The agent that talks is the easy part. What's hard, and what brands using WhatsApp as a revenue channel won't build in-house, is everything around it: orchestration, integration, governance, measurable outcomes, and custom solutions.
A Meta Business Agent is an AI agent that Meta runs inside WhatsApp for you.
A customer messages your number, Meta's own agent writes the reply, and Meta bills you per token.
It arrives with a pricing change. Meta is splitting non-template replies (the messages you send inside an open customer service window) into two paid buckets. One message only ever falls into one bucket, so there's no double charging.
| Meta Business Agent | Service message | |
|---|---|---|
| Who writes the reply | Meta's own AI agent | Your own AI, a third-party agent, or a human |
| How it's billed | Per token | Per message |
| Goes live | [1 Aug 2026] | [1 Oct 2026] |
| Benchmarked against | Token usage | Utility / auth template rates by market |
Source: Pricing Update for Business Agent
One detail worth knowing: this is per message, not per conversation. The old conversation-based window isn't coming back. A 24-hour customer service window still exists and resets each time a customer messages you.
Meta's headline figure is about $2 per 1 million tokens.
It estimates a typical message at roughly 24,500 tokens, which it puts at around 4 to 5 cents per message in the US. Cost scales with complexity: a one-line answer is cheap, a long multi-turn chat that solves a real problem costs more.
Two things matter for your budget.
First, there's no double-charging: each message lands in exactly one bucket, so a reply powered by Meta's agent is billed per token only, not also a per-message service or marketing fee.
Second, pricing now tracks what a reply does. Non-template replies inside the customer service window are free today; once these buckets go live they carry a cost, per token for Meta's agent or per message for a service reply, alongside the per-message template rates Meta has charged since 1 July 2025.
With cost tied to usage, the question for each conversation becomes what value it creates, and that's the question worth running the channel on.
Meta is positioning Meta Business Agent as the lower-friction, adoption-friendly path, a sign of how much it wants brands using it.
The number that matters for you is the value each conversation creates, and that's what charles is built to grow.
We've broken down the full pricing change and what it does to your cost per outcome in a separate post on the new WhatsApp pricing. Here, the point is what the agent adds, and what you put around it.
For a simple agent that answers common questions, yes. Meta offers two paths to a Business Agent, and the split tells you a lot.
Self-serve is a toggle-on agent you switch on in-app and configure with Meta's no-code builder (Weave).
It's built for any business that wants instant AI help, with minimal to moderate customisation.
The Enterprise API path is the one for mid-market and enterprise brands with custom needs, and Meta's own description of it is direct: direct API integration, engineering work done in-house or through a Business Solution Provider (BSP), full custom logic, data connectors and SLAs.
So even on Meta's own terms, running an agent at enterprise scale means engineering and a partner.
charles is a Meta Business Partner, so that's the work we already do. Meta will keep adding to what its agent can do on its own. What stays hard, and what Meta won't build for each brand, is running it as a coordinated layer: several agents working together, connected to your full tech stack, kept inside your compliance rules, measured against revenue, and working across every channel. That's the work, and it's the part enterprise and mid-market brands won't build in-house.
The agent is the small part. Four jobs around it decide whether agentic messaging makes money.
Real setups run several agents at once: one for product questions, one for order status, one for returns, plus scripted flows and human handover. Something has to route each conversation to the right place, decide when a person takes over, and now also decide which billing bucket each turn lands in: Meta's per-token agent, a per-message service reply, or your own agent. charles runs multiple agents under one head agent with handover built in, routes per turn, and keeps a per-user record across all three buckets. A single Meta agent doesn't do this.
A reply that knows the customer's last order, their size, what's in stock and where their parcel is comes from deep connections to your commerce, catalogue, order and CRM systems. A standalone agent has none of these by default. charles already connects them, so the agent answers with real account and order data rather than guesses. See how the charlesAI agent uses connected data.
For enterprise, and heavily in the EU and DACH, you need consent and opt-in management, brand-safety controls, and a full audit trail of what the agent said and why. A platform that records consent and keeps an audit trail makes that defensible. A raw agent on Meta's infrastructure leaves you to prove it after the fact. You also decide when Meta's agent handles a conversation and when to keep it out of sensitive or high-value turns; with a standalone agent, that control sits with Meta.
This is general information, not legal advice. Check your specific setup with your DPO or legal adviser.
Run WhatsApp as a performance channel and you tie conversations to revenue, ROAS and retention. A standalone agent answers and stops there, with no view of what a conversation was worth. charles reports messaging against revenue, so you can see which conversations and flows actually pay back.
Enterprise setups need tailoring to your stack, your catalogue and your service rules. charles runs an in-house AI team and an in-house creative agency, so we build the agents and the campaigns that run on them together with you, not just the technical wiring.
charles is the orchestration, integration, governance and measurement layer for agentic messaging.
Meta makes it easy to build conversational AI agents. charles connects it to your commerce and CRM stack, routes between agents, flows and humans, keeps consent and audit in order, and reports it all against revenue.
The pieces Meta is adding to its agent are ones charles already runs in production across many brands. As an official Meta Business Partner, with an in-house AI team and creative agency, we build the setup around your stack rather than handing you a bare agent.
charles will also support building and launching Meta Business Agent inside the platform, as one more agent running under your head agent. You get Meta's agent where its economics fit, with orchestration, integration, governance and measurement around it, in one place. For a brand that wants to run WhatsApp as a revenue channel with more than one agent, that's the full setup rather than a single bot on Meta's pipes.