Since announcing Copilot Studio last year, Microsoft claims it has achieved significant efficiency gains across multiple business units using its tools. Starting next month, customers will be able to put those claims to the test.

The service provides a no-code-style interface for building task-specific AI agents to automate employee tasks.

If you're not familiar with the concept of AI agents, this refers to applications that use a combination of large language or vision models, sometimes more than one, and conventional automation frameworks to proactively process information — a sales lead for example — without human interaction.

Although Copilot Studio can be used to build custom agents to automate any number of tasks, Microsoft is making 10 ready-made ones available under its Dynamics 365 offering, which targets everything from sales and customer service, to finance and supply chain management.

In the case of its supply chain agent, Microsoft claims it can "minimize costly disruptions by autonomously tracking supplier performance, detecting delays, and responding accordingly," adding that it will free teams from monitoring and firefighting.

Naturally, these services are designed to be plugged into customers' existing Microsoft 365 services to extend their knowledge base, presumably through a technology called retrieval augment generation (RAG) — a concept we've explored in depth here.

Long term, many industry hopefuls believe the technology will help to automate entire departments. However, for the moment, Microsoft's ambitions appear a little more grounded, with Redmond pitching its service as an efficiency driver that will help existing workers get more work done or focus on higher value tasks.

Microsoft says Honeywell has seen productivity gains equivalent to adding 187 full-time employees from the use of Copilot, which really begs the question of whether these tools are really aimed at improving worker productivity or driving workforce reductions.

That's exactly what Klarna is doing, but rather than letting the AI hand out pink slips, the Fintech outfit is using its agents as an excuse not to replace departing workers.

In its own testing, Microsoft claims at least one of its sales teams has seen 9.4 percent higher revenue per seller and closed 20 percent more deals using Copilot. Meanwhile, the software giant says its AI HR assistant is now helping answer questions with 42 percent higher accuracy.

Of course, we recommend taking these claims with a grain of salt. Percentages aren't always as impressive as they look, especially when you don't know what the baseline is. A 42 percent increase in accuracy is a lot if the HR assistant was only answering questions correctly three-quarters of the time. But, if the HR agent already responded accurately 9 out of 10 times, that's a much smaller improvement.

For all the benefits Microsoft prescribes to these agents, it remains to be seen whether enterprises will actually let these agents make decisions on their own or if they'll just add to the cacophony of other alerts already demanding these teams' attention. There's certainly plenty of reason to believe it'll be the latter.

Earlier this month, Redwood Research's CEO Buck Shlegeris shared with The Register how an AI agent designed to scan his network, identify a computer, and connect to it, went a little off the rails and began pushing updates to machines which it promptly botched.

To be clear, this particular agent was custom-written in Python and used Anthropic's Claude LLM as its backend. So, perhaps Microsoft's Copilot Studio will have better guardrails built in to prevent this kind of thing from happening in production. ®

https://www.theregister.com//2024/10/21/microsoft_copilot_agents/