If you’re currently on Windows 11 25H2 or 24H2, you’ll be getting Windows 11 26H2 later this year, likely around October 2026. There’s also “Windows 11 26H1,” but it’s only meant for new Snapdragon X2 and possibly other ARM CPUs. Microsoft has clearly said that version 26H1 is only for new silicon/CPUs, and version 25H2 is the latest version for now.

“26H1 is not a feature update for version 25H2 and only includes platform changes to support specific silicon. There is no action required from customers,” Microsoft wrote in a blog post dated November 7, 2025.

At that point, Microsoft also confirmed “25H2 remains the primary place for new features,” and there were no plans to release more than one feature update in a year.

“Windows 11 continues to have an annual feature update cadence, with releases in the second half of the calendar year,” the company said.

This still stands true. While both Windows 11 26H1 and 26H2 are real, they’re not the same thing. Version 26H1 is specifically for new Snapdragon X2 chips, as it includes platform changes (performance or scheduler optimization for new ARM chips). On the other hand, Windows 11 26H2 is for all non-ARM PCs, including existing and upcoming PCs.

Windows 11 26H2 confirmed

Now, on January 27, 2026, Microsoft rolled out Windows 11 Build 26300.7674, which is a significant jump in the build number and likely signals a new Windows release (not a cumulative update). In fact, Microsoft made that quite obvious by confirming it’s testing “behind-the-scenes platform changes” for Windows Insiders.

If you’re part of the Windows Insider Program and use Build 26300.7674 or newer (in the 26300 series), open the Windows Update history page. Then, go to uninstall updates, and you’ll notice references to Windows 11 26H2 “enablement package.” I’ve captured a screenshot of the page if you don’t have the build installed:

This is the first time we’re seeing references to “26H2,” as all support documents only mention “26H1” or “25H2.”

At the moment, the Windows 11 26H2 reference does not show up in Settings > System > About or Winver (via Windows Run). But that will likely change as newer builds roll out.

Also, the Windows Update page describes Build 26300.7674 as an “enablement” package for Windows 11 26H2. An enablement package usually means it’ll be a minor release, but it’s probably still too early to decide. Also, an enablement package release does not necessarily mean Windows won’t get new features in 2026.

For example, Windows 11 25H2 was also an enablement package update for version 24H2, but last year, Microsoft still shipped some major changes, including new colorful battery icons, a new Start menu, and, of course, Copilot-related improvements.

2026 is going to be a big year for Windows 11

Copilot inside File Explorer

One of the most notable and probably controversial changes could be Copilot in File Explorer.

As Windows Latest previously reported, Microsoft is testing “Copilot” chat inside File Explorer. Right now, File Explorer has “Ask Copilot” in the context menu, which only sends your files or images to Copilot.

When Copilot rolls out inside File Explorer, you’ll be able to interact with Microsoft’s AI in the side pane, similar to how you can preview files. That means Copilot will appear on the right side, and you’ll be able to switch between Details, Preview, and Copilot panes. However, the feature is going to be optional, so if you don’t like it, you don’t have to use it.

Copilot on the taskbar

The second biggest change is a new Windows Search experience powered by Copilot, called “Ask Copilot.”

I’ve already tested Ask Copilot in preview builds, and it’s actually better than the existing Windows Search UI. While it’s based on the same Windows Search index, it’s less cluttered and feels smoother.

Ask Copilot on the taskbar is also going to be an optional feature, and it won’t be turned on by default.

When you turn it on, Copilot replaces the Windows Search UI.

Agenda View is coming to the Windows 11 notifications center

Windows 10’s Agenda view is being added back to the notifications center, and it’ll include Copilot integration, but the overall experience will be more or less the same. You’ll be able to directly view Outlook agendas in the Windows 11 notifications center.

However, unlike Windows 10’s Agenda View, Windows 11’s implementation is going to be WebView2-based.

That does not necessarily mean a bad experience, but Windows Latest confirmed that Agenda View, when used or opened, adds another 100MB of memory usage, as Windows starts Edge/Web-related processes.

Modern Run Dialog

Windows 11 is also testing a new optional modern Windows Run that replaces the legacy Run when you specifically turn on “Modern Run” in Settings > Advanced.

The new modern Run uses WinUI and matches the look and feel of Windows 11, thanks to the Mica effect.

But we don’t think the modern Run is only about a new interface, since Microsoft might plan to add new features. This is quite obvious given the fact that the feature is part of Advanced Settings (for developers) and is optional.

Other changes coming to Windows 11 include the Xbox Full Screen Experience (FSE), support for .webp as desktop backgrounds, and more.

