and WhatsApp enquiries
Many Malaysian SMEs receive customer enquiries through WhatsApp, Facebook, Instagram, websites, and phone calls. Staff may spend much of the day answering repeated questions about operating hours, delivery areas, appointment availability, product specifications, or required documents.
An AI agent could classify incoming enquiries, search approved information, draft a reply in English, Bahasa Malaysia, or a mixture that matches your customers, and flag cases that need personal attention. You should still require approval before the system confirms unusual requests, changes an order, or makes a commitment on behalf of your company.
This is especially useful when your business serves different customer groups. A clinic, tuition centre, renovation contractor, wholesaler, or online retailer can create separate response rules while keeping a human involved for sensitive matters.
2. Sales follow-up and quotation preparation
Sales opportunities often disappear because nobody follows up at the right time. An agent can read enquiry details, identify the customer’s requirements, prepare a checklist for a quotation, and create reminders for your sales team.
For a Malaysian distributor, the workflow might begin when a customer asks for several products. The agent can extract item names, quantities, delivery location, and requested timing. It can then prepare a draft quotation using your approved product information. Your staff member reviews availability and terms before sending anything.
For a contractor or service provider, the agent could turn an initial enquiry into a site-visit checklist. It may identify missing information, such as measurements, building type, location, or preferred completion date. This reduces back-and-forth communication and gives your sales team a more complete starting point.
3. Internal administration and document handling
Administrative work is another suitable starting point because it often follows clear rules. You could use an AI agent to organise supplier documents, summarise meeting notes, compare purchase requests with internal policies, or prepare a weekly list of outstanding tasks.
For example, a small trading company may receive invoices, delivery orders, and purchase confirmations in different formats. An agent can extract key fields and identify missing information for staff review. It should not automatically approve payments unless your controls are strong and the action is reversible.
Anthropic also says its new model has more precise safeguards and is less likely to block basic biology questions than the earlier version. However, the company states that certain cybersecurity activities, including penetration testing and exploit generation, may still be redirected to other models or restricted. Source: The Verge For an SME, this is a reminder that model capability and safety rules are separate considerations. You need both useful output and sensible controls.
Where to Start
Do not begin by trying to automate the whole company. Choose one workflow that is frequent, structured, and easy to measure. A good first process might involve customer enquiry sorting, document summarisation, follow-up reminders, or preparation of internal reports.
| Workflow | Suitable first action | Human approval needed? | Useful measure |
|---|---|---|---|
| Customer enquiries | Classify and draft replies | Yes, for unusual cases | Response time and correction rate |
| Sales follow-up | Create reminders and summaries | Yes, before commitments | Follow-up completion rate |
| Document handling | Extract fields and flag missing details | Yes, before approval | Data accuracy and processing time |
| Internal reporting | Summarise approved business data | Yes, before distribution | Report preparation time |
The figures in the table are workflow categories and recommended controls, not industry benchmarks. Your own baseline should come from observing the process for a normal working period before automation.
Practical Takeaways
- Choose one repeatable process. Avoid starting with a broad instruction such as “run my business better”. Select a task with a clear beginning, middle, and end.
- Write down your rules. The agent needs approved information about products, service areas, response tone, escalation conditions, and staff responsibilities.
- Keep approval points. Require a person to review sensitive customer replies, quotations, refunds, contracts, hiring decisions, and operational changes.
- Measure reliability. Track how often staff must correct the output, how long the task takes, and whether follow-ups are completed.
- Protect business information. Check where data is stored, who can access it, how long it is retained, and whether it can be used for model improvement.
- Test languages and tone. Ask your team to check Bahasa Malaysia, English, names, addresses, local terminology, and customer-facing politeness.
- Plan for failure. Every automated workflow should have a clear route to a human when the AI is uncertain or a connected system is unavailable.
Anthropic says its Enterprise Frontier Safeguards will store customer data on customer cloud servers and begin rolling out later in the year. Source: The Verge You should treat this as a vendor claim to evaluate, not as a substitute for your own data governance review.
The Bigger Picture
The long-term change is that AI tools are moving from “help me create something” towards “help me complete a controlled business process”. That creates a different responsibility for you as the owner. You are no longer evaluating only whether an answer sounds good. You also need to ask whether the agent followed the right procedure, used current information, protected customer data, and stopped at the correct point.
For a company with one to 50 employees, this can improve consistency without forcing every task through the owner or one experienced staff member. A well-designed workflow can help new employees follow standard procedures and give senior staff visibility into exceptions.
However, automation should not hide weak processes. If your product information is outdated, your approval rules are unclear, or your customer records are incomplete, an AI agent will simply process those problems faster. Clean up the workflow first, then automate the parts that are stable.
Claude Fable 5.1 may be one option for testing this approach, but the broader lesson applies to any AI platform. Start with a narrow business problem, use your own data carefully, retain human accountability, and expand only after the results are dependable.
Ready to Streamline Your Operations?
Your business should run itself. AutoRunBiz deploys AI agents to automate your daily operations — WhatsApp orders, invoicing, customer follow-ups, and accounting. Book a free 15-min ops audit to see where automation fits your business →