Today's jobs Microsoft thinks could use an AI assist: Researchers and analysts
(2025/03/27)
- Reference: 1743063611
- News link: https://www.theregister.co.uk/2025/03/27/microsoft_365_copilot_reasoning_agents/
- Source link:
Microsoft on Wednesday introduced out two "reasoning agents" it claims can handle research and analysis projects.
In a departure from Redmond's [1]convoluted naming conventions , they're identified by their function, [2]Researcher and [3]Analyst .
Much like the [4]11 Security Copilot agents introduced earlier this week, Researcher and Analyst can make decent predictions and automate multi-step tasks that utilize the often adequate predictive competency of machine learning models and whatever data sources have been surrendered to the AI.
[5]
"Recent advances in LLMs allow AI to contribute like a team member," said Ashok Kuppusamy, corporate VP for M365 Core Experiences, in an [6]explanatory video . "We've taken one of these models – OpenAI's o3 deep research – along with advanced orchestration in Copilot and deep search algorithms and optimized it for work."
[7]
[8]
If your job involves development of product strategies, or writing reports about business operations, take note because Researcher can compile such documents.
The Register is unaware of any company that has achieved improved results using an AI-authored business plan, though presumably that's the goal. At least if Researcher's [9]sample advice [PDF] to fictional wearables biz Proseware flops in the market, a human scapegoat may not be necessary.
OpenAI and Google drop fresh AI model updates
The "Days since the last AI model was announced" timer has reset to zero again after OpenAI and Google announced AI model updates on Tuesday.
OpenAI's new offering [10]adds image generation to GPT-4o, which can now [11]sometimes , render readable text within images - a task previous models often butchered. Evidently, it works well enough to generate images of [12]fake identification cards .
Google meanwhile talked up [13]Gemini 2.5 Pro . Available in Google AI Studio, it has a million-token context window, with an upgrade to two million coming soon. It apparently compares favorably to the competition in assorted benchmarks.
Google has published [14]a video demo of how a single prompt can instruct the model to produce the executable code for an endless runner game.
Perhaps one day, we'll see a model that can help AI companies understand how to [15]make a profit .
Researcher, according to Kuppusamy, takes a prompt and then tries to answer it using a combination of OpenAI's deep research model, Microsoft Graph data, and external sources including the web, and data stored in applications including Salesforce, ServiceNow, and Confluence.
Analyst is oriented toward data analysis. Kuppusamy in a video showed off how it might help visualize an Excel spreadsheet of customer data.
[16]
"Usually, to make sense of this data I'd need to ask my colleague who knows Python," he [17]explained . "But let's try the Analyst agent. I don't have to spend time writing the perfect prompt to get what I'm looking for. I'll just ask it to help me come up with an easy way to learn and visualize my customer base."
Through its chain-of-thought process, the Analyst agent will work through the prompt in a series of steps and even generate Python code to create the requested graph.
"I can always click and see its chain-of-thought reasoning, as well as the Python code it's running in real time," Kuppusamy continued. "This allows me to validate and trust in its thinking and approach."
[18]
The two agents will be available next month through Microsoft’s new "Frontier" program that makes Copilot enhancements available to customers while they're still in development.
[19]AI agents swarm Microsoft Security Copilot
[20]Microsoft goes native with Copilot. Again
[21]You begged Microsoft to be reasonable. Instead it made Copilot reason-able with OpenAI GPT-o1
[22]Linux rolls out the welcome mat for Microsoft's Copilot key
Meanwhile, Microsoft has expanded the capabilities of its Copilot Studio agent-building tool by adding [23]deep reasoning and agent flows .
Deep reasoning enables CoPilot’s underlying AI model to break down complex instructions into a sequence of steps and evaluate each one in the context of the task's overall goal. This step-by-step reasoning takes more processing time but can deliver more accurate and context-aware results.
Agent flows allows agents to follow predefined logic paths, helping ensure interactions stay on track and aligned with business rules. For instance, an AI agent handling customer feedback could be configured to apply specific escalation policies and trigger a human response for high-priority issues.
