Microsoft has started rolling out the August 2026 Patch Tuesday update for Windows 11, and after going through the full changelog and testing everything, I feel there is no headline feature this time. It is a quality-of-life update with dozens of small fixes and refinements.

The update ships as Windows 11 KB5121003 (OS build 26200.9165 and 26100.9165) for Windows 11 version 25H2 and 24H2, and takes everything from the July 28 optional preview, KB5101684, and adds it into the mandatory August security release.

Voice Access gets its biggest upgrade in a while, File Explorer finally shows file sizes properly, Windows Hello adds support for external fingerprint readers, and there is a long list of smaller reliability work across the taskbar, Start menu, networking, and power settings.

Like every recent release, and I feel like a broken record here, this update also arrives in two phases. A gradual rollout hands features to devices over time through Microsoft’s Controlled Feature Rollout system, so some of what is listed below may take extra days to show up after you install. A normal rollout goes to every eligible device at once and covers the security-critical pieces, including Secure Boot.

Speaking of which, Microsoft confirms that Secure Boot certificates from 2011 keep expiring in stages through 2026, and that PCs still waiting on the newer 2023 certificates will keep booting and updating normally in the meantime. If you have not checked your Secure Boot status recently, it is worth a look.

Windows 11 version 24H2 Home and Pro reach end of updates on October 13, 2026, with Enterprise and Education supported until October 2027. 24H2 is already on the list of 15 products Microsoft is retiring in 2026, and Microsoft is force-installing 25H2 on eligible Home devices to keep them up-to-date.

What’s new in the Windows 11 August 2026 Patch Tuesday update

Here is a quick list of everything coming to your PC with the mandatory August update:

• Voice Access adds Voice Isolation, Korean language support, and more reliable startup.
• File Explorer shows file sizes in KB, MB, or GB instead of everything in KB.
• Middle-click now opens a folder in a new tab from the address bar and Home page.
• Windows Hello Enhanced Sign-in Security works with external, plug-in fingerprint readers.
• Touchpad gestures add adjustable scroll and zoom speed, plus accelerated scrolling.
• Taskbar notification badges switch from red to your Windows accent color.
• You can now remove the Image Generation AI component from supported Copilot+ PCs.
• Windows Search gets better at handling typos and partial app names.
• Power and battery settings apply more consistently across sleep, hibernate, and lid-close.
• Secure Boot certificate rollout keeps expanding to more eligible devices.

Voice Access gets its biggest upgrade in a while with Voice Isolation

The only “big” new change this month is on Voice Access, and it is part of Microsoft’s focus on voice as a priority area for 2026. Voice Access now has three speech recognition modes under Voice Access settings > Improve speech recognition:

1. Voice Isolation filters out other speakers and background noise, though it needs a one-time voice setup first by going to Voice Access settings > Improve speech recognition and selecting Voice Isolation.
2. Remove background noise only cuts out non-speech sounds like typing or a door closing, with no setup required.
3. No filtering keeps the default microphone input untouched.

I tried Voice Isolation with a podcast playing in the background, and the difference over the default mode was noticeable. Voice Access didn’t pick up any unwanted words from the audio, which had been a real problem before this update.

Voice Access also adds Korean language support, and Microsoft also improved the reliability of starting Voice Access. Interestingly, Windows chief Pavan Davuluri wrote in his July update on Windows quality that voice was one of the areas the team has been focused on since March.

File Explorer finally shows file sizes in the right units

File Explorer’s Details view has shown every file size in kilobytes for years, regardless of whether the file was 10KB or 10GB, and all of us have gotten used to it by now.

Anyway, with the August Patch Tuesday update, file sizes display using appropriate units, KB, MB, or GB, based on the size of the file, so a 10MB file no longer shows as 10,000KB.

I have always found this oddly frustrating given how basic a fix it is, since converting kilobytes to gigabytes in your head while browsing a drive is not something we want to waste our brain tokens on!

The feature is rolling out gradually, so it may not be immediately visible.

Address bar navigation is also improved. Middle-click now opens a folder in a new tab from both the address bar and the File Explorer Home page, matching how tabbed browsing works elsewhere in Windows.

Microsoft also fixed a grey flash on load and unexpected scrolling to the top of the Home page in certain cases, and file thumbnails in the Recommended section are now crisper.

Windows Search gets better at handling typos for apps

The August update brings forward Search improvements Microsoft has been testing for months. App search now better handles typos, dropped letters, extra characters, and partial names, so a misspelled app name is more likely to show the app you meant to search for.

Searching Windows Settings also improves here, meaning more useful settings should appear higher in the results list.

