AI in UCaaS Just Changed: From Building It to Switching It On

For the past couple of years, getting real value from artificial intelligence (AI) has meant building it. Crafting the prompts, grounding the model in the right data, designing agents around a specific business process and wiring them into the systems that run the company. It was, and still is, skilled work. I think of it as automating the business process, and it remains where the deepest value sits.

But something has shifted in UCaaS this year, and it changes what "adopting AI" means for a lot of organisations. The vendors have started shipping AI you don't build at all. You just switch it on.

The Switch-On Era Has Arrived

Look at what the big three UCaaS platforms are putting into the product right now.

Microsoft is building AI directly into Teams Phone. Copilot Call Delegation gives every user an AI that answers and screens their incoming calls, filters the spam, works out what's urgent, transfers the caller if it is, and books a follow-up with a summary if it isn't. And the newly announced Teams Phone Agent goes a step further: a customer-facing AI receptionist that greets callers, handles the routine requests and routes with context, in more than 60 languages. No project. No integration workshop. It's a feature.

Zoom has taken a similar path. AI Companion is bundled into paid plans, quietly summarising calls, capturing follow-ups and reading sentiment. Zoom Phone now offers a Virtual AI Receptionist, an always-on answer point for the calls that used to ring out or land in a shared voicemail nobody checks.

Cisco has done the same across Webex. The AI Assistant sits on live calls, drafts the summaries, suggests responses and, in the contact centre, scores the quality of every single interaction rather than the 2% sample a human team could manage.

None of this needed a data scientist or a prompt. It ships in the platform organisations already pay for, and in a striking number of them it is sitting there switched off.

Quality of Life First

I'd describe this first wave as quality-of-life AI. It doesn't transform a business model. It gives people their time back. Nobody types up call notes any more. The urgent call no longer gets buried between two spam calls. The receptionist role that was impossible to staff around the clock is now covered at 2am, in Portuguese if needed.

Individually these are small gains. Collectively they add up, and, more importantly, they normalise AI inside the organisation. The team that has lived with call summaries and an AI receptionist for six months is a far easier audience for the bigger conversation that comes next.

Expect much more of this. The platform vendors have found their delivery model for AI, which is to make it a toggle, and the pace of these releases is only increasing.

Where it Goes Next: The Business Process

Here's the thing, though. Switch-on AI is becoming table stakes. When every organisation on Teams, Zoom or Webex can turn on the same features, the features stop being a differentiator. They're the entry ticket.

The differentiation, and the serious value, still lives where it always has: in the build. Connecting AI to the actual business process. Handing callers to AI agents that can resolve, not just route. Putting real-time assistance in front of human agents with the customer's full context. Automating the workflow behind the conversation, wired into the systems of record. That is customer experience territory, and it's where the craft skills of the last two years, the prompting, the data grounding, the integration, really earn their keep.

The good news is that the tools for that build have grown up fast. Microsoft Copilot Studio now lets you create a custom AI voice agent and connect it straight into Teams Phone, grounded in your own business data. In the contact centre, the specialist platforms have gone further still. Genesys has brought agentic AI into the heart of its Cloud platform, with virtual agents and copilots that can act autonomously and interoperate with other AI agents and outside systems through emerging standards like MCP. Five9 has built its Genius AI suite around the same idea, with AI agents that go beyond answering questions to completing end-to-end tasks, sitting on a governance layer that keeps them within the rules you set. And dedicated AI agent platforms like Parloa now let you describe an agent's role and boundaries in plain language, then simulation-test it against thousands of AI-played customer conversations before it ever speaks to a real one.

What all of these have in common is the direction of travel: from AI that assists a conversation to AI agents that complete a process. The skill of describing a business process precisely, grounding an AI agent in the right data, and testing it until you trust it is fast becoming the most valuable capability in the channel.

So the advice I find myself giving partners and customers is simple. Switch on the simple stuff now. It's already in your licence, it pays back in weeks, and it builds the organisational muscle. Then take that momentum into the harder, more valuable question: which business process should AI run next?

The organisations that win won't be the ones that talked about AI the longest. They'll be the ones that switched it on, learned from it, and then built.

Ryan Osborne is Telecom Product Owner at CallTower, a global UCaaS and communications provider, where he leads AI across the unified communications, collaboration and contact centre portfolio.