Customer-support tools moved AI from assistant to operator this week, and started metering it.

The week in customer-support

The sector's AI work crossed a line this week: several vendors stopped shipping assistants that suggest and started shipping agents that act, then began attaching a meter to the compute behind them. Xurrent put Sera AI's first two autonomous agents into production, moving from a chat widget to software that closes requests without a person in the loop. Plain opened its Sidekick agent to any MCP server, so its reach is now defined by the customer's own systems rather than a connector roadmap. Respond.io pushed Copilot into every workspace, including an agent builder that drafts and test-runs an AI Agent from a description. The common move is to make the agent the default way to operate the product, not a bolt-on.

The commercial half of that bet showed up in the same window. Respond.io converted AI from a capped plan feature into a metered consumable with AI Credit On-Demand, and Formbricks provisioned a metered price for workflow runs on its Scale tier. Two vendors, in one week, deciding that automation volume — not seats or plan — is what they charge for. Underneath both threads sits a quieter discipline: grounding the AI in the customer's own data, from Copilot answering strictly out of the Help Center to ManageEngine ServiceDesk Plus letting Resolution Assist draft from similar resolved tickets rather than the model's general knowledge.

Leaders

Respond.io shipped the week's densest pair: Copilot now sits in every workspace as product Q&A, cross-module navigation and an AI Agent Builder, while AI Credit On-Demand starts billing usage past a plan's included credits automatically from the next cycle. Copilot's grounding rule — it says so when the docs do not cover a question — signals they expect it to carry real support load. The two releases are one decision from both ends: make AI the default, and make sure it cannot run out mid-workflow.

Twilio moved US Enhanced Branded Calling to general availability on T-Mobile and Verizon, attaching a business logo and stated call reason to outbound calls, and took its Salesforce Service Cloud Voice integration to GA on the AppExchange. The rest of the changelog is parity and plumbing — WhatsApp Card and Flows reaching Conversations, dedicated IPs for Twilio Email — but the branded-calling GA is the clearest expression of a trust layer being sold back on top of the network.

Plain connected Sidekick to any MCP server, the release that removes the ceiling from a month of autonomy work, then cleared the path in front of it with custom importers that pull conversation history from any tool without waiting for a built-in connector. Capability, then autonomy, then extensibility, then onboarding — Plain is declining to sit as the bottleneck between a customer's systems and its own product.

Formbricks provisioned a metered workflow_runs price on Scale subscriptions in 5.3.0-rc.2, its first billing-side evidence of a shift off plan-based pricing. The same RC tightened MCP OAuth challenge parity and added audit logging, continuing enterprise-tenancy hardening that has run line by line across API keys, surveys, quotas and now email recipients.

Xurrent shipped Sera AI's first two autonomous agents alongside knowledge-article readiness checks and readable CI reconciliation — the data-quality substrate an agent needs before it can be trusted to close a request against a CMDB. Building the substrate first is why the agent work landed without the usual rollback noise.

Wildcards

HelpCenter.io shipped FAQs.me, a standalone free FAQ builder that needs no account, carries no watermark, and emits one self-contained snippet with no scripts calling home. It is the vendor's first product that isn't HelpCenter.io — a free tool solving the entry-level problem while keeping analytics and AI deflection on the paid side, which reads as an acquisition funnel more than a side project.

Hatz AI shipped Activate, a rollout console that equips the reseller rather than the end user: it scans a client environment for shadow AI, refreshes use cases on an ongoing basis, and adds a Sales User role so partner staff get access without admin rights. Against a window otherwise made of model additions, Hatz is shipping the managed-service provider's selling motion, not another AI feature.

Themes that compounded

• Agents crossed from suggesting to acting, with MCP as the extensibility layer — visible in Plain, Xurrent, Respond.io and Frill.
• Metered, usage-based pricing for AI and automation arrived at Respond.io and Formbricks in the same week.
• Vendors grounded AI in the customer's own corpus — Help Center, ticket history, CMDB — at Respond.io, ManageEngine ServiceDesk Plus and Xurrent.
• The MSP channel deepened as a first-class surface at Hatz AI and INKY, from rollout consoles to Autotask ticketing.
• MCP kept spreading as table stakes, reaching general availability on every plan at Frill and OAuth parity at Formbricks.

Watch this week

The open question the data raises is governance, not capability. Respond.io and Formbricks both turned AI into a billable meter this week without shipping the per-team limits or consumption reporting an admin needs to control it — watch for usage-governance surfaces to follow the pricing. On the agent side, Xurrent's success-metrics tooling and Plain's importer groundwork both point to the same next step: proving the agents work and feeding them migrated history, rather than shipping more of them.