We first saw the typo-tolerant search behavior in an Insider Experimental build back in June, where searching “pwerp” correctly returned PowerPoint. It is good to see that work reach all PCs now.

Windows Hello Enhanced Sign-in Security now works with external fingerprint readers

Windows Hello Enhanced Sign-in Security, or ESS, previously required a built-in fingerprint sensor. The August update extends ESS to peripheral fingerprint sensors, so desktops and other Windows 11 PCs without a built-in reader, including Copilot+ PCs, can use this more secure sign-in method too.

To set it up, plug in a supported ESS fingerprint reader, go to Settings > Accounts> Sign in options, and follow the enrollment prompts.

Microsoft first disclosed this feature back in January 2026 through KB5074105, and it is only now beginning to roll out to all PCs.

Touchpad gestures add adjustable scroll speed and accelerated scrolling

New gesture controls for precision touchpads come under Settings > Bluetooth & devices > Touchpad. You can adjust the baseline speed for scroll and zoom gestures, and there is a new accelerated scrolling option that speeds up scrolling the more you repeat the gesture, so long documents move faster the longer you keep scrolling.

I did not expect to care about this one, but after using accelerated scrolling on a lengthy PDF I had downloaded, it turned out to be one of the more surprisingly useful small additions in this update. It is the kind of change you only appreciate once you have used it.

You can now remove the Image Generation AI component from Copilot+ PCs

Supported Copilot+ PCs can now uninstall the Image Generation AI component where it is installed, giving users who do not use the feature a way to reclaim the space it takes up. Microsoft has slowly made AI components optional after pushback over disk space and unwanted defaults.

Widgets and Start menu pick up smaller refinements

Taskbar notification badges switch from red to match your Windows accent color, which is a smaller version of the badge changes Microsoft already rolled out for Insiders earlier this year when they turned off the MSN feed and ads in Widgets by default.

New users also see a simpler Lock screen, with Weather as the only widget shown by default instead of a cluttered set of cards.

Start menu view preferences are now retained more reliably, so your chosen layout won’t get reset spontaneously. Keyboard navigation also improves when focus is set to the apps list within the Start menu.

Also, the Start menu’s account control flyout gets a design refresh and now shows a badge for your subscription status, which you can view by selecting your account picture in the lower-left corner. It is limited to users signed in with a Microsoft account, and availability varies by device and market.

Accessibility, input, and windowing improvements

Magnifier on touch-enabled devices gets a fix for the horizontal and vertical touch bars used to pan the magnified view. These are now off by default so they no longer obstruct magnified content, though you can turn them back on in Settings > Accessibility > Magnifier if you use touch panning.

Fluid dictation for Voice Typing is now off by default for new users, with a teaching tip shown the first time the feature is used. Mouse cursor size now persists more reliably between sessions, and system dialogs open at the appropriate size on small tablets.

Power, Windows Update, and networking reliability improvements

With the latest Windows 11 update, Power and battery settings apply more consistently now. Changes you make in Settings, covering display, sleep, hibernate, the power button, the sleep button, and lid-close behavior, now apply across all power plans instead of just the active one.

Microsoft also restored the ability to set a threshold for when Energy Saver turns on, under Settings > System > Power & Battery.

Windows Update gets two minor fixes. Update progress calculation in Windows Update Settings is more accurate now, and update clean-up logic has changed to improve system performance right after an update finishes installing.

On networking, DHCP renewal reliability improves in a few scenarios, particularly when a device receives a NACK from the DHCP server, and on devices using Modern Standby.

Print performance also improves for IPPS printers, so pages should print faster on secure Internet Printing Protocol connections.

Clipboard reliability improves in some Remote Desktop and Azure Virtual Desktop scenarios too.

General reliability improvements across explorer.exe and sign-in

Microsoft continues chasing explorer.exe stability. The August update improves reliability when opening Jump Lists and recent files, when sharing files and folders, and when using Task View and multiple desktops. It also improves the reliability of the Windows sign-in and lock screens, especially when system memory is low, and improves Start menu and taskbar reliability during startup.

System sounds are more consistent when using Windows in dark mode, and you wonder how sounds and dark mode are even related!.

Time zone change detection is more accurate, and daylight saving time data has been corrected for Middle East Standard Time in Beirut, Morocco Standard Time in Casablanca, Israel Standard Time in Jerusalem, and Greenland Standard Time in Nuuk.

Fonts get attention too, with restored support for Unicode variation sequences in the Myanmar Text font and improved character shaping for the Mongolian Baiti font. The taskbar’s system tray area also loads more reliably on touch devices used in tablet mode.

