Happy Wednesday.

This week felt different.

Not because a new model scored higher on a benchmark.

Not because another company announced a chatbot.

Different because some of the biggest players in AI quietly revealed what comes next.

And it looks a lot less like "asking ChatGPT questions" and a lot more like giving AI a permanent desk in your company.

For the past few years, we've treated AI like an employee who only works when spoken to.

You ask.

It answers.

You prompt.

It responds.

You leave.

It waits.

Microsoft just proposed a completely different future.

One where AI never stops working.

One where your digital assistant is preparing for meetings while you sleep, organizing your tasks while you're on vacation, and surfacing opportunities before you even realize they exist.

The era of reactive AI may already be ending.

The era of agentic computing has officially begun.

At its Build 2026 conference, Microsoft stepped out from OpenAI's shadow and delivered one of the most ambitious AI visions we've seen yet.

For years Microsoft benefited enormously from its OpenAI partnership.

This week they made it clear they're building their own future.

The centerpiece was a new AI agent called Scout.

Unlike traditional assistants, Scout doesn't sit around waiting for instructions.

Built on Microsoft's OpenClaw technology, Scout continuously monitors your Microsoft 365 environment, proactively preparing meeting materials, identifying action items, organizing tasks, and helping coordinate work before you ever type a prompt.

Think less chatbot.

Think digital chief of staff.

Microsoft wasn't done there.

They also announced:

The strategy couldn't be clearer.

Microsoft wants Windows and Microsoft 365 to become the operating system for AI agents.

A place where agents:

This isn't an Office strategy.

It's an infrastructure strategy.

The company that owns the environment where agents work could become the next dominant computing platform.

Not surprisingly, OpenAI has no intention of letting Microsoft own that future.

This week OpenAI expanded Codex far beyond coding.

The company launched six role-specific plugin collections spanning:

But the most interesting launch may have been Codex Sites.

Imagine describing a website in plain English and receiving a live, deployable application complete with a shareable URL.

No coding.

No hosting setup.

No technical expertise.

Just an idea and a prompt.

For small businesses, startups, consultants, agencies, and entrepreneurs, this removes yet another layer between concept and execution.

The future increasingly looks like this:

Have an idea.

Describe it.

Deploy it.

Of course, there's a catch.

Actually, a very expensive catch.

While everyone is celebrating AI productivity gains, companies are discovering something finance teams have been quietly worrying about for months.

AI usage isn't cheap.

Uber became this week's cautionary tale.

The company reportedly burned through its entire projected 2026 AI budget in just four months.

The result?

A hard cap of $1,500 per month per employee per AI tool.

Let that sink in.

The same companies that encouraged employees to use AI as much as possible are now discovering what happens when usage scales.

Some organizations even gamified adoption.

Uber reportedly created internal leaderboards ranking teams based on AI usage.

Employees were rewarded for consuming more AI resources.

The budget consequences arrived shortly afterward.

Even worse, many companies still struggle to answer a simple question:

Did all those AI tokens actually create measurable business value?

Uber CEO Dara Khosrowshahi says AI agents now write roughly 10% of Uber's code.

That's impressive.

But executives are still trying to determine whether spending millions on AI usage actually improves outcomes in ways customers notice.

The AI adoption phase is ending.

The AI accountability phase is beginning.

Another surprise emerged this week.

Developers who had become accustomed to predictable AI subscription costs suddenly discovered the downside of usage-based billing.

GitHub Copilot's shift toward token-based pricing created reports of monthly costs jumping from approximately $29 to several hundred dollars, and in some cases much higher.

Predictably, the internet responded exactly as the internet always does.

Reddit exploded.

Twitter complained.

Memes were created.

The phrase "what a joke" appeared repeatedly across developer communities.

The reality is simple.

AI usage is becoming more like cloud computing.

The more power you consume, the more you pay.

That may be economically rational.

It just isn't emotionally pleasant.

Another major trend accelerated this week.

Human authentication.

For years CAPTCHAs served as the internet's gatekeepers.

Click the bicycle.

Select the traffic light.

Find the crosswalk.

Problem solved.

Except now AI can solve many of those challenges too.

AI can generate convincing faces.

Convincing documents.

Convincing identities.

Convincing social media profiles.

The result is an emerging crisis of trust.

That's where World ID enters the conversation.

The project from Tools for Humanity aims to verify human identity using biometric scanning technology.

The concept sounds futuristic.

It may soon become necessary.

