AI meeting agents that act inside the call, not after it

Why vendors moved the agent into the live meeting

The interesting move in AI agents is not better note-taking. It is the decision by vendors to put the agent inside the live meeting rather than waiting for the transcript to arrive afterwards. That choice changes what the technology can do, and it raises a sharper question for any organisation already paying for several AI subscriptions.

The first wave of enterprise AI agents worked on artefacts after the fact. They drafted emails, generated documents and tidied calendars once a meeting had ended. The newer products treat the conversation itself as the source of value, on the basis that this is where business context is richest and where decisions are made rather than merely recorded. A representative product in this category sits inside the meeting, reads live context from connected enterprise systems, and completes downstream work without a manual handoff. It monitors conversations across video platforms, connects to systems such as Salesforce, ServiceNow, Jira and Slack, and then produces finished deliverables, updates records and books follow-ups while the conversation is still running. Pricing is set at twenty dollars per user per month.

Our position is straightforward. The live-conversation model is genuinely different from after-the-fact tooling, and for some teams it will close real gaps. But the gap between a vendor demo and a production deployment with your data, your access controls and your governance is wide, and it is where most of these tools show their limits.

Where the model earns its place and where the pressure sits

The capability set divides into three parts. There is cross-platform search that queries live conversations, connected systems and the web at the same time. There is workflow orchestration, where agents watch the meeting, detect the next required step and trigger actions across multiple systems on their own. And there is content creation, turning transcripts and enterprise data into presentations, spreadsheets and project documentation. The appeal is easy to picture. On a sales discovery call, the agent updates the opportunity record, drafts the follow-up proposal and schedules the next meeting from what was actually discussed. On a service desk video call, it routes the ticket, pulls the customer history and writes the escalation summary before the call ends.

The pressure sits in two places. The first is integration quality at scale. Enterprise environments are not uniform. Access controls, data residency rules and custom configurations differ at every organisation, and the quality of an agent operating across six or seven SaaS platforms at once is exactly where these products tend to fail. The second is overlapping spend. Many Australian organisations already pay for AI features inside Microsoft 365, Google Workspace, Salesforce and ServiceNow. Another subscription at twenty dollars per user per month needs a business case showing measurable time saved or workflow improvement that the existing tools cannot deliver. If meeting outcomes are not being actioned consistently, an agent that automates record updates and task creation may earn its place. If the real problem is that decisions are never written down in the first place, more automation will not fix it.

This pattern reaches well past the internal meeting room. The same model is appearing in contact centre technology, where agents are positioned to watch live customer interactions, surface the relevant knowledge article, suggest the next-best action and complete post-call work. The appeal is much the same in each case. An agent inside the conversation sees more than one that reads the transcript later. So are the hard questions. How does it handle edge cases? How does it reconcile conflicting data across systems? How clearly can it explain why it took an automated action?

We advise Australian organisations on AI agents for both internal collaboration and customer-facing contact centres, and our advice is consistent. Treat the demo as a starting hypothesis, not a result. Ask for a proof of concept against your own CRM customisations, your access policies and your data residency requirements. Map your current AI spend before adding to it, so the case rests on incremental value rather than duplicated function. The teams that get real value from these agents are the ones that arrive with clear use cases, sober expectations and a willingness to iterate. The promise of an agent that finishes the work your meetings start is appealing. Whether it pays off depends far less on the technology than on how well you configure, govern and integrate it.