The Agentic Web: Dueling Philosophies From Google and Microsoft

May 2025 is a milestone moment in AI and it’s critical that everyone (not just marketers and technologists) pays attention to the dueling perspectives Google and Microsoft laid out. 

One painted a world where AI is integrated into everyday life, with purchasing, writing, and debatably thinking decisions made on our behalf. The other outlined a world in which we build our own AI integrations and define how they interact. 

Essentially the choice is prescription or collaboration. 

I’m not going to weigh in on which one I think is right because at the end of the day, we’ll only know that for certain once both have critical mass adoption, along with empirical data. 

However, I do think it’s useful to outline the major announcements and how they fit into these two perspectives. Whether intended or not, Microsoft Build took place at the same time as Google I/O and Google Marketing Live (GML). Microsoft Activate, the virtual event for advertisers in the partner network, also took place this week. That meant the influencers, builders, and media buyers who got invited to one event effectively lost the mental bandwidth to pay attention to announcements from the other. 

Those of us attending the events virtually had the luxury of being able to see both perspectives laid out, and by extension, what Google and Microsoft thinks of its customers. 

This post is not going to outline all announcements - if you’re looking for a full breakdown, here are the links from them:

Microsoft Build

Google I/O

GML

The Agentic Web: Balancing Privacy & Utility

One of the few areas where most AI oriented companies agree is the future lies in the agentic web. Most humans scratched their head at that: what is the agentic web?

Agents are tools trained on data from a brand or person to be able to understand what that brand or person would want. Rather than searching for information and having to do all actions yourself, the agentic web would automate “grunt work” in our interactions. Think tasks like scheduling meetings, purchasing items/services at agreed upon prices, or summarizing emails and prioritizing “to-dos” from them. 

While GML and I/O painted a very ecommerce oriented picture of the agentic web, there are a lot of lead gen oriented applications. 

• Booking an appointment with a vet off of a query around a pet eating something they shouldn’t and if there isn’t an appointment fast enough, offering to book an appointment with a nearby ER. 
• Proposing a workout playlist off of a query on how to get ready to run a marathon, as well as checking if you have any active gym memberships.
• Offering a travel agenda when traveling on vacation based on interests complete with booking on your behalf if approved, or even suggesting a few restaurants that might be useful for business meals if traveling for work.. 

How useful the agentic web is relies heavily on how much information you are willing to share with your own agents and the agents of brands trying to win your business. Additionally, there’s a certain amount of trust in delegating tasks like these to a literal virtual assistant. 

Google’s outlook on the agentic web prescribes their agents integrated throughout their toolset and allowing them to all speak to one another. These agents can write emails for you, purchase items you ask it to monitor on your behalf, and many other arguably useful functions. The price: letting Google into your email and spend your money without direct oversight. 

The “Try It On” tool is objectively cool and should help brands sell more clothing to more consumers. However, when taken in the context that Google will also A/B test pricing and offers for brands who opt into that feature in Google Merchant Center next, it’s not hard to imagine a world in which Google over optimizes for its own commissions and undercuts retailer profits. Combined with the Smart Bidding Exploration, which can adjust return on ad spend (ROAS) or cost per acquisition (CPA) goals up to 30%, there’s a perfect storm of well intentioned AI on the horizon. 

The agentic web will likely push more people to put up firmer barriers between their work and personal handles. It’s not a huge ask to share professional communications with an approved AI to help with work tasks. It’s a much different question when asking people to hand over decades of personal communications to build consumer profiles. 

I’ve shared before the different experiences I get when I use my different Google handles with different privacy checks. In case you missed it, it’s linked here. I can see Google’s agentic web evolving to honor work vs personal profiles with drastically different (ideally smarter) answers and outcomes than the traditional SERP outputs. Today, the only real difference is frequency of ads - tomorrow, we might see truly different tasks and user interfaces.   

If Google’s vision of the agentic web is an AI prescription, Microsoft’s agentic perspective is rooted in a choose your own adventure mindset.

While Microsoft’s agentic web functions in a lot of the same ways as Google’s. Agents perform tasks on your behalf based on information you share. However, unlike Google, Microsoft built for privacy and work/personal profiles from the jump. 

All agents have the ability to adjust access permissions and there’s empathy for encryption that just isn’t available in Google’s tools yet. Additionally, Microsoft is putting the power in users’ hands, letting us build our own agents to interact in ways we get on board with. 

It’s also worth acknowledging how much Microsoft learned from Microsoft Clarity (the behavioral analytics tool). In case you’re unfamiliar with it, Microsoft Clarity is a free behavioral analytics tool. 

Over one million websites have Microsoft Code installed with the express goal of capturing how users interact on those sites. Imagine what Microsoft will be able to build and empower us to build with multiple AI companies hosting their platforms with Azure…

There’s no forcing of any particular vision of AI on consumers or companies. Simply offering the building blocks and in the process building themselves into every possible outcome. 