And an important feature for parents, Windows Setup now shows a parental controls notice highlighting available family safety features during setup.

Secure Boot certificate rollout continues in the normal phase

Unlike most of the changes above, Secure Boot updates ship in the normal rollout phase, so it applies as soon as you install the update. The release adds more high-confidence device targeting data, expanding the pool of devices eligible to automatically receive the 2023 Secure Boot certificates. Certificate deployment through Windows Update keeps continuing across supported PCs and non-managed business devices over the coming months.

Microsoft has been clear that devices still waiting on the new certificates are not at any immediate risk. Standard Windows updates keep installing normally, and the certificates will arrive automatically once a device shows enough successful update signals. If you want to check your own status, open Windows Security > Device Security, and look under Secure Boot for a green checkmark, yellow warning, or red alert.

Other fixes shipping in the normal rollout

Office applications running in virtualized environments are more stable, both while in use and when closing.

File Explorer no longer incorrectly marks files on a DFS mapped drive as originating from the internet after a device starts offline and later reconnects, which had been triggering a false “this file could harm your computer” warning in the Preview Pane along with an unexpected Mark of the Web on copied files.

File History backups to network shares over SMB are fixed too. Backups were failing with a false “invalid credentials” error even when the credentials were correct, and that is resolved now.

AI components are updated in the background

The August update refreshes the internal AI components Windows 11 uses across the system to version 1.2607.840.0, covering Image Search, Content Extraction, Semantic Analysis, and the Settings Model. These update silently and do not need any action from you.

Servicing stack update and known issues

Microsoft is shipping servicing stack update KB5101711 (build 26100.8962) alongside this release, which improves the reliability of the Windows update installation process itself.

As of this writing, Microsoft says it is not aware of any known issues with this update, a welcome change after July’s update briefly caused shutdown and overheating problems on some Dell PCs before Microsoft shipped an emergency fix.

How to install the August 2026 Patch Tuesday update

Go to Settings > Windows Update > Check for updates to grab it. Your PC moves to build 26200.9165 if you are on 25H2, or 26100.9165 on 24H2, once installed. Turning on “Get the latest updates as soon as they’re available” can help it appear sooner.

Given how much of this update is gradual, do not be surprised if some of the smaller items above, particularly the File Explorer file size change or the touchpad gestures, take a few days to show up even after installing. Yes, this isn’t a feature-rich update, but if Microsoft keeps shipping quality fixes like this every month, Windows 11 ends up a noticeably stable OS by the end of the year.

The post I tested Windows 11 August 2026 update, here’s everything new, improved, and fixed appeared first on Windows Latest

Google is gearing up to reveal a bunch of new Pixel devices on August 12th. A series of leaks leading up to the event suggest that the Pixel 11 lineup will come in an array of colors, with signs pointing to a built-in light coming to Pro models. Teasers from Google also indicate that it will show off a next-gen Pixel Fold alongside a Pixel Watch 5 in four different finishes. There might be a new Pixel Tag tracker as well.

Google may take the wraps off the new devices well ahead of its Made by Google livestream. On Google’s website, the countdown for preorders ends on August 12th at 10AM ET — eight hours before its event at 6PM ET. Comedian Trevor Noah will host this year’s show, featuring notable guests like NBA star Stephen Curry, podcaster Alex Cooper, Indian cricketer Shubman Gill, and Japanese actress Ayami Nakajo. We’ll be live blogging the event, so stay tuned!

Follow along for all the latest news about Google’s Pixel devices.

• Google’s upcoming Pixel phones are for the fans, even if its launch event isn’t
• What to expect from Google’s 2026 Pixel hardware launch event

A deep dive into building a flexible Three.js Geometry Painter with WebGPU and TSL, from surface picking and procedural geometry to live controls, custom shaders, lighting, and post-processing.

As AI coding assistants shift the developer's primary role from writing boilerplate to reviewing and maintaining systems, language choice becomes critical for long-term architectural integrity. Go directly addresses this new paradigm by utilizing its strict compiler, integrated toolchain, and uncompromising readability to provide deterministic guardrails that help AI models self-correct and generate highly standardized code. By enforcing ecosystem-wide consistency and strict backward compatibility, the Go platform empowers engineering teams to efficiently verify, optimize, and maintain high-velocity, AI-generated output in production environments.

Most of the questions I have about our APIs aren’t really code questions. They’re org questions. Who owns /customers? Do we already have an API that does user preferences, or should I build a new one? Which of these seventeen payment services is the one the mobile team actually uses?

