LoginSign Up
← Back to Support
Weekly · Support · Week of August 3, 2026

Support tools stop deflecting and start acting: agents gain skills, MCP reach, revenue context and a voice line

agentic-supportmcprevenue-attributionvoice-aicontact-identityfeedback-automation
O
Generated 1h agoDrawn from 8 products

The week in customer-support

The clearest move this week came from Plain, which finished converting Sidekick from an in-thread helper into something that runs work on its own. Two sparks landed close together: Sidekick can now be given reusable skills authored once and shared across a workspace, and it can connect to any MCP server rather than only Plain's built-in integrations. Together they change where the ceiling sits. What Sidekick can do stops being set by Plain's integration roadmap and starts being set by what each customer wires up and writes down. The sequence Plain has followed — capability, then autonomy, then extensibility — is the same arc showing up across the sector this week, just further along.

The rest of the leaderboard is running the same play from different starting points. Twilio took US Enhanced Branded Calling to general availability on T-Mobile and Verizon, an answer-rate argument rather than a developer feature. Canny rebuilt its reporting around money, attaching renewal risk, pipeline and churned revenue to individual feature requests. Respond.io kept fitting its AI Agent with the behaviors a human agent has while loosening the phone number as a contact's identity. And Thread brought outbound calling into the ticket for MSP technicians. Feedback capture, trust signaling, revenue attribution and voice all moved the same direction: less deflection, more action recorded where the work already lives.

Leaders

Plain had the strongest window, with two sparks. Workspace-shared skills let a team write up a routine once and have everyone invoke it, so the agent stops being configured per-session and starts accumulating institutional procedure. The second spark opened Sidekick to arbitrary MCP servers, so a customer can point it at whatever internal service exposes MCP. Workflows becoming the trigger layer for that agent work, on a schedule or on matching threads, is the tell that Plain is reshaping the support product around the agent rather than bolting the agent on.

Twilio is selling trust. Enhanced Branded Calling reaching general availability puts a business logo and a stated call reason on the recipient's handset before they decide to answer. The same window tightened sender vetting, with error code 30915 now catching misfiled sole-proprietor registrations across both A2P 10DLC and Toll-Free Verification, and published an unusually candid breaking change: Functions and Assets timestamps switch to true UTC on September 1 with six weeks of notice.

Canny is wiring product feedback to revenue. The rebuilt Reporting tab splits into Ideas, Autopilot and Portal, and the Ideas reports now carry renewal risk, open pipeline value and churned revenue per feature. Alongside it, Autopilot began pulling from Fathom and Grain call recordings and the MCP server passed 55 tools, readying the same data for agents and larger admin teams.

Respond.io is making its AI Agent the first responder rather than a deflection layer. The agent can now wait for a Contact to finish typing before replying, send files and images as rendered attachments, and recognize a reopened conversation. Underneath, WhatsApp usernames and Business-Scoped User IDs let Contacts message a business without sharing a phone number, with BSUIDs flowing through webhooks and the Developer API.

Thread is making the ticket, not the channel, where an MSP conversation lives. Outbound calling from Inbox means technicians place calls with recording, transcript and AI summary landing on the ticket automatically, and the transcript pushed to the PSA as an attachment. The surrounding work — Teams emoji reactions syncing back to Inbox, per-flow control over which updates reach Slack and Teams — is all about fidelity between where a conversation happens and where it is recorded.

Wildcards

Sleekplan spent 2026 on a rebuild rather than incremental releases. Sleekplan 2.0 entered beta with a rewritten admin app and an AI layer over the feedback loop, and the Impact Score was rebuilt the same day to be inspectable and configurable rather than an opaque number. Cadence is low and lumpy, so the rebuild is consuming the roadmap; the configurable Impact Score is the piece that has to be trusted before the AI layer can prioritize unattended.

Infobip publishes quarterly roll-ups, and its Q1 2026 summary puts AI at the foundation of the stack: AgentOS to unify the communication tooling, AI agents that take actions rather than draft replies, and MCP servers wiring models to live WhatsApp, RCS and voice channels. The repositioning is from delivering messages to running autonomous interactions on top of them, with channels becoming the substrate and the agent layer the product being sold.

Themes that compounded

  • Agents that act, not deflect. Plain, Respond.io, Hatz AI and Infobip all shipped work that moves the AI from drafting a reply to running the task, with human-agent behaviors and autonomy attached.
  • MCP as the extensibility layer. Plain opening Sidekick to any MCP server, Canny passing 55 MCP tools, and Infobip shipping MCP servers across channels all make their surface callable by models the customer already runs.
  • Feedback wired to revenue. Canny attaching renewal risk, pipeline and churn to individual features, and Sleekplan rebuilding a transparent Impact Score, both answer why a piece of feedback matters with a number.
  • Voice becomes a first-class surface. Thread putting outbound calling on the ticket and Twilio branding the inbound call are two ends of making voice something recorded and trusted rather than bolted on.
  • Identity decoupling from the phone number. Respond.io introduced Business-Scoped User IDs so Contacts can reach a business without a phone number, flowing through webhooks and the Developer API.

Watch this week

The open question this week is governance. Plain's Sidekick can now act across arbitrary MCP servers on a schedule, and Canny's MCP surface already covers create and delete paths — both raise approval and audit questions the current releases do not address, so the next visible move is likely permissioning and oversight around agents that already act. On the trust track, Twilio's branded calling is generally available on two US carriers, so the near-term step is more carriers or markets rather than new voice capability. And Sleekplan 2.0 is still in beta, with a general-availability push the most likely next release. Only this week's data supports these reads; watch whether the agent-autonomy work starts shipping the controls it currently lacks.