The post Windows 11 26H2 confirmed for later this year, as references spotted on Windows Update page appeared first on Windows Latest

Microsoft says Copilot usage is booming…. which was sort of expected after it integrated Copilot into nearly every product it owns, including Windows, Outlook and Office. We’ve some numbers here, thanks to Microsoft’s quarterly report.

As per the report, Microsoft 365 Copilot’s active users increased by 10X, and average conversations per user doubled YoY (year-over-year).

Now, we need to understand that Microsoft 365 Copilot is more than just ‘Copilot,’ as it lets you manage or view Office files and even act as a Microsoft Lens replacement on mobile. At the same time, the Microsoft 365 app has pivoted to Microsoft 365 Copilot, so you’re going to see Copilot here and there.

Either way, Microsoft 365 Copilot is growing, and Microsoft credits features like Researcher Agent.

“Microsoft 365 Copilot is also becoming a true daily habit, with daily active users increasing 10X year-over-year,” Microsoft said. “We are also seeing strong momentum with the Researcher agent, which supports both OpenAI and Claude, as well as Agent Mode in Excel, PowerPoint, and Word.”

Microsoft also revealed that it added more Microsoft 365 seats for enterprises. It’s up by 160% in a year, which is quite impressive.

“We saw accelerating seat growth quarter-over-quarter and now have 15 million paid Microsoft 365 Copilot seats, and multiples more enterprise Chat users,” Microsoft noted.

Microsoft is also reporting wide adoption of Microsoft 365 Copilot among commercial customers. For example, the company says it has tripled the number of customers, reaching up to 35,000 seats. Those customers include NASA, the US Department of the Interior, and Westpac, all of which purchased over 35,000 seats.

Copilot consumer app daily users nearly tripled, and it’s moving into shopping

Some of you might argue that the above numbers only refer to Microsoft 365 Copilot, which is being used for other purposes. But Microsoft also says that Copilot (consumer app) usage is up. For example, the company confirmed that daily users of the Copilot app increased nearly 3X year-over-year.

We don’t have any other numbers for Copilot, and I do believe it’s true that Copilot usage increased 3X YoY due to the fact that it’s pre-installed on Windows 11, among other places.

“In consumer, for example, Copilot experiences span chat, news feed, search, creation, browsing, shopping, and integrations into the operating system,” Microsoft said.

Copilot is gaining momentum across several categories, including shopping, which is why the company partnered with PayPal, Stripe, and Shopify to integrate an AI-powered shopping experience.

While these numbers sound good on paper, the reality is that Copilot on the web is struggling, as it has close to 1% market share. On the other hand, ChatGPT and Gemini dominate more than 65% of the chatbot market.

The post Microsoft says MS 365 Copilot is now a daily habit, Copilot for consumers daily users up 3X after telling everybody to use it appeared first on Windows Latest

The “Black Monday” market crash in October 1987, when the Dow plunged more than 22%. The antitrust ruling against Microsoft in April 2000, when a federal judge declared the company a monopolist. The Surface RT writedown in July 2013, when its $900 million bet on tablets went bad. The COVID crash of March 2020, when the world shut down and the market fell off a cliff.

Now add to the list of Microsoft’s worst single-day stock plunges: the company’s better-than-expected earnings for the second quarter of fiscal 2026, last Wednesday, Jan. 28.

Microsoft’s shares fell as much as 12% in trading the day after the earnings report, before closing at $433.50, down 10%, erasing $357 billion from its market value.

It was the largest single-day dollar loss in Microsoft’s history and the seventh-largest percentage decline since the company went public in 1986. The last time Microsoft’s shares fell this much after an earnings report was that Surface RT debacle 13 years ago. The stock barely moved on Friday, leaving the massive losses intact going into the upcoming week.

On the surface, the quarter itself was strong. Revenue rose 17% to $81.3 billion. Adjusted earnings hit $4.14 per share, above the $3.91 consensus. Operating margin was 47.1%. Microsoft Cloud revenue was more than $50 billion for the first time.

“Even in these early innings, we have built an AI business that is larger than some of our biggest franchises that took decades to build,” Microsoft CEO Satya Nadella said on the earnings call.

The concerns behind the numbers

So what happened? Several factors are playing out behind the scenes.

Microsoft’s Azure cloud platform grew 38% in constant currency, ahead of its own forecast. But Wall Street’s “whisper number” was 39.4%, and that miss was enough to shake the market.

Capital spending soared to $37.5 billion in the quarter, up 66% from a year ago, illustrating the size of the risk Microsoft is taking. The company is racing against Amazon, Google, and others to build data centers and buy chips to compete in AI.

Microsoft 365 Copilot, the AI assistant that the company has built into its Office apps, has 15 million paid users. That sounds like a lot until you consider that Microsoft 365 has 450 million paid seats. Copilot has reached a little more than 3% of them.