The core problem with how Microsoft announced this objectively more interesting and privacy compliant perspective on the agentic web is that it didn’t make it clear how consumers could come along for the journey. Here are the updates that are actually interesting and useful to non-technical folks:

AI Coworkers, Not Just Chatbots: Smarter Teams Agents

In Microsoft Teams, Copilot is getting agents that can remember previous conversations, share context, and work together. Think of them as virtual coworkers collaborating with you.

For marketers, these agents could surface campaign insights or manage routine reporting. For business owners, multiple agents could help untangle operational complexity, from HR onboarding to finance workflows.

The value here isn’t just novelty, it’s scalability. AI that learns once and applies that learning across multiple interactions is exponentially more useful than chatbots stuck in a first-date loop of introductions.

Build-Your-Own Copilots, Without Steep Learning Curves

Microsoft’s Copilot Studio now includes simplified toolkits and pre-built templates for creating custom AI agents across apps and websites. Full-stack dev team not required! 

This democratizes AI development, which is a win. For small teams, it means faster launches and reduced reliance on external dev shops. For marketers, it’s a path to customized chatbots that speak to campaign goals. The catch? As with any "easy" tool, the quality of the output depends on how well you know what you need it to do.

So while the barrier to entry is lower, the strategic bar remains high. Smart inputs will still separate the savvy from the sloppy.

Microsoft 365 Copilot Gets a Brain Boost

The Microsoft 365 Copilot suite is stepping into research and reasoning with new agents like Researcher and Analyst. These AI teammates can now analyze documents, summarize information, and even generate content or imagery based on prompts.

For content marketers, data visualization no longer needs to be a time intensive process. For analysts, it's instant data triage.

A word of caution: while these features enhance productivity, they also amplify whatever bias or blind spots exist in the source material. Human oversight isn’t optional; it’s more important than ever.

Copilot in Outlook: AI That Actually Preps You

Managing inboxes and calendars isn’t fun, and Microsoft is offering a very similar inbox management as Google. The key difference: agent level permissions that respect encryption and different access levels. Copilot in Outlook now summarizes threads, highlights attachments, and auto-preps you for meetings with relevant docs.

For marketers in client services, this means less time digging and more time delivering. For owners running lean, it’s a back-office assistant that’s actually helpful.

The operational relief is real, but needs to be managed appropriately.

No-Code AI: Power Platform Gets Even More Accessible

The Power Platform now supports natural language commands to build apps and data models. Specify what you want to achieve, and the tool generates functional components.

This shift is monumental for under-resourced teams. It empowers non-technical folks to prototype solutions without gatekeeping from IT. For marketers, that’s quick landing pages. For SMBs, it’s internal tools that actually reflect how your business runs.

The risk? Over-reliance on default logic and templates. As always, educated human oversight is critical.

AI Agents on Your Website: Real-Time Customer Help

Power Pages now lets you embed AI agents directly into your site or portal. These bots don’t just chat, they perform actions like updating orders or checking accounts.

The obvious win is 24/7 intelligent support without increasing headcount. However, the real magic is in conversational UX and data collection. You’ll be able to collect much more specific feedback about where your customers are getting stuck and trends in how to help them. 

For marketers, this could become a stealth conversion engine. For ecomm, it’s cart recovery in real time.

NLWeb: Turning the Web Into a Conversation

Microsoft’s introduction of NLWeb represents a major shift: a standard for embedding natural language AI interfaces into websites.

This could shift user behavior from “click and search” to “ask and receive.” Marketers would be wise to start thinking about content as dialogue, not just copy. What questions should your homepage be answering? How can you design for curiosity?.

AI That Lives on Your PC: Windows AI Foundry

Microsoft is rolling out a unified AI layer across Windows 11 that lets apps run AI models locally. That means faster performance, offline functionality, and enhanced privacy (translation: no constant pinging to the cloud).

This is a sleeper hit. For marketers building creative tools, it means fewer lags. For business owners with sensitive data, it means control. And for developers? Local dev environments just got a whole lot smarter.

GitHub Copilot Grows Up: Autonomous Coding Agents

GitHub’s Copilot is now more than just a code autocomplete. In preview, it can fix bugs, refactor legacy code, and add new features on its own. I admit this was where I got the most excited as a non-coder. 

For dev teams, this is a huge step toward AI pair programming at scale. For marketers managing websites or technical SEO, it could mean fewer bottlenecks and faster turnarounds.

It’s also a powerful reminder: if you're not exploring how AI can assist your workflows, you're leaving efficiency (and money) on the table.

Which agentic web perspective will win out? I’d like to believe the Microsoft outlook will win hearts and minds because it is focused on keeping minds critical and hearts open to different perspectives. However, there’s no denying humans, at our core, are lazy. As such, Google’s easy to understand and opt into tools may get the edge. 

Time will tell if humans choose choice or prescription.  

Creative Appropriation: Leaning Into AI