Live meeting AI arrived in one week. All of it is Enterprise-gated.

For about two years, "the AI tells you what to say while you are still talking" was a thing you could demo but mostly could not buy. That ended in a 48-hour window, and it ended twice.

We had just finished a competitive review of this category when both landed, which meant rewriting a good deal of it. The conclusion that survived is not the one we expected. The interesting question is no longer whether live assistance exists. It is what it costs, who is allowed to have it, and where the words on your screen actually come from.

What actually shipped

Otter Live Assist, announced 21 July 2026, is positioned by Otter as "the first real-time AI coaching agent that joins any call and delivers glanceable, in-the-moment guidance." It works across Zoom, Google Meet, and Microsoft, or through Otter's desktop app. It is available on Otter Enterprise.

Zoom Revenue Accelerator Sales Assist, announced 22 July 2026, surfaces objection guidance, discovery prompts, battlecards, and competitive intelligence during a live call. It arrives in two new tiers: Essentials from $66 per user per month and Premium from $99.99 per user per month, billed annually. Essentials is dated for August, so the tier that carries Sales Assist is announced rather than purchasable as of this writing.

For completeness, the rest of the tier had already crept toward this. Avoma's Real-time Answer Assistant sits inside a Conversation Intelligence add-on that lands around $53 to $74 per user per month all-in. Fathom's desktop app shows live summaries and lists Real-Time Coaching as "Coming soon." Read AI's live dashboard is behavioural -- sentiment, engagement, talk time -- rather than tactical.

The part worth arguing about

Once several products do the same job, the feature name stops carrying information. What separates these is where the guidance comes from, and there are now three genuinely different answers.

Zoom pre-configures it. This is the best-documented of the three. Per Zoom's own support documentation, an administrator enters questions and guidance per topic category -- discovery, objection handling, pricing, competitor talking points -- attaches knowledge collections, and rule-based signals plus timer reminders drive what appears. It is a well-built system for showing a rep the card that someone already wrote.

Otter grounds it in your material. Otter's Agent Builder lets you point Live Assist at playbooks, SOPs, past meetings, and internal resources, then surfaces tip cards visible only to you and checks objectives off as the call proceeds. That is a meaningfully different design from a static card, and closer to retrieval over content you supply.

Backchannel generates it from the conversation. Our objection handler runs roughly every ten seconds over the trailing ninety seconds of transcript. It emits a response you could say in the next ten seconds, paired with the underlying concern and the strategic angle behind it. There is no card library, because there are no cards. The text is written against the objection that was actually raised.

None of those three is automatically better. Pre-authored guidance is consistent, reviewable, and lets a sales leader enforce a message -- which is exactly what some organizations want and what a generated response cannot promise. Generated guidance handles the objection nobody wrote a card for, and says something specific to the sentence just spoken. Those are different products for different problems, and the honest version of this comparison says so.

What we are not claiming

We are not going to characterize Clari Copilot's internals. Their current documentation confirms live cues and long-monologue alerts and is silent on the matching mechanism. The widely-repeated description of how it selects a card traces back to community posts from 2023 that we could not corroborate, so we dropped it rather than repeat it. If you want a concrete example of the pre-authored pattern, use Zoom, which documents it publicly and is current.

We are also not claiming live assistance is novel any more. It was a reasonable thing to say a month ago. It is not one now, and a comparison page that still says so is out of date.

So what is left

Price and permission. Otter's is Enterprise. Zoom's starts at $66 per user per month and only inside Zoom. Avoma's needs an add-on on top of a base seat. Every one of them processes the call in the vendor's cloud, and every one of them is a per-seat line item that someone has to approve.

Backchannel is the same category of feature with the opposite deployment. It is MIT-licensed, it runs on hardware you control, and it costs nothing per seat. Point the agents at a self-hosted OpenAI-compatible endpoint such as Ollama or LM Studio, pair it with local ONNX transcription, and the whole pipeline runs without a key from anyone.

The honest trade, since this piece has been demanding honesty of everyone else: we have no compliance certifications, no CRM sync, and our setup is a Docker Compose stack rather than a button. If you want managed conversation intelligence with a support contract, the vendors above are selling exactly that and you should buy it from them.

If what you actually wanted was the guidance, without the seat licence and without the recording leaving your network, that is the thing we built.