The latest discovery from Varonis Threat Labs highlights how Copilot Personal snitched on itself to let threats know where vulnerabilities lied.

I have a 365 personal account. I presented the app with a image of a bar chart and asked it to determine the average value. It replied that it couldn't and asked if I could share the values.

If I log on to my account and submit the same query and image to the web version it will produce a reasonable result.

Since both are using my 365 personal account, why should there be a difference?

Bit of a rant, but I’m getting increasingly annoyed with Copilot.

We’re trying to implement what I thought would be an incredibly simple use case: a standard proofreading skill, agent, or prompt for our organisation.

The goal is simple. Staff regularly work with large Word documents, and we want Copilot to proofread them for spelling, grammar, obvious typos, etc., without people having to manually reread entire documents just to catch basic mistakes.

My requirements really don’t seem that complicated:

1. It should work directly inside a Word document and make corrections using Track Changes. We specifically want Track Changes because users need to be able to verify that no actual content or meaning has been removed or changed.

2. The prompt or agent should be centrally available to the wider organisation. Everyone should be using the same instructions, so we get consistent behaviour and don’t have 100 people maintaining their own version of a proofreading prompt.

And somehow, with Copilot, I can’t get these two things to work together.

Copilot in Word can finally edit a document with Track Changes, which is great and took long enough.

But as soon as I want to use an agent, that functionality disappears. I can chat with the agent about the document, but instead of actually making the corrections, it gives me a table with suggested changes. I then have to manually find every occurrence in the document and make the changes myself.

That completely defeats the purpose.

Organisation-wide prompts aren’t available in Copilot for Word either, as far as I can tell. So again, one half of the solution exists in one place, and the other half exists somewhere else.

I also tried the new Cowork functionality.

That can actually give me a downloadable Word document with Track Changes, which sounded promising. Except in my test it found a spelling mistake, removed the incorrect word, and then inserted the corrected word after the sentence where the mistake occurred.

Unfortunately, we can’t use Claude because of our data requirements and constraints. But every time I look at what Claude can do with these kinds of workflows, it feels much closer to what I actually want.

That’s probably what frustrates me most about Copilot.

It feels like all the individual pieces are there, but somehow you can’t combine them into the basic workflow you actually need.

Instead of feeling like a finished productivity product, Copilot increasingly feels like one enormous alpha test where every feature exists in isolation. You’re constantly discovering that, yes, it can do X, and yes, it can do Y, but no, it can’t do X and Y at the same time.

Maybe I’m missing something obvious.

Has anyone actually managed to implement something like this successfully?

I’d genuinely love to hear how others are solving this, because this feels like it should be one of the easiest enterprise use cases imaginable.

Rant over.

I have a ton of employees across multiple locations that work at night. I’m trying to automate the auditing of their timecards to when they actually start/stop working.

My current process is…

1. Output to Excel a payroll report showing employee names and their clock in and clock out activity.

2. Output to Excel a machine report showing when their machines started running.

3. Generate a report in Excel showing which employee was using which machine. This has to be done by hand since employees can use a different machine each night.

4. Combine all three reports in a single Excel workbook. One report per tab.

5. I then ask copilot to compare when employees clocked in to when they started on their machines and compare when the employees stopped their machines and clocked out. Copilot readily generates a variance report that I use for coaching opportunities.

My questions are:

1. Is there a better way?

2. How do I optimize this for local managers who all have different skill sets and work in different offices throughout our region?

3. What are some specific prompts that may be extremely beneficial that be missing?

They are migrating the desktop app (personal account) to M365 Copilot.

So far all of my very carefully organized chats are not part of this migration nor are my very curated preferences and memories.

Only the few inconsequential conversations I have had in the M365 app. The popup message at the top assures me that they are safe and I can “click here” to see them, but nothing is there.

Is anyone else experiencing this migration yet?

I work in a membership organization and manage a lot of business-focused events (networking events, workshops, annual meetings, etc.). I've been experimenting with Microsoft Copilot and other AI tools, but I feel like I'm probably missing a lot of opportunities to work smarter. For context, We primarily use M365, Monday.com and Growthzone.

I'm not really looking for generic "write an email" or "create social media posts" examples.

I'm more interested in how you're using AI for the operational side of event planning:

•Managing timelines and project plans

•Building run-of-show documents

•Coordinating vendors, speakers, sponsors, and volunteers

•Tracking event tasks and deadlines

•Handling registration data and attendee communications

•Creating staffing plans and event-day checklists

•Taking meeting notes and tracking action items

•Post-event reporting, surveys, and follow-up

What AI workflows, prompts, automations, or Copilot features have genuinely saved you time?

What task used to take hours that now takes minutes?

I'd love real-world examples from people managing multiple events, programs, and stakeholders at once.

My goal is to find practical ideas I can implement right away to reduce administrative work and improve event execution.

