AI Assistants Are Moving From Chat to Action
You may already use AI to draft messages, summarise documents or create social media ideas. The harder question is what happens when an AI assistant can do more than suggest an answer. What if it can read your business applications, contact people, arrange tasks and act on your instructions?
That is the direction highlighted by Instinct, an AI assistant startup founded in 2025. The company reportedly raised US$250 million in a Series B round, bringing total funding to US$350 million and its valuation to US$2.5 billion, according to TechCrunch. Its assistant connects to apps and devices and lets users communicate through text and calls.
For you as a Malaysian SME owner, the important lesson is not the size of the funding. It is the change in expectations. AI tools are being designed to carry out connected tasks, not merely provide information. That creates useful opportunities, but also requires careful control over access, privacy and approvals.
TL;DR
AI assistants are moving towards handling multi-step work across your business applications.
Start with low-risk tasks, limit permissions and require human approval before an assistant sends messages, changes records or makes commitments.
What This Means
A traditional chatbot responds to a prompt. You ask it to write a customer reply, and it produces text for you to review. An AI agent or action-oriented assistant aims to go further. It may interpret your request, access connected systems, complete several steps and report back.
For example, “Help me prepare tomorrow’s delivery schedule” could involve checking orders, identifying delivery locations, grouping jobs, drafting driver instructions and placing the proposed schedule in your operations system. The assistant is not simply writing a paragraph. It is coordinating work across several tools.
Instinct reportedly allows users to connect apps and devices and communicate with the assistant through text and calls. Early examples described by its founder include planning travel, buying groceries, purchasing tickets and cancelling subscriptions, as reported by TechCrunch. These examples show the convenience of an assistant that can complete a sequence rather than answer one question.
However, the same capability introduces risk. An assistant that can access your calendar, customer records, email, messaging tools or business systems may have broad visibility and authority. The TechCrunch report also noted concerns about the permissions requested and the potential implications of the startup’s terms of use. That is a useful warning: convenience must be balanced with clear boundaries.
A simple comparison
| Type of tool | What it usually does | Business example | Control needed |
|---|---|---|---|
| Chat assistant | Generates or explains information | Draft a quotation follow-up | Human review before sending |
| Workflow automation | Runs predefined steps | Send a reminder when an invoice is overdue | Clear rules and exception handling |
| AI agent | Interprets a goal and coordinates actions | Review enquiries, classify them and assign follow-ups | Limited access, approval gates and activity logs |
How This Applies to Malaysian SMEs
In a Malaysian retail or e-commerce business, an AI assistant could help monitor customer enquiries from WhatsApp, email and website forms. It might identify whether someone is asking about product availability, delivery status or a return. It could prepare a suggested response, find the relevant order details and assign the case to the right staff member. You would still decide whether the assistant is allowed to send replies automatically, especially when the message involves refunds, complaints or unusual requests.
For a service business such as an agency, repair company, tuition centre or cleaning operator, the assistant could help turn enquiries into scheduled work. It might collect the customer’s preferred date, check staff availability, prepare a job brief and place a proposed appointment on the calendar. In Malaysia, you may also need to account for public holidays, different operating hours, language preferences and locations across states. These details should be captured in your rules before automation is allowed to act.
For wholesalers, distributors and small manufacturers, the most useful application may be internal coordination. An assistant could review incoming purchase orders, compare them with inventory records, flag items that need replenishment and prepare a delivery plan. It should not independently confirm stock, promise delivery dates or alter official records unless your team has defined the conditions clearly. A wrong automated commitment can create more work than the original manual task.
Professional firms can also benefit. An accounting practice, consultancy or recruitment company might use an assistant to organise documents, identify missing information and prepare follow-up lists. Sensitive files require extra care. You should understand where information is stored, who can access it, how long it is retained and whether it is used to train a service. Malaysia’s Personal Data Protection Department provides guidance connected to personal data protection obligations through its official portal at PDP.
For many SMEs, the best first step is not giving an AI assistant access to everything. It is connecting one controlled process. For example, begin with enquiry classification or meeting summaries. Once you can see how the tool behaves, review its accuracy and expand only when the controls are working.
The more actions an AI assistant can take, the more important it becomes to decide what it is not allowed to do.
Practical Takeaways
- Choose one repetitive workflow. Start with a task that consumes staff time but does not involve irreversible decisions.
- Separate reading from acting. Let the assistant review information before you allow it to edit records or send messages.
- Use approval gates. Require a person to approve refunds, quotations, contracts, staff instructions and customer commitments.
- Limit permissions. Connect only the applications required for the chosen workflow. Avoid granting full access by default.
- Keep an activity log. You should be able to see what the assistant read, changed, recommended or sent.
- Define escalation rules. Tell the assistant when to stop and hand a case to a person, such as an angry customer, uncertain payment or unusual order.
- Test with old cases. Use previous enquiries or completed jobs to check whether the assistant classifies and prepares work correctly.
- Review data handling. Read the provider’s privacy terms, retention policy and security information before connecting customer or employee data.
- Train your team. Staff should know that an AI-generated answer is a draft unless your business has approved automatic sending for that specific process.
A practical readiness checklist
| Question | If your answer is “yes” | If your answer is “no” |
|---|---|---|
| Is the process clearly documented? | Convert the steps into rules for testing. | Document the current process first. |
| Can a person review important actions? | Add an approval stage. | Do not automate the final action yet. |
| Can you trace what happened? | Keep logs and review exceptions. | Choose a tool with usable records. |
| Is the data access limited? | Connect only necessary systems. | Reduce permissions before testing. |
The Bigger Picture
The Instinct story suggests that investors and users are paying attention to assistants that can operate across several parts of daily life. The reported US$2.5 billion valuation and US$350 million in total funding are figures supplied in the TechCrunch report, not a guarantee that every AI assistant will deliver reliable business results. Still, the direction is clear: software providers are competing to become an active layer between you and your business systems.
That may change how SMEs think about productivity software. Instead of opening separate applications and moving information manually, you may increasingly describe an outcome and let connected tools prepare the work. The advantage will not come from using the most advanced assistant. It will come from having clean records, clear procedures and sensible approval rules.
There is also a management lesson. AI adoption is not only an IT decision. It affects customer service, operations, finance, human resources and compliance. You need someone responsible for reviewing permissions, monitoring mistakes and deciding when a process should remain human-led.
For now, approach agent-style AI with curiosity and discipline. Pick a narrow workflow, measure whether it reduces delays or repeated work, and expand gradually. If an assistant cannot explain what it did or cannot be stopped before a sensitive action, it is not ready for an important business process.
As a Malaysian SME owner, you do not need to follow every AI announcement. You do need to prepare your business for tools that can act on your behalf. The companies that benefit most will likely be those that combine automation with practical safeguards, keeping people responsible for decisions that affect customers, employees and the business itself.
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 →