As bots become indistinguishable from humans, proving you're real could become one of the most important pieces of digital infrastructure built over the next decade.

While most attention focused on product launches, policy shifted too.

The Trump administration signaled a softer approach toward AI oversight.

Rather than relying heavily on executive orders and regulatory restrictions, the new framework emphasizes voluntary industry participation while maintaining national security reviews for frontier models.

The approach appears designed to balance innovation with safety.

Whether it succeeds remains an open question.

But the message to AI companies is clear:

Move fast.

Just don't break national security.

One of the more uncomfortable developments this week came from GitLab.

The company announced a 14% workforce reduction while simultaneously discussing preparation for an "agentic future."

This creates an increasingly difficult narrative for the AI industry.

Publicly, many leaders continue emphasizing augmentation rather than replacement.

Privately, companies are redesigning workflows around AI capabilities.

Meanwhile, Bernie Sanders proposed a radical idea:

If AI companies create unprecedented wealth using data generated by society, should the public own part of that value?

His proposal suggested Americans receive ownership stakes in major AI companies.

Whether realistic or not, it highlights a growing debate.

Who benefits from the AI economy?

And how should that value be distributed?

Lost amid the headlines was another important development.

Claude continues evolving from chatbot to operating environment.

New features now include:

Increasingly, Claude is becoming less of a conversation tool and more of a digital workspace.

And that's exactly where the broader market appears headed.

The biggest lesson from this week isn't about Microsoft.

Or OpenAI.

Or Anthropic.

It's about the direction of computing itself.

We're moving from:

AI as a tool

to

AI as an environment.

The winners of the next decade may not be the companies with the smartest chatbot.

They may be the companies that create the place where billions of AI agents live, work, collaborate, and make decisions.

That's a much bigger opportunity.

And a much bigger battle.

Most marketers use ChatGPT to do general research and then call it an AI strategy. The ones outperforming them are deploying specialized agents built for specific jobs.

We put together 22 plug-and-play ChatGPT marketing agents that handle the work eating your week, each with built-in instructions and structured outputs ready to go in under 5 minutes.

Subscribe to Marketing Against the Grain and get all 22 free.

Inside you'll find:

Your competitors are already running agents like these.

Get 22 ChatGPT Marketing Agents free when you subscribe to Marketing Against the Grain today.

Get The Guide

Microsoft's Scout signals a shift from reactive assistants to proactive AI agents that continuously work in the background.

Organizations are discovering that AI adoption without spending controls can destroy budgets much faster than expected.

As AI-generated content becomes indistinguishable from reality, proving human identity online is becoming mission-critical.

The next winners won't simply build AI tools. They'll build the environments where AI agents operate.

Companies are increasingly restructuring around agentic workflows even as public messaging remains focused on augmentation.

Claude continues expanding beyond chat with Skills, Cowork mode, workspace connectors, advanced analytics capabilities, and desktop organization features. One of the most compelling capabilities is its ability to create reusable workflows that improve over time through evaluation loops and memory.

Explore Microsoft's latest vision for agent-first computing, including Scout, MAI models, agent infrastructure, and future workplace automation concepts.

Codex is rapidly evolving into a workplace platform with specialized workflows, integrations, and the ability to generate deployable applications from prompts.

A useful extension for monitoring Claude token usage, remaining capacity, and context consumption in real time. Particularly valuable as AI usage costs become more important to manage.

An all-in-one AI workspace combining writing, research, design, image generation, video creation, and multiple model access under one interface.

An open-source AI desktop agent capable of managing tasks across Slack, email, messaging apps, and other workplace tools with persistent memory.

An AI-powered career assistant that helps identify opportunities, customize applications, optimize resumes, and automate portions of the job search process.

Prompt:

Prompt:

Prompt:

Prompt:

Prompt:

Prompt:

A few years ago we worried AI would become smarter than us.

This week, we learned something arguably more interesting.

AI is becoming organized.

It now has calendars.

Workflows.

Budgets.

Authentication systems.

Dedicated hardware.

And apparently its own expense reports.

At this rate, the first AI agent to submit a reimbursement request for excessive token usage may be only a few software updates away.

When that happens, somewhere in a finance department, a CFO is going to discover that the robot not only spent the money...

...it also approved its own purchase order.

🧠 If you enjoyed tonight’s deep dive, forward it to someone in your network who wants to fully grasp AI in 5 minutes per day. They’ll thank you later.

Your slightly self-deprecating, definitely human narrators,
Anicia & Shane