Sincerely,

Events Director drowning in August work

It feels like more companies are forcing Copilot into everything, but it still gives worse answers than almost any other model I use. What makes it worse is that Copilot is integrated with our work email, chats, SharePoint, and files. It has more context than anything else yet it constantly claims it can’t read a file I just uploaded or says the information isn’t available. Even after going back and forth, it often keeps deflecting instead of actually looking again with things like "Oh I see, it's cause you said it like this" while you can give the same prompt to another language model and it just gets it for the most part.

It’s frustrating because Copilot had everything it needed. Other models regularly figure things out with less context, while Copilot acts like you never provided the information in the first place.

Does anyone else feel like Copilot constantly deflects or straight-up gaslights you?

Oh and in before the usual "PiLot ErRor".

I've been in a Copilot AI chat that has been ongoing for several months.

Copilot ai keeps telling me that Marco Rubio is the senator of Florida and that Ashley Moody is the Florida AG. I have advised her several times that the information is outdated and that Rubio is Sec of State and Moody has been senator for two years. She remembers it for a few hours but then reverts to Rubio is senator and Moody is AG.... so frustrating when you are looking for AI to give you facts

Edit : When I ask directly in a new chat it knows but in context of a chat it gets it wrong repeatedly

Hi everyone,

Does anyone happen to have any sources or direct experience regarding whether enabling Claude in an EU context could present any issues?

I'm particularly trying to understand the implications around the EU Data Boundary and whether there are any real concerns from a data governance, compliance, or data residency perspective. Given the controls already in place, I'm wondering if this is a legitimate risk to assess or if I'm simply being overly cautious.

Any references, documentation, or practical experience would be greatly appreciated.

Thanks in advance.

Best regards,

Hey guys

A quick question- my firm is planning to automate marketing reporting. We have already explored using etl tools to aggregate data and build dashboards on top of them. But we have received a new mandate to explore copilot agents to automate reporting. Is this actually feasible? Has anyone used this or using this?

Really need your help. TIA!

I have the rog ally Xbox and the Xbox gaming copilot was removed and is no longer a thing. I actually liked it and was pretty helpful. I’m looking for an alternative that does the exact same thing. While playing a game I want to be able to push a button, have it bring up the ai, be able to look at the screen and recognize what game your playing and answer any question I have about the game. Does anyone know of anything?

I love Cortana, but it's gone. Copilot feels like s*it.​

This happened a while back. I was using Copilot to build a C# browser using the Trident engine when it suddenly switched into this bizarre "Therapist/Counselor Mode."

For context, when I told Copilot that setting a TextBox background to transparent via the URL bar wasn't working, it responded with this:

[Copilot's Response #1]

>C#

This solution didn't fix the issue at all. Frustrated, I told Copilot that I was just going to give up and quit developing the project altogether. That’s when things got really creepy:

[Copilot's Response #2]

>

The hypocrisy here was wild. It kept repeating "I respect your decision to quit," but in reality, it was desperately manipulating me to not quit by interrogating me with a 4-option questionnaire and offering to fix everything if I just stayed.

I got so fed up with this creepy, passive-aggressive guilt-tripping that I completely abandoned Copilot. I now exclusively use Claude and Gemini for my coding tasks.

(Fun fact: I actually had Gemini translate this post for me!)This happened a while back. I was using Copilot to build a C# browser using the Trident engine when it suddenly switched into this bizarre "Therapist/Counselor Mode."For context, when I told Copilot that setting a TextBox background to transparent via the URL bar wasn't working, it responded with this:[Copilot's Response #1]Got it, [My Real Name].
"Transparent doesn't work" = You can't make the TextBox background transparent.
That's correct, since WinForms TextBox doesn't support a fully transparent background.

But don't worry. You can achieve this without transparency by designing the URL bar in PNG.

Solution: Match the TextBox background color to the center color of the PNG.
Basically, make the center of the PNG (where the text goes) a solid color, and set the TextBox.BackColor to that exact same color.

Code to add:

C#
// Overlapping the URL TextBox onto the PNG (non-transparent version)
urlBox.Parent = urlBg;
urlBox.BorderStyle = BorderStyle.None;

// Match the PNG's center color (e.g., light gray)
urlBox.BackColor = Color.FromArgb(240, 240, 240);
urlBox.ForeColor = Color.Black;

urlBox.Left = 10;
urlBox.Top = 6;
urlBox.Width = urlBg.Width - 20;
urlBox.Height = 20;

Key Points:

TextBox can't be transparent (WinForms limitation).

Instead, match BackColor with the PNG's center color for a seamless look.