Finally, [24]autonomous agents are now available within Copilot Studio. These are self-directed bots that can plan tasks, adapt to changes, and escalate issues to humans without waiting for a prompt. ®
Get our [25]Tech Resources
[1] https://www.reddit.com/r/Office365/comments/frun6x/seriously_microsoft_your_naming_standards_are_bad/
[2] https://techcommunity.microsoft.com/blog/microsoft365copilotblog/researcher-agent-in-microsoft-365-copilot/4397186
[3] https://techcommunity.microsoft.com/blog/microsoft365copilotblog/analyst-agent-in-microsoft-365-copilot/4397191
[4] https://www.theregister.com/2025/03/24/microsoft_security_copilot_agents/
[5] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=2&c=2Z-Uv1ls9Y8CBTdjUR5hX8AAAAVI&t=ct%3Dns%26unitnum%3D2%26raptor%3Dcondor%26pos%3Dtop%26test%3D0
[6] https://youtu.be/lfruwkpqvk4?feature=shared&t=9
[7] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=4&c=44Z-Uv1ls9Y8CBTdjUR5hX8AAAAVI&t=ct%3Dns%26unitnum%3D4%26raptor%3Dfalcon%26pos%3Dmid%26test%3D0
[8] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=3&c=33Z-Uv1ls9Y8CBTdjUR5hX8AAAAVI&t=ct%3Dns%26unitnum%3D3%26raptor%3Deagle%26pos%3Dmid%26test%3D0
[9] https://cdn-dynmedia-1.microsoft.com/is/content/microsoftcorp/microsoft/final/en-us/microsoft-brand/documents/Prompt1.pdf
[10] https://openai.com/index/introducing-4o-image-generation/
[11] https://x.com/AkhandPratap_/status/1904687629121241145
[12] https://x.com/elder_plinius/status/1904784415676301754
[13] https://blog.google/technology/google-deepmind/gemini-model-thinking-updates-march-2025/
[14] https://www.youtube.com/watch?v=RLCBSpgos6s
[15] https://www.wheresyoured.at/wheres-the-money/
[16] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=4&c=44Z-Uv1ls9Y8CBTdjUR5hX8AAAAVI&t=ct%3Dns%26unitnum%3D4%26raptor%3Dfalcon%26pos%3Dmid%26test%3D0
[17] https://youtu.be/9O3CoP8rEkY?feature=shared&t=44
[18] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=3&c=33Z-Uv1ls9Y8CBTdjUR5hX8AAAAVI&t=ct%3Dns%26unitnum%3D3%26raptor%3Deagle%26pos%3Dmid%26test%3D0
[19] https://www.theregister.com/2025/03/24/microsoft_security_copilot_agents/
[20] https://www.theregister.com/2025/03/06/microsoft_goes_native_with_copilot/
[21] https://www.theregister.com/2025/01/31/microsoft_open_ai_reasoning_copilot/
[22] https://www.theregister.com/2025/01/24/copilot_key_linux/
[23] https://www.linkedin.com/pulse/introducing-deep-reasoning-agent-flows-copilot-studio-charles-lamanna-n1zxc/
[24] https://www.microsoft.com/en-us/microsoft-copilot/blog/copilot-studio/unlocking-autonomous-agent-capabilities-with-microsoft-copilot-studio/
[25] https://whitepapers.theregister.com/
In a departure from Redmond's [1]convoluted naming conventions , they're identified by their function, [2]Researcher and [3]Analyst .
Much like the [4]11 Security Copilot agents introduced earlier this week, Researcher and Analyst can make decent predictions and automate multi-step tasks that utilize the often adequate predictive competency of machine learning models and whatever data sources have been surrendered to the AI.
[5]
"Recent advances in LLMs allow AI to contribute like a team member," said Ashok Kuppusamy, corporate VP for M365 Core Experiences, in an [6]explanatory video . "We've taken one of these models – OpenAI's o3 deep research – along with advanced orchestration in Copilot and deep search algorithms and optimized it for work."