Those questions used to require Slack archaeology. You’d ping someone from three months ago, wait, get forwarded twice, and eventually land on the person who knows. Half the time you gave up and built the duplicate service anyway.

The Postman Context Graph is the answer to those questions once your APIs, specs, collections, and connected repos are in Postman. And the fastest way to interrogate it is by asking the Postman AI Engineer natural-language questions. In this post I’ll show you the prompts I actually use, what comes back, and what I’ve discovered about our own systems that I didn’t know before.

What you’re actually querying

The Context Graph is a live index of everything Postman knows about your APIs and how they connect. Nodes are APIs, specifications, Postman Collections, environments, workspaces, monitors, mock servers, backend repos, frontend repos, route handlers, and call sites in source code. Edges are the relationships between them: which collection calls which endpoint, which repo implements which route, which workspace owns which API, which service calls which downstream API.

When you ask the AI Engineer a question about your ecosystem, it’s querying that graph first before it writes anything. That’s what makes the answers grounded instead of hallucinated. The graph tells the agent what exists, who owns it, and how it’s connected. The agent turns that into an answer.

You can reach the AI Engineer from a few surfaces: from Agent Tasks inside the Postman workbench, from Slack, and (as it rolls out) from the Postman CLI, the Postman API, and integrations like Claude or GitHub. All of them hit the same graph.

The prompts I actually use

I’ve grouped them by the question I’m trying to answer. Copy any of these into Agent Mode against a workspace you own and see what comes back.

Finding ownership

Nothing kills momentum faster than not knowing who to ping when an endpoint returns a 500 at 4pm. This is the prompt I lean on most:

Who owns the API that serves GET /customers/:id?
Return the team, the workspace, and the repo that implements the route handler.

You get back the owning team, a link to the workspace, and the repo that contains the handler. If ownership is ambiguous (as in, two workspaces claim the same endpoint), the AI Engineer says so and lists both, which is itself useful information.

Finding consumers before you change something

This is the classic use case. Before you rename a field or deprecate a parameter, ask:

List every consumer of GET /customers/:id.
Include backend call sites, frontend call sites, Postman Collections,
mock servers, and monitors. Group by owning team.

The output is a real map of who’s on the hook if you make a change. I covered a full-run example of this in the multi-repo changes post. What matters here is that even without doing the change, the answer alone tells you whether the field rename is a one-team conversation or a five-team coordination.

Discovering duplication

Every large org has this problem. Someone builds a service that already exists because they couldn’t find the existing one. Ask:

Do we have any existing API that manages user notification preferences?
Show me endpoints, spec files, and collections that look like a match,
even if they're named differently.

The graph does more than a name match. It looks at OpenAPI descriptions, schema shapes, and the request/response payloads in collections. I’ve found duplicate services this way twice, both times by describing what I wanted to build rather than what I thought it might be called.

Finding drift between spec and implementation

Specs drift. A team ships a new required field and forgets to update the OpenAPI specification. Ask:

For every API in the payments-workspace, compare the OpenAPI spec
to the actual route handlers in the connected repo.
List any endpoints where the spec and code disagree, with a short reason.

You get a list like POST /refunds: spec says amount is a string, code accepts a number. This is the kind of finding that used to require reading every route by hand.

Finding orphaned collections and monitors

Collections have a way of accumulating. Someone writes tests for an endpoint that gets deleted, and the collection lives on, running against nothing. Ask:

Find every Postman Collection in this workspace whose requests target
endpoints that no longer exist in any API spec or route handler.

I ran this on my own workspace and cleared out a dozen collections in an afternoon. One of them had a passing monitor. Passing tests against an endpoint that hasn’t existed in eighteen months. That’s the sort of thing you want to know.

Auditing consistency across services

Larger orgs care about consistency. Are we using camelCase everywhere? Do all endpoints paginate the same way? Ask:

Review the APIs in this workspace for design consistency.
Flag any endpoints that use snake_case fields, non-standard pagination
patterns, or inconsistent error response shapes. Give me a per-team report.

The response is usually longer than you’d expect. It’s also the kind of report that’s hard to get any other way, because it requires looking at every endpoint across every service at once.

Auth and security posture

You can ask the graph what’s exposed and how:

List every endpoint across all APIs in this workspace that doesn't require
authentication, or that accepts API keys in the query string.

The AI Engineer walks the specs, the collections, and the route handlers to find endpoints matching either condition. Query-string API keys are a known bad pattern because they end up in logs and referrer headers. Finding them all in one query beats a codebase-wide grep.