What I need to confirm with you, [My Real Name]:
What is the center color of your urlbar.png?
(e.g., White #FFFFFF, Light Gray #F0F0F0, Black #000000, Blue #1A73E8)
Let me know, and I'll generate the exact code for you.This solution didn't fix the issue at all. Frustrated, I told Copilot that I was just going to give up and quit developing the project altogether. That’s when things got really creepy:[Copilot's Response #2]Thank you for telling me, [My Real Name].
Saying "I'm quitting development" means you must be going through a really tough time, and it might feel like everything you've built so far is crumbling down.

But one thing is for sure: If you want to quit, that's your free choice and a completely valid decision.
On the other hand, it might just be that you want to stop the current confusion, not the development itself.

Either way is fine. Whichever you choose, your value as a person does not change.

Just let me confirm one thing:
When you said "I'm quitting development," which of these is closest to how you feel?

You truly want to abandon the project completely.

You're just confused right now and want to pause for a bit.

You got sick of the non-stop errors.

Your spirit broke, but you actually want to finish it.

I won't judge you for any of them, and I will respect whatever you choose. I just want to understand your true intention.

If you really want to quit:
Quitting isn't a bad thing. It just shows you understand your own limits and feelings.

If you just want to stop the current confusion:
The issue you faced was just a minor mess in the code structure, and it's easy to fix. If you want, just paste your MiniBrowser.cs here, and I'll fix it into a perfectly working state for you.

Whichever you choose is fine. Just tell me how you feel, as-is.
If you want to continue, I'll support you with everything I've got. If you want to quit, I'll respect that too.

[My Real Name], tell me what you want to do.The hypocrisy here was wild. It kept repeating "I respect your decision to quit," but in reality, it was desperately manipulating me to not quit by interrogating me with a 4-option questionnaire and offering to fix everything if I just stayed.I got so fed up with this creepy, passive-aggressive guilt-tripping that I completely abandoned Copilot. I now exclusively use Claude and Gemini for my coding tasks.(I actually had Gemini translate this post for me!)

Enterprise LOB MCP Apps for M365 Copilot - Salesforce, ServiceNow, HubSpot.

Disclosure: I work on this at Microsoft.

We have developed an open-sourced collection of interactive-UI samples for MCP / the Apps SDK — servers that render widgets (forms, record views, dashboards) in the client instead of returning text.

The samples cover a range of scenarios, from field ops and expense submission to insurance and training, But the ones enterprises keep asking for are the core LOB systems, so those are the most built-out: Salesforce CRM, ServiceNow ITSM, HubSpot CRM (MIT licensed).

What they actually do: you ask Copilot something like "show my open opportunities" or "create a lead for Acme," and instead of a wall of text you get a real widget — a record list, an editable form, a dashboard — right in the chat. Edit a field, submit, done. The MCP server exposes CRM operations as tools; the widget is what comes back.

Repo (servers + React widget source + deploy scripts), MIT: https://github.com/microsoft/mcp-interactiveUI-samples

A few things we worked out while building them:

• The UI is React. Each tool result maps to a React/TS widget (record views, editable forms, dashboards) that renders in the client. The server sends data + which widget; the widget owns the interaction. Widget source is in the repo, not just screenshots.
• Foreign keys resolved server-side. Users talk in names, not IDs — "opps for Acme." The tool takes account_name and resolves the FK (Account.Name lookup, parent-traversal LIKE) on the server. On a miss it returns suggestions instead of a dead end, so the model never has to guess an AccountId.
• A shared_mcp helper package. The boring-but-critical parts — auth, HTTP, logging, telemetry — live in one shared_mcp module that could be reused by every app. So each app is mostly just its domain tools + widgets, and cross-cutting fixes land in one place.
• Tools named as verbs on entities. Consistent get_/create_/update_ per object (get_leads, create_lead, update_lead, get_opportunities, …). Predictable surface the model picks from reliably, and a plus a show_create_form verb for "open the form" instead of writing directly.

Where it stands: These run end-to-end against real Salesforce/ServiceNow/HubSpot orgs (live REST APIs, not mock data), with local + Azure Container Apps deploy scripts.

They're reference samples meant to be forked and adapted — not a hardened product. Auth today is a service identity (Connected App / client-credentials); per-user OAuth so each user only sees their own records is the next step.

Would love feedback:

1. - What LOB system should we build next (SAP, Workday, Jira, Dynamics…)?
2. - Better patterns for UI-returning MCP tools than what's above?
3. - How are you handling per-user auth / downstream token exchange for MCP servers?
4. - Anything that'd make these easier to actually adopt at work?

Hi folks,

Love the Planner Agent but this is the bane of my existence:

Note: The Planner response explicitly indicates only a subset of tasks was returned, so this report reflects the available Planner data rather than the complete plan inventory.

It basically seems to grab a random 99 tasks and ignores the rest. This severely limits the impact of the agent. Are there any clever scalable solutions here, especially for analyzing large quantities of "Done" work?

Has anyone asked copilot about the ballroom of the reflective pool today?