Microsoft is rolling out ‘a new set of capabilities that enable you to build autonomous agents’, which will ‘understand the nature of your work and act on your behalf – providing support across business roles, teams, and functions’.
The agents, which will be in ‘public preview’ in late November, will be created through the Microsoft Copilot Studio, and interestingly the tech giant mentioned yesterday that ‘McKinsey, Thomson Reuters, [and] Clifford Chance’ have been using the Studio to become ‘true AI-first companies. AI-first companies leverage a combination of people, Copilot, and agents to be more efficient, improve customer engagement, and improve employee experience’.
All well and good, but what is this about? Agents – as you know – are programs, usually connected to an LLM, which allow for a series of ‘chained tasks’ to be performed. There are stop and review points built in, but the hope is that eventually those review points will also be automated, although we are not there yet. Even in the most basic form, agents should provide a big leap in efficiency gains as they help you to speed through multiple tasks.
Or as MS puts it: ‘[This allows] agents to act independently, initiate events, and automate complex business tasks.’
Here are some key aspects of the MS agents:
- ‘Autonomous triggers: Agents can automatically respond to signals across your business and initiate tasks. They can be configured to react to events or triggers without human input that instead originate from various tools, systems, and databases, or are even scheduled to run hourly, daily, weekly, or monthly.
- Dynamic agent plan: Each business process can have different paths since agents create dynamic plans on the fly to handle and complete tasks. Users can view the underlying logic for each of the agents’ paths, which includes key details, steps, and systems involved. This provides insight into why the agent chose a particular method, its decision process, and context, along with detailed steps, including variables and outputs, which are crucial for debugging.
- Activity overview: Trust, dependability, and precision are at the core of establishing agent autonomy in organizations. The true strength of these agents is their ability to monitor activities across real-world scenarios, which in turn enhances business outcomes. To further transparency and accountability, we’re launching an “Activity” tab that enables makers to access a complete log of past agent runs, including progress tracking, identification of blocks, trend analysis, and review of earlier decisions.’
A key part of this, MS said, will be the OpenAI o1 series, which is intended to bring higher level reasoning to the work of LLMs. That said, there have been conflicting views on how good that reasoning is so far on complex unstructured data tasks, e.g. understanding a contract. But, this site and many others expect things to improve, just as the early GPT models improved.
So, is this a big deal?
Although Copilot has seen a lot of uptake in the legal sector, most legal innovation experts will tell you it’s not that great for legal work, primarily because it’s not been refined for specific legal use cases, nor is it necessarily going to be connected to legal data to refer to that it needs to be accurate and useful to lawyers.
That said, lawyers live in Word and other MS tools like Outlook email and Teams. So, if you have all of that legal activity in the MS suite, and then you have MS agents running through it, connected to Copilot…..then, maybe, just maybe, we’re going to see an effective tool.
Moreover, there is nothing to stop law firms and inhouse house teams refining and improving what can be done with better application layers, although that’s not an easy task and no doubt many will try to use the ‘vanilla’ version as it is, because it’s simpler and faster to do so – even if the results are less legal-centric.
On the flip side, some may say: why use MS Copilot’s agents when there are legal specific tech companies focused on leveraging LLMs and agents to try and give you results you may feel more comfortable with? Why go vanilla and generic when you can have something that’s custom-made for lawyers?
Hmmm….we will probably see some of both, and in many cases lawyers may dip in and out of a more generic agent tool, then go over to a more legal specific tool, depending on the use case. If the goal is just to move emails around and then summarise them, then MS’s agents are probably good enough. If the goal is to do a series of super-accurate contract analysis tasks in a chain then, well….maybe not.
Either way, agents are very much now at the top of the priority list and the tech there will only get better.
(Plus, see the story about Luminance this morning about the launch of their agent.)
—