AI News for Business Owners: WhatsApp Replies Get Billed From 1 October


Monday morning at a building-materials distributor in Bekasi, around Rp 800 million a month, four staff sharing one WhatsApp Business number. The owner opens the app to clear the weekend backlog and finds most chats already answered overnight. Product questions handled. A price list sent. One appointment booked. Nobody worked late: someone on the team tapped an approval prompt that appeared in the app in August, without telling them. The replies are polite and mostly correct. One quotes a discount that no longer exists.
AI agent platforms for business changed shape in mid-August 2026. They are no longer a product you shop for and buy; they arrive as a feature inside subscriptions you already pay for — WhatsApp Business, Google Workspace, Microsoft 365. So the first move is not a vendor shortlist. It is an audit: what is already switched on in your accounts, who can switch it off, and what the bill looks like when usage climbs.
Meta launched Meta Business Agent for every business in Indonesia on 11 August 2026, announced at the fourth WhatsApp Business Summit Indonesia in Jakarta, and it works inside WhatsApp conversations (Source: ANTARA News, 2026). When the feature went global on 3 June 2026, Meta said more than 1 million businesses were already using it, free to activate, with paid tiers to follow (Source: Meta Newsroom, 2026). At that summit Meta also presented figures showing 71 percent of Indonesians message a business every day and 83 percent feel an AI agent gets them answers faster (Source: ANTARA News, 2026).
The brakes shipped alongside the accelerator. Google announced agent security controls for Workspace on 17 August 2026: agent identities on least privilege, an admin dashboard to suspend a flow or revoke its OAuth scopes, human approval steps, and DLP limits on what an agent may touch in Drive (Source: Google Workspace Updates, 2026). Ten days later came Skills in Workspace Studio, no-code automations for recurring work called with an @mention from Gmail, Docs or Chat (Source: Google Workspace, 2026). Microsoft extended its Planner Agent to all group-based Planner plans for Copilot-licensed users on 11 August, and still requires admin review before a self-built agent can reach an organization's Agent Store (Source: Microsoft Learn, 2026).
Notice the pattern. Admin approval and a kill switch are not compliance decoration; they exist because an agent acting inside your accounts is a real risk surface. You keep control only if someone holds that switch.
This month's news mixes two different things: what you can use this week, and what you cannot buy from Indonesia. Anthropic released an open-source commerce-agent blueprint on 2 September 2026 — a shopping agent and a merchant agent, with reference implementations for retail, travel, telecom and ticketing (Source: Anthropic, 2026). What shipped is code, and to be useful it has to sit on the Claude API, Amazon Bedrock or Google Cloud Vertex AI and be wired into your catalog, prices and order system by an engineering team.
The success numbers need careful reading. At the summit Meta showed its own customers' results: creative agency Mastung reported sales up 65.4 percent after activating the agent, while beauty brand Queensi reported up to 250 percent more customer conversations, and upselling to a higher package reached 60 percent of its customers (Source: ANTARA News, 2026). Anthropic claims carts up to 35 percent larger and shoppers 60 percent more likely to complete a purchase (Source: Anthropic, 2026). All of them come from the seller, with no comparison period and no published methodology. That does not make them wrong. It makes them a poor basis for your projection.
The gap between the news and the reality is measurable. Some 40 percent of Indonesian companies have adopted AI, up from 25 percent a year earlier, yet only 12 percent describe their implementation as fully integrated and 56 percent of adopters are still exploring (Source: AWS & Strand Partners, 2026). The button exists in plenty of accounts. What is missing is the process that should run on it.
In our experience at Agentiq, what is usually missing is not model intelligence but raw material. When a client switches on a platform's built-in agent, it knows only the conversation: not warehouse stock, not the special price agreed with a customer, not the payment status sitting in a finance spreadsheet. The answers sound confident and go wrong on the two most expensive things, price and availability. Exactly like that expired discount in the Monday morning chat.
Start with cost, which is never the tidy "from Rp X a month" of a vendor comparison. Microsoft prices its agent add-ons per user per month, in dollars, even on the Indonesian storefront: USD 18.00 per user per month for Microsoft 365 Copilot Business on annual commitment, USD 30.00 for Microsoft 365 Copilot, and a separate USD 200.00 a month for a 25,000-credit pack if you build your own agents in Copilot Studio (Source: Microsoft, 2026). Against a local yardstick: the 2026 Jakarta minimum wage is Rp 5,729,876 a month (Source: Pemprov DKI Jakarta, 2025). For a 20-person office that is a permanent cost line in a currency you do not earn.
Billing has two shapes. Two models can run at once: a fixed per-seat cost on the office side, and a variable per-conversation cost on the WhatsApp side, whose billing scheme changes from 1 October 2026. Data is the third piece. Once a platform agent reads your customer conversations and acts on them, your company remains the data controller under Indonesia's Personal Data Protection Law (UU PDP) and stays accountable for that processing, whoever owns the agent.
One thing has stopped being a reason to buy: Indonesian-language capability. When the big platforms give it away, what is left is your data, your process, and your rules.
Before you draw up a vendor list, do these four things. All of them fit in one working week.
If you want to know which agent features are worth switching on in your business, and which parts belong on your own data, tell us about your process through our contact page. The Agentiq team will answer with concrete recommendations, not a template.



Tell us what you need, and we will help turn it into a real application.