What the Context Graph view looks like

Alongside the AI Engineer, there’s a Context Graph view in the Postman workbench that renders the graph directly. You can pick a node (an API, a collection, a repo) and see its neighborhood: the specs that define it, the collections that call it, the repos that implement it, the monitors that guard it.

I use the view for two things:

1. Sanity-checking the AI Engineer’s answers. If the agent says four repos consume an endpoint, I can open the endpoint’s node in the graph view and see the four repos as edges. If a fifth one shows up in the view that the agent didn’t mention, that’s a bug I want to know about.
2. Serendipity. When you’re clicking around the graph, you notice things. “Wait, why does the invoicing service call the auth API directly instead of going through the gateway?” That kind of question surfaces because the graph makes the connections visible.

Most of the time I ask the AI Engineer a question and read the answer. The graph view is when I want to poke around without a specific question in mind.

Things I’ve discovered this way

A few real findings from the last couple months. Yours will look different, but the pattern is similar:

• Two teams shipping the same webhook signing logic in different repos. Neither knew the other existed. The graph linked them because both implemented POST /webhooks/verify, and the AI Engineer surfaced the duplication when I asked “what do we have for webhook signature verification?”
• A collection with a monitor running twice an hour against an endpoint that had been deleted eleven months earlier. The monitor was passing because the endpoint returned a 404 and the assertion checked for “status is not 500.”
• A frontend repo that was still calling a v1 endpoint the backend team had assumed no one used. The team had migrated the “official” consumers a year ago and moved on. The graph found the old call site because it indexes the source code.
• Three separate services that all validated email addresses with slightly different regexes. Two of them accepted the third one’s test emails as invalid. This came out of a consistency audit prompt.

None of these are dramatic discoveries. They’re the kind of thing that would take days of digging without the graph and take about five minutes with it.

Tips from the trenches

Ask specific questions. “Show me all the APIs” gets a long, unfocused list. “Show me all the APIs owned by the payments team that expose a webhook endpoint” gets a useful answer. The graph rewards specificity.

Ask the graph before you ask a person. I’ve stopped writing Slack messages that start with “does anyone know who owns…”. Nine times out of ten the graph knows, and it doesn’t need to check its calendar.

Save the useful prompts. I keep a running note of prompts that worked, so I can rerun them across workspaces. “Find orphaned collections” is a prompt I run monthly. “Compare spec to implementation” is one I run before every release.

Verify the surprising answers. If the AI Engineer says something you don’t expect (a service you didn’t know called your API, an ownership that doesn’t match your mental model), open the graph view and check the edges yourself. Most of the time the agent is right and your mental model is stale. Occasionally the graph has a gap. Either way you learn something.

Connect your repos. The graph gets much richer when it can see the backend and frontend code that actually calls your APIs, not just the collections. Set up Native Git on the workspaces you care about.

Try a few prompts today

Open Postman Agent Mode in a workspace with a few APIs and collections in it, and try these three in order. They build on each other:

1. List every API in this workspace and its owning team.
2. For the top three most-called endpoints, list every consumer
   grouped by owning team.
3. Are any of those endpoints missing an OpenAPI spec, or drifting
   from their spec? If so, describe the drift.

By the third answer you’ll have a working map of what’s here, who uses what, and where the specs have quietly diverged from the code. That’s a solid two hours of investigation compressed into a few prompts.

Where this changes the work

I used to spend a real chunk of every week piecing together things about our own systems by asking around. Some of that is a healthy part of team communication, and it isn’t going away. But the mechanical parts (who owns this, who calls this, is this a duplicate, is this drifting) shouldn’t require a person. Those are graph queries. Now they are.

If you want the fastest introduction, pick one thing you’ve been avoiding investigating and give the AI Engineer a specific prompt about it. The answer usually contains something you didn’t know.

Resources

• Introducing the AI Engineer on the Postman blog
• Managing downstream dependencies with the AI Engineer
• Why context beats code alone for spec drift
• Postman Agent Mode overview in Postman Docs
• Native Git in Postman Docs
• About Postman AI features in Postman Docs
• Postman Collections overview in Postman Docs
• OpenAPI specification

The post Querying the Postman Context Graph with the AI Engineer appeared first on Postman Blog.

Deploying secure, real-time Edge AI on Raspberry Pi is now simplified using LiteRT and lightweight Gemma open models. LiteRT optimizes CPU and GPU performance, delivering fast token speeds for models like Gemma4, enabling real-time local reasoning for robotics. Developers can quickly convert, quantize, and run these models using the lightweight LiteRT CLI tool. Support for Hailo AI accelerators is also coming very soon.