Microsoft’s outlook for the upcoming quarter didn’t help. Its forecast for the Windows and Devices business was more than $1 billion below what analysts expected, as the wave of PC upgrades sparked by Windows 10’s end of life runs its course and loses steam.

But the number that seemed to concern investors most was this: 45% of Microsoft’s $625 billion in remaining performance obligations (RPO) is tied to OpenAI. 

RPO represents contracts that customers have signed but Microsoft has not yet fulfilled. It is a measure of future revenue already locked in. Microsoft’s report showed that roughly $281 billion of that backlog is committed to a single customer, a company that is still burning cash and searching for a sustainable business model.

The question is whether that investment will ultimately pay off in demand from Microsoft’s business customers, not just from OpenAI and other AI companies.

In a post before earnings, Judson Althoff, CEO of Microsoft’s commercial business, pointed to customers using AI to solve real problems: Epic generating 16 million patient record summaries a month, Land O’Lakes building an assistant that turns an 800-page crop guide into instant recommendations for farmers, Mercedes-Benz using AI agents to cut factory issue diagnosis from days to minutes.

A ‘prove it’ moment

But analysts at UBS reflected the broader market skepticism, as noted by CNBC. Although Microsoft 365 commercial revenue was up 16% in the quarter, to more than $24.5 billion, that’s not because of Copilot, they said, citing their checks with customers. 

“We think Microsoft needs to ‘prove’ that these are good investments,” they wrote.

Others were more optimistic. Morningstar maintained its $600 fair value estimate for Microsoft, calling results “consistent with our long-term thesis.” The stock, which closed Friday at $430, “remains one of our top picks,” analyst Dan Romanoff wrote.

William Blair analyst Jason Ader titled report “A Lot to Like” and noted that demand for AI and cloud services continues to outpace supply. The firm also pointed to accelerating enterprise adoption, observing that the number of customers with more than 35,000 Copilot seats tripled year-over-year.

Wedbush analyst Dan Ives maintained an “Outperform” rating but lowered his price target to $575. He cited friction between long-term investments and short-term investor patience.

“The Street wanted to see less cap-ex spending and faster cloud/AI monetization,” Ives wrote, “and coming out of the gates it’s the opposite.” He described 2026 as an “inflection year” for the company and called the selloff a buying opportunity.

Rick Sherlund has watched Microsoft go through these plunges before. The longtime Wall Street analyst, who started covering the company when it went public, observed in an appearance on CNBC that the market seemed to be in “a foul mood” last week.

Consumer AI tools like ChatGPT get the attention, but businesses actually pay. The real driver, Sherlund said, will be agentic AI, where software agents interact with enterprise systems and with each other, burning enormous computing cycles in the process.

In terms of demand, he said, the enterprise AI market is “really just getting started.”

Judging from the market’s mood, Microsoft is no longer getting the benefit of the doubt.

A few weeks ago, a small open-source project called Clawdbot started floating around my timeline. The clips were interesting — AI agents running commands, interacting with real systems — but I’ve been around long enough to know that most viral demos don’t survive first contact with real usage.

Then it changed names.
First to Moltbot.
Then to OpenClaw.

Normally, that’s a warning sign.

This time, it wasn’t.

Instead of fading out, the project matured. The intent got clearer. The boundaries got sharper. And the conversation shifted from demos to actual workflows.

That’s when I decided to run it.

The Name Changes Were Maturation, Not Chaos

Clawdbot was the spark.
Moltbot was the project shedding its original identity — partly driven by trademark pressure and partly by a need to reset expectations. OpenClaw is the stable form.

By the time it landed on OpenClaw, the framing was explicit:

This isn’t a chatbot.
This is a local-first AI agent framework with controlled execution.

How I’m Running It (Deliberately, Not Recklessly)

I’m not running OpenClaw on my primary machine.

I deployed it on an old Ubuntu box with intentionally limited permissions, wired it to a dedicated email account used solely for the agent, and built a Telegram bot as the control surface.

That Telegram bot is paired to my user only.
It’s not fully hardened — and I’m not pretending it is — but it’s restricted enough to be a safe starting point.

Only my Telegram user ID can issue commands. No public access. No group chats. No broad exposure.

That’s the right first step:
limit surface area first, harden iteratively.

Setup Was Refreshingly Boring

The setup didn’t fight me:

• Install OpenClaw
• Attach a mode (I’m using Claude Sonnet 4.5 through OpenRouter)
• Start the local gateway
• Pair the Telegram bot and restrict it to my user ID

No cloud lock-in.
No opaque control plane.
No permissions I didn’t understand.

The gateway runs locally. Telegram is just an interface — not the executor. If the machine goes down, the agent disappears. That’s the correct failure mode.

