When the agent moves into your sales and service desk

The announcements that landed today share one thread: the AI agent is being pushed out of the demo and into the parts of a company that face a paying customer, sales, service, and the video you send them. The more interesting question in each case is not what the agent can do, but what has to be true around it, the connections, the oversight, and the skill of the person signing off its work, before a business can let it near the customer at all.

Salesforce frames the agentic customer experience

Start with a definition, because the word is now attached to almost everything. An *agentic* system is software that takes an action on your behalf, drafts the reply, updates the record, books the follow-up, rather than simply showing you data and waiting for you to act. In customer experience (CX) that is a meaningful shift. A dashboard that tells a service rep a customer is unhappy is a reporting tool. Software that answers that customer directly is something else, and it carries the company's name when it does.

Salesforce has published a roundup positioning this move as the direction of travel for the field: "Salesforce's Agentic CX Vision: 8 Dreamforce Announcements CX Leaders Need to Act On" [1]. The framing is the point. When a vendor tells CX leaders there are announcements they "need to act on", it is signalling that agents in service and support are moving from optional to expected.

The trade-off a business should hold onto is oversight. An agent that can act is an agent that can act wrongly, and the cost of a wrong action lands on the customer before anyone internal sees it. Before you let software reply, refund, or close a case on its own, decide which of those actions it may take unattended and which always require a human. We wrote about where that line sits in decisions automation should never make. The announcement is a reason to open that question now, not after the agent is live.

Zoom and the rise of revenue orchestration

The second item is a category most buyers have not had to name yet. A *revenue orchestration platform* sits across the sales process and coordinates the activity in it, the calls, the follow-ups, the handoffs, so that a deal does not stall because someone forgot a step. It is the sales-side cousin of the service agent above.

Zoom reports that it has been placed in an analyst's evaluation of that category: "Zoom was named a Strong Performer in The Forrester Wave: Revenue Orchestration Platforms For B2B, Q3 2026" [2]. It is worth understanding what that phrase means before you weight it. A Forrester Wave is a structured comparison of vendors in a defined category, and a tier such as "Strong Performer" sits below the top "Leader" band. That is useful context, not a verdict for your business. An analyst scores a product against a general set of criteria; you are buying for a specific team with specific habits.

The practical lesson is that "recognised by an analyst" and "right for us" are different tests, and only you can run the second one. Recognition tells you a product is serious enough to be measured. It does not tell you whether your sales team will actually use the orchestration it offers, which is the only measure that changes revenue.

Gemini voiceovers arrive in Google Vids

The third item is smaller but telling. Google is bringing upgraded synthetic speech into its video tool: "Create more natural, expressive AI voiceovers in Google Vids with upgraded Gemini 3.8 Flash Lite TTS" [3]. Text-to-speech (TTS) turns a written script into a spoken voice, and "voiceover and avatar narration" means that the voice in an internal training clip or a customer explainer need not be a person reading into a microphone.

For a business, this quietly lowers the cost of producing video, which until now was the expensive medium. A product update, an onboarding walkthrough, or a sales explainer can be assembled from a script rather than a studio session. That is genuinely useful for teams that avoided video because of the effort.

The trade-off is disclosure and tone. A synthetic voice narrating a message to a customer is a choice about how the company presents itself, and some audiences care that a voice is generated. There is no single right answer, but there is a wrong one, which is deciding by accident. If you adopt generated narration, decide where you will say so and where the human voice still matters.

Zapier connects Gemini Enterprise to everything else

An assistant is only as useful as the data it can reach, and most company data does not live in one place. Zapier's guide to Gemini connectors speaks to exactly this: "Gemini is everywhere in Google apps: it can read your emails, pull in data from your calendar, and find that file you accidentally saved to your work Drive" [4]. The unstated second half is that most businesses run on far more than Google's own apps, and a *connector* is the bridge that lets the assistant reach the tools outside that walled garden.

This is the plumbing question underneath every agent announcement today. An agent that cannot see your CRM, your billing, or your support queue can only act on part of the picture, and acting on part of the picture is how an agent gives a confident, wrong answer. We have argued before that disconnected tools are a quiet tax on a business in why your tools do not talk to each other. The point stands with agents in the picture: an assistant's usefulness is set by what it can reach, and reach is a decision you make when you choose and wire up your tools.

GitHub on the skill the agent cannot replace

The last item is the human one, and it is the thread pulled tight. GitHub's advice to developers working alongside AI is to "direct AI agents, critically review their output, and keep technical judgment at the center of your workflow" [5]. Read it past the developer audience and it is a general rule for anyone adopting agents this week.

Each of the items above describes software that can now act. GitHub's point is that someone still has to direct it and judge the result, and that this judgement is the skill worth building rather than the one to hand over. An agent that drafts a customer reply, orchestrates a deal, or narrates a video has produced a first draft of a decision. The value you add is the review. We set out where that human line belongs in what AI should and should not do in your business, and today's announcements are a reminder that the line gets more important as the agent gets more capable, not less.

The common lesson across all five is simple to state and harder to practise. The agent is arriving in the places that face your customer. What determines whether that helps or harms you is not the agent, it is the oversight you keep, the fit you test for yourself, the connections you build, and the judgement you refuse to delegate.