[7]
[8]
If your job involves development of product strategies, or writing reports about business operations, take note because Researcher can compile such documents.
The Register is unaware of any company that has achieved improved results using an AI-authored business plan, though presumably that's the goal. At least if Researcher's [9]sample advice [PDF] to fictional wearables biz Proseware flops in the market, a human scapegoat may not be necessary.
OpenAI and Google drop fresh AI model updates
The "Days since the last AI model was announced" timer has reset to zero again after OpenAI and Google announced AI model updates on Tuesday.
OpenAI's new offering [10]adds image generation to GPT-4o, which can now [11]sometimes , render readable text within images - a task previous models often butchered. Evidently, it works well enough to generate images of [12]fake identification cards .
Google meanwhile talked up [13]Gemini 2.5 Pro . Available in Google AI Studio, it has a million-token context window, with an upgrade to two million coming soon. It apparently compares favorably to the competition in assorted benchmarks.
Google has published [14]a video demo of how a single prompt can instruct the model to produce the executable code for an endless runner game.
Perhaps one day, we'll see a model that can help AI companies understand how to [15]make a profit .
Researcher, according to Kuppusamy, takes a prompt and then tries to answer it using a combination of OpenAI's deep research model, Microsoft Graph data, and external sources including the web, and data stored in applications including Salesforce, ServiceNow, and Confluence.
Analyst is oriented toward data analysis. Kuppusamy in a video showed off how it might help visualize an Excel spreadsheet of customer data.
[16]
"Usually, to make sense of this data I'd need to ask my colleague who knows Python," he [17]explained . "But let's try the Analyst agent. I don't have to spend time writing the perfect prompt to get what I'm looking for. I'll just ask it to help me come up with an easy way to learn and visualize my customer base."
Through its chain-of-thought process, the Analyst agent will work through the prompt in a series of steps and even generate Python code to create the requested graph.
"I can always click and see its chain-of-thought reasoning, as well as the Python code it's running in real time," Kuppusamy continued. "This allows me to validate and trust in its thinking and approach."
[18]
The two agents will be available next month through Microsoft’s new "Frontier" program that makes Copilot enhancements available to customers while they're still in development.
[19]AI agents swarm Microsoft Security Copilot
[20]Microsoft goes native with Copilot. Again
[21]You begged Microsoft to be reasonable. Instead it made Copilot reason-able with OpenAI GPT-o1
[22]Linux rolls out the welcome mat for Microsoft's Copilot key
Meanwhile, Microsoft has expanded the capabilities of its Copilot Studio agent-building tool by adding [23]deep reasoning and agent flows .
Deep reasoning enables CoPilot’s underlying AI model to break down complex instructions into a sequence of steps and evaluate each one in the context of the task's overall goal. This step-by-step reasoning takes more processing time but can deliver more accurate and context-aware results.
Agent flows allows agents to follow predefined logic paths, helping ensure interactions stay on track and aligned with business rules. For instance, an AI agent handling customer feedback could be configured to apply specific escalation policies and trigger a human response for high-priority issues.