What I Actually Use It For

Even in a constrained environment, the utility shows up quickly.

Remote Awareness Without Logging In

From my phone:

• “What’s running on that Ubuntu box?”
• “How much disk is left?”
• “Anything unusual since yesterday?”

No SSH. No dashboards. No context switching.
This is still an experiment — not how I plan to operate day-to-day — but it clearly demonstrates what’s possible when intent goes straight to execution.

Ambient Utilities With Memory

I don’t want another app.

I want to message:

• “Check the weather”
• “Summarize system status”
• “Tell me if something changed”

And not have to restate context every time.

OpenClaw persists memory locally. I can inspect it. I can delete it. That matters.

File Hygiene Without the UI Tax

“Find files I touched last week.”
“Group them.”
“Move them somewhere sane.”

Not magic. Just fewer steps between intent and execution.

Memory I Can See and Control

OpenClaw doesn’t pretend memory exists — it writes it to disk. If I want to version it or wipe it, I can.

That’s the difference between an assistant and something you can actually trust.

Using Cron for Daily Briefings (Early, but Promising)

One feature I’ve started experimenting with — and plan to expand — is cron-based daily briefings.

OpenClaw includes a built-in cron scheduler, which means the agent doesn’t just respond when you message it. It can wake itself up on a schedule, run a task, and report back.

At the moment, I’ve only had this running for about two days, but even at this early stage it’s useful.

Each morning, on a fixed schedule, the agent:

• runs a lightweight system check on the Ubuntu box
• summarizes anything noteworthy since the previous run
• sends a concise briefing to my paired Telegram bot

No prompts. No follow-ups. It just shows up.

This is where the model shifts from interactive assistant to proactive agent.

I’m deliberately keeping the scope small for now — no inbox scanning, no external APIs, no aggressive automation. The goal isn’t autonomy; it’s reliable, predictable behavior on a cadence I control.

Even this minimal setup proves the point:

An AI agent that can act on a schedule — without being prompted — is categorically different from a chatbot.

Cron turns OpenClaw into something that works for you, not just with you.

A Quick Reality Check: This Is Still Experimental

Just because you can get OpenClaw up and running quickly doesn’t mean you should treat it casually.

You are giving an AI agent execution capability. That comes with real risk.

Even in a local, sandboxed setup, you need to be thoughtful about:

• security boundaries
• prompt injection
• what your machine has access to
• what the agent could indirectly reach if something goes wrong

If the agent can read files, it can read sensitive files.
If it can run commands, it can run dangerous ones.
If it has network access, it can be influenced in ways you didn’t intend.

This isn’t hypothetical — it’s basic threat modeling.

That’s why I started with:

• a locked-down Ubuntu box
• limited permissions
• a dedicated email account used solely for the agent
• a paired Telegram bot restricted to my user
• and a deliberately narrow scope of actions

This is not something I’d recommend running on a primary workstation or any machine with broad credentials without additional hardening.

The upside is real — but so is the blast radius if you’re careless.

Treat OpenClaw like any other powerful automation tool:
use least privilege, assume failure, and expand slowly.

That caution isn’t a weakness of the system.
It’s the cost of letting AI actually do things.

The Real Insight: It Was Never “AI Can’t Do Things”

After using this for a bit, the conclusion is obvious:

AI was never incapable of acting.
We just didn’t design systems to let it — safely.

LLMs have been able to plan and reason for a while now. But humans remained the execution layer by default — copying commands, clicking buttons, stitching workflows together.

Not because of technical limits.
Because of fear and product decisions.

OpenClaw makes a quieter, more important argument:

AI should be allowed to act — deliberately, observably, and with constraints.

Not full autonomy.
Not chaos.
Just fewer unnecessary handoffs.

Why This Didn’t Flame Out Like Other Agent Projects

Most agent frameworks collapse because they chase autonomy without discipline.

OpenClaw survived because it went the other way:

• local-first execution
• explicit permissions
• human confirmation for risky actions
• clean separation between interface (Telegram) and execution (gateway)

It doesn’t try to be smart everywhere.
It tries to be useful where you allow it.

That’s why the Clawdbot → Moltbot → OpenClaw evolution worked instead of imploding.

This Is Officially a Weekend Project

Once you see this working — even on a locked-down Ubuntu box, with a paired Telegram bot and early cron jobs — your brain starts asking better questions:

• Where does human-in-the-loop actually belong?
• Which guardrails should be architectural, not policy-driven?
• How many workflows are manual simply because “that’s how it’s always been”?

That’s when you know something’s real.

This is now a next-weekend project for me.

Not because it’s flashy.
Because it’s directionally correct.

References:

• OpenClaw - Personal AI Assistant
• Integrations - OpenClaw