Support tooling crossed from AI that drafts to AI that acts, with Plain, Thread, and Hatz pushing agents into production.
The week in customer-support
The clearest move this week was support AI shifting from advisory to operational. Plain made it explicit: its Sidekick assistant, until now a drafter and summarizer, can now take actions across connected tools and inside Plain itself. That is the harder, higher-trust step, and several other vendors leaned the same direction in the same window. Thread turned on Voice AI outbound calling for MSP techs, ManageEngine ServiceDesk Plus promoted its Zia-powered Workflow Assist to GA as a workflow author rather than a chat helper, and Hatz AI paired a new AI output surface with a full audit trail. The pattern is less about new chat boxes and more about letting the AI investigate, execute, and be measured.
The second, quieter story is the plumbing that has to exist before any of that autonomy is trustworthy: identity, billing, audit, and contact-resolution. Twilio shipped standards-based organization identity, Respond.io absorbed WhatsApp's phone-free contact model, and LiveAgent kept building metered AI-credit provisioning under a heavy maintenance cadence. None of those are AI features on their own, but they are the rails the agentic layer runs on. The sector is building the guardrails and the meters at roughly the same speed it is building the agents.
Leaders
Plain is the sharpest signal of the week. Sidekick can now execute actions across connected tools and inside Plain on the agent's behalf, not just draft or summarize — the pivot point of the company's agentic story. Two supporting improvements reinforce it: Sidekick auto-starts on threads that match workflow rules, so investigation and a draft reply are waiting when a human opens the thread, and workflows can now fire on a recurring schedule with no triggering thread at all. The direction is proactive, workflow-driven support with the human moved to review.
Thread extended its MSP service desk from text into live voice. Voice AI outbound calling went live: technicians click-to-call from the Inbox, and every call is recorded, transcribed, and dropped onto the ticket with an AI summary, while the customer always sees the verified main number regardless of device. Paired with new mid-call controls (pause transcription, force-end stuck calls) and a six-dashboard Magic Analytics suite, the work is about making the agents production-safe and measurable, not just live.
Hatz AI advanced its MSP platform on two fronts that fit its governance thesis. Artifacts lets users ask the assistant to build a report, deck, tracker, or web page and refine it in-conversation; Compliance & Logs records every AI invocation — user, tenant, model, content — as a filterable, exportable audit trail. For a multi-tenant gateway selling to regulated MSPs, an exportable record of what every tenant's AI did is the feature that closes deals, and the routine model refresh (Fable 5, GPT-5.6) keeps the catalog current.
ManageEngine ServiceDesk Plus made Workflow Assist generally available across every edition and taught Zia, Zoho's hosted LLM, to generate workflows from attached images and write executive summaries of existing ones. Admins can go from a screenshot to a drafted workflow and understand a workflow's governance without reading its full config. It moves Zia from conversational add-on to builder — the most directional AI step in ITSM this stretch, atop the usual change-management refinements.
Respond.io made the deepest architectural change of the set, absorbing WhatsApp usernames and Business-Scoped User IDs so contacts can message a business without ever sharing a phone number. The platform auto-identifies those contacts by BSUID and threads the identifier through its webhooks and Developer API so integrations keep syncing. Its AI Agent also gained inline file attachments — incremental, but consistent with pushing more front-line handling onto the agent.
Wildcards
Twilio is off the agentic pattern entirely: its window is platform-maturity work for large, regulated buyers. OAuth 2.0 client credentials for the Organization APIs hit GA alongside public Roles and Role Assignments APIs and Entra SCIM, making account provisioning programmable and standards-based. Add RCS error-code hygiene, WhatsApp username parity, and non-US Branded Calling beta, and it reads as steady hardening, not a directional AI bet — the identity layer everyone else's agents will eventually authenticate against.
Richpanel posted no spark but a dense integration cadence: RingCentral voice landed in the inbox with call recordings and optional AI transcript/summary, alongside the AfterShip tracking-returns-warranty trio and SellerCloud. The bet is breadth — become the single console where an agent acts on every downstream ecommerce system without tab-switching — with the AfterShip Tracking AI-reply hook hinting at where the AI agent goes next. Coverage over depth, for now.
Themes that compounded
- AI crossed from drafting to acting: Plain's Sidekick actions, Thread's live voice calls, and ManageEngine's workflow-authoring Zia all move the assistant from suggesting to executing.
- Production-hardening trailed the features: audit logs (Hatz), mid-call controls and analytics (Thread), and approval-shaped workflow triggers (Plain) show vendors making autonomy safe to trust.
- Contact identity is migrating off phone numbers, with Respond.io adopting WhatsApp BSUIDs and Twilio surfacing the same usernames across its messaging APIs.
- Commercial and identity plumbing kept pace with AI features, from LiveAgent's metered AI-credit pools to Twilio's programmable org identity.
- MSP-focused desks (Thread, Hatz) are converging on an agent-plus-voice-plus-governance operating model rather than single headline features.
Watch this week
The products that shipped agent actions this week — Plain, Thread, ManageEngine — are the ones to watch for the guardrails that have to follow: approval controls, permission scoping, and audit surfaces that make execution reversible. Hatz already paired its output surface with logging, and that ordering (capability, then provable oversight) is the tell for who is selling into regulated buyers. On the identity side, watch whether WhatsApp BSUID handling spreads past Respond.io and Twilio into automation and reporting surfaces, since phone-free contacts break any workflow keyed on a number. LiveAgent's metered AI-credit work is worth tracking as the quiet indicator of when AI-agent usage turns customer-facing and billable.