Finally, [24]autonomous agents are now available within Copilot Studio. These are self-directed bots that can plan tasks, adapt to changes, and escalate issues to humans without waiting for a prompt. ®
Get our [25]Tech Resources
[1] https://www.reddit.com/r/Office365/comments/frun6x/seriously_microsoft_your_naming_standards_are_bad/
[2] https://techcommunity.microsoft.com/blog/microsoft365copilotblog/researcher-agent-in-microsoft-365-copilot/4397186
[3] https://techcommunity.microsoft.com/blog/microsoft365copilotblog/analyst-agent-in-microsoft-365-copilot/4397191
[4] https://www.theregister.com/2025/03/24/microsoft_security_copilot_agents/
[5] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=2&c=2Z-Uv1ls9Y8CBTdjUR5hX8AAAAVI&t=ct%3Dns%26unitnum%3D2%26raptor%3Dcondor%26pos%3Dtop%26test%3D0
[6] https://youtu.be/lfruwkpqvk4?feature=shared&t=9
[7] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=4&c=44Z-Uv1ls9Y8CBTdjUR5hX8AAAAVI&t=ct%3Dns%26unitnum%3D4%26raptor%3Dfalcon%26pos%3Dmid%26test%3D0
[8] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=3&c=33Z-Uv1ls9Y8CBTdjUR5hX8AAAAVI&t=ct%3Dns%26unitnum%3D3%26raptor%3Deagle%26pos%3Dmid%26test%3D0
[9] https://cdn-dynmedia-1.microsoft.com/is/content/microsoftcorp/microsoft/final/en-us/microsoft-brand/documents/Prompt1.pdf
[10] https://openai.com/index/introducing-4o-image-generation/
[11] https://x.com/AkhandPratap_/status/1904687629121241145
[12] https://x.com/elder_plinius/status/1904784415676301754
[13] https://blog.google/technology/google-deepmind/gemini-model-thinking-updates-march-2025/
[14] https://www.youtube.com/watch?v=RLCBSpgos6s
[15] https://www.wheresyoured.at/wheres-the-money/
[16] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=4&c=44Z-Uv1ls9Y8CBTdjUR5hX8AAAAVI&t=ct%3Dns%26unitnum%3D4%26raptor%3Dfalcon%26pos%3Dmid%26test%3D0
[17] https://youtu.be/9O3CoP8rEkY?feature=shared&t=44
[18] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=3&c=33Z-Uv1ls9Y8CBTdjUR5hX8AAAAVI&t=ct%3Dns%26unitnum%3D3%26raptor%3Deagle%26pos%3Dmid%26test%3D0
[19] https://www.theregister.com/2025/03/24/microsoft_security_copilot_agents/
[20] https://www.theregister.com/2025/03/06/microsoft_goes_native_with_copilot/
[21] https://www.theregister.com/2025/01/31/microsoft_open_ai_reasoning_copilot/
[22] https://www.theregister.com/2025/01/24/copilot_key_linux/
[23] https://www.linkedin.com/pulse/introducing-deep-reasoning-agent-flows-copilot-studio-charles-lamanna-n1zxc/
[24] https://www.microsoft.com/en-us/microsoft-copilot/blog/copilot-studio/unlocking-autonomous-agent-capabilities-with-microsoft-copilot-studio/
[25] https://whitepapers.theregister.com/
Pardon?
Mike 137
how it might help visualize an Excel spreadsheet of customer data. "Usually, to make sense of this data I'd need to ask my colleague who knows Python"
Only if the spreadsheet is snake oil, I guess.
But joking apart, if a spreadsheet of customer data is so badly designed that you need a programmer to explain it, maybe that's where the problem really resides. In my consulting life I've seen many appallingly badly designed manual processes rigorously replicated in expensive automation, with predictably poor results.
AI in Research
Pascal Monett
I predict it will soon declare that it has discovered how to create a warp drive.
You'll just have to go through 10,000 pages that will basically be all of Scotty's Star Trek declarations with a mix of Star Wars hyperdrive description and maybe some Babylon 5 or The Expanse for good measure.
Good luck building an actual warp drive with that.
I'm waiting for its first plan
Neil Barnes
1) Brilliant idea!
2) ???
3) $$$
well
IamAProton
If you need marketing stuff, reports about business operations, product strategies etc. sure, 'AI' might be as good (or as bad) as humans I'm ok with that, perhaps in most instances it's doesn't really matter as long as a report is produced.
In some cases I would rather have an "intelligent agent" do the stuff for free rather that knowing somebody is getting paid to bullsh!t me (sales in general, real estate is the first coming to mind)
But please, do not design bridges with AI!
I can always click and see its chain-of-thought reasoning
I suppose it's more than you can say for Rachel Reeves.
What will be quite interesting is whether companies treat obviously bad advice from a machine any differently to obviously bad advice from human sources, because they seem to be suckers for the latter.