Microsoft Copilot - supporting research

 This article, was published mid-2025, providing an overview of the impact of Copilot researcher and analyst agents on work. 

Having now used both for 6 months plus, my take on these two agents, is that they can be useful, but need to used with care. 

1) working out what is to be done is important. Responses from researcher or analyst agents, arise through the prompts provided. Careful structuring of prompts is therefore essential in obtaining the types of actions required. Otherwise, researcher may go off topic quite easily.

2) providing context is important and delimiting the researcher/analyse agents to specific papers, websites or attachments help ringfence the direction of responses.

3) using the prompt writing agent in Copilot can be useful for tightening prompts.

4) ensuring that 'work or 'web' is selected, along with depth of responses (auto, quick, think deeper) helps again to ensure responses fit expectations.

5) Triangulation is always required, to check the validity of the responses. Limiting resources also help to save time, so that triangulation is restricted and does not have to go all over the place.

6) Draw on the advantages of AI. Summarising key points, comparison of key points across papers/sites, drilling deeper to extend insights, providing 'neutral' perspectives on conceptualisations, frameworks etc.

7) be aware of 'cognitive debt' / 'cognitive atrophy' and how this can come about very easily when your critical thinking is replaced by AI doing the work. You still need to read deeply, cogitate, make your own judgements and come up with your own synthesis. Then use AI to cross check these and see if it provides other insights which are viable.

8) Continue to learn how to manage AI to draw on it's ability to automate some research processes. The key is to use AI as a thinking partner, not to replace your own effortful thinking.

Copilot agents

A bit of 'research' into Copilot agents this morning, following on from yesterday's session on building agents in CoPilot. 

Microsoft 365 Copilot Premium (now often simply M365 Copilot) offers "declarative agents" designed for lightweight, personalized tasks using organizational data, managed within the M365 environment.  on the other hand Copilot Studio enables creation of advanced, custom agents with complex, multi-step workflows, API connections, and full lifecycle governance. 

Microsoft Copilot Studio in New Zealand is generally licensed as a tenant-wide capacity 

pack, starting at approximately NZ$323.60 per pack/month. This subscription provides 25,000 "Copilot Credits" monthly, which are consumed when agents complete actions or responses

The cost of Microsoft 365 Copilot in New Zealand varies depending on whether you are 

purchasing for an individual or a business, with prices generally starting around NNZ$34–$40 per user/month for business subscriptions

Microsoft 365 Copilot for Education is priced at approximately NZ$30–$32 for academic institutions. This specialized academic license provides AI-powered assistance for faculty, staff, and higher education students, acting as an add-on to existing Microsoft 365 A3/A5 license.

Templates for business type agents include a Quiz tutor template and a series of  'assistants' for various job tasks including travel planning, peer feedback agent, training content writer.

A guide for educators  provides ideas for using agents in educational settings. 

Overall, there are good possibilities :) 

Microsoft - introduction to building agents

 The last in the 'back to campus series' offered through Microsoft (based in Australia). 

As with the other two sessions I attended, Jennifer Ruan is the facilitator and Victor Kochetkov does the demonstration.

Notes taken during the session:

Covering what are agents, what tools exist, to through building an agent and Q & A. 

Explained Copilot as a user interface for AI. Currently there are agents accessible through Copilot and also customised agents can be build in individual copilot accounts.

Copilot is for human augmentation, private, personal assistant and 1:1 interaction.

Agents connected to Copilot, chat or autonomous and can contact with other agents.

Out of the box agents on Copilot include surveys, researcher, analyst etc. If (frontier) then they are in initial phase.

Copilot is off the shelf, retrieval only and connects to Microsoft 365 only. Copilot Studio provides task/autonomy and for developers with more complex task. Developers can also build customised agents with Copilot Studio + Azure AI.

Went through Copilot - using 'new agent' on the left menu to access. Generally, describe what you want to do, configure the agent by adding knowledge and capabilities, try out as it is being build and then use/share with others. 

Copilot agents can be shifted across to Copilot Studio to make them more sophisticated. Studio has more comprehensive tools, wide selection of models, able to publish to multiple channels and has more powerful orchestration. 

Agents generally retrieve information and reason, summarise etc. They can also take action to automate workflows and replace repetitive tasks. Automous agents will operate independently, plan, orchestrate other agents, learn and escalate. 

Example in education, an IT helpdesk agent, devise refresh agent, research tracker agent, budget management agent, study guide agent, student support agent etc.

Note: agents build in premium cannot be used by others who only have access to Copilot basic without extra billing!!

Demonstrated how to create an agent. I create one to compare Learning Outcomes against NZQA criteria in 2 minutes! Also, ask prompt agent or chat to help create the agent, providing the best prompts to use. Copilot studio has help and examples, along with templates for creating agents. Analytics are also available. Evaluation also possible, to test an agent. Remember to use AI to do the work where required. 

Discussion on who will see or have access to the agent as various people in the institution have access to various type of Copilot (basic/premium). Also with each type, there are functions that some have access to and others do not :(

A link from Moodle for a agent will work - if the learners have the right version of Copilot and use it within their Ara digital environment. Otherwise, Copilot Studio needs to be used. 

Another series later this year is seeking pre-registration - Future ready with Copilot.

Cowork by microsoft - alternatives

 Yesterday's microsoft presentation introduced Cowork. To have this available in my institution's Copilot premium, requires the turning on of access to Claude. I can see some advantages to having Cowork, as it automates tasks which are carried out, often across several apps, regularly.

An AI query (on Google) reveals that all the other main LLMs have equivalents. Claude itself has Claude Cowork. Open AI has 'operator' which costs US$200 a month. Google has Project Mariner. Manus and even Amazon (Nova Act) also have similar capabilities. 

Plus, there are a whole host of  enterprise tools - Eigent, OpenClaw,  Composio etc. O-mega offers an overview of 10 Claude Cowork alternatives 

Therefore, agentic AI is now mainstream and will have further reaching implications on work and education than Gen AI itself. 

Microsoft - Copilot chat for researchers

 Notes from a session with Microsoft on how researchers can use Copilot chat (M365 copilot (licenced) for researchers (aka Copilot premium).

The same microsoft representatives hosts and deliver the workshop.

Jennifer Ruan is the facilitator and Victor Kochetkov does the demonstration.

The session explores literature reviews, data analysis and research documentation.

Shared recent CSIRO's six month M354 Copilot trial with 7,400 Australian Public Service staff. In general, 1 hour saved each day with ability to reallocate 40% of time to higher level research activities (analysis, discovery and strategic thinking). Copilot can assist with literature reviews, grant applications, meeting and collaboration time, data analysis bottleneck, document drafting.

Demonstrated upload of document to interrogate it with the standard chat. In Auto, there is ability to select different LLMs (we seem to only have ChatGPTs but access to Sonnet, Claude etc. possible). Used notebook (only on premium) to store responses along with how to bring the sources /suggested references and sites into the notebook. Then how to share notebook. Audio summary is available (quick create button) similar to capability in Notebook LM. Study guide is available (but not on my version of  Copliot). The chat next to the notebook has the option of bringing in the Researcher agent (button not to the +). 

Then when through how to use Analyse agent to summarise trends in an uploaded Excel files. Then transfer of these into pages by using the edit in pages button (icons at bottom of response).

Copilot notebook is like Onenote (storing a variety of files). Copilot pages stores responses. Onenote files can be uploaded into notebook or chat to be used for analysis etc. by Copilot. 

Excel and powerpoint agents are also not available on my version of Copilot :( 

Previewed the 'cowork' agent which will be avaialble (for some probably!!) next month. This seems to automate tasks like organising your inbox, organising your week, prep for a meeting, research a company, find files and merge or bring files together in a certain way or for a certain purpose etc. basically pre-prompted tasks. Needs Claude to work so if this is not in your LLM list then need to work with IT to see if it will be available. 

Need to check with local organisational IT for access to LLMs other than ChatGPT and also for Excel/ Pp. However, for Excel and ppt, Copilot integrated into these and can be used when each is opened, just not connected to the Copilot premium chat.

A bit of a marketing at the beginning and usual whizz through all the various items. Will view the recording again to catch up on missed items!

Generative AI in education: Theories, applications and ethical frontiers - overview

 This book collates 21 chapters on how Gen AI is shifting education and the role of educators in understanding and utilising AI to support teaching and learning.

The book is edited by G. Durak, S. Cankaya and M. Sharples with chapter authors from Europe and Asia. It is not open access and the short pdf available provides a  preface, and list of chapters and authors.

A good selection of topics is collated through the book including:

- sustainable education - chapter 2

- classroom orchestration - chapter 4

- lesson planning - chapter 5

- deep fakes to voice cloning - chapter 7

- learning analytics - chapter 8

- smart learning environments - chapter 9

- VR and AR - chapter 10

- AI literacy - chapter 11

- Prompt design - chapter 12 & 13

- dialogic feedback - chapter 14

- creative learning - chapter 19

- ethical considerations - chapter 20

- teacher training and PD - chapter 21

Will recommend the book to the library and add chapter summaries after I have looked through them. 

MIT - unified framework of five principles in AI society

 This article, A unified framework of five principles for AI in society, written by L. Floridi and J. Cowls in 2019, provides a synthesis of other frameworks. The work predates the arrival of ChatGPT but the principles are even more important now as Gen AI moves towards becoming Artificial General Intelligence (AGI).

Table 1 in the article summarised where each principle is also present in 9 frameworks. The principles are beneficence, non-maleficence, autonomy, justice and explicability. These concur with the principles published of late, calling for increased ethical actions as Gen AI and agentic AI take hold. 

From personalised to precision learning

This paper from Educause was in George Siemen's presentation last week. It proposes a tightening of personalised learning through 'precision learning'.

The paper was published in Novermber 2025 with authors from the U S of A - 

Tanya Gamby, David Kil, Rachel Koblic, Paul LeBlanc, Mihnea Moldoveanu and George Siemens.

The article proposes a 'computed curriculum' using AI to generate and deliver content and activities dynamically.

An important first step is to review institutional data architectures so that they can be drawn on to support the demands of adaptive and personalised learning. Real-time adaptivity is required so that 'just-in-time' feedback is availed to learners. Hence, in the article, the following is required:

"Delivering real-time recommendations and interventions that improve learning outcomes requires a fundamentally reimagined architecture. Instead of static, siloed data collection, institutions need systems capable of dynamic ingestion, immediate processing, and centralized analysis of learner interactions and performance. Such an architecture might leverage event streaming platforms, robust APIs, cloud-native databases, or other technologies to make data available and actionable the moment it is created. With this shift, educational systems can identify learning gaps, trigger personalized recommendations, and adapt curricula in real time—moving from post hoc analysis to proactive support."

The combination of agile learner data and LLM can then allow for real-time support and intervention. Then, combining the above with learning profile and curriculum leads to 'precision learning'.

The paper provides a way forward, explained in lay language for those with little computer science background. Now the first steps are to find out how my institute stores their learner data and see how complex or large the task ahead will be to format and provision precision learning!

The precursors to this article are also worth a read - Ai tsunami is here (2025 - Sept); and Dialogue at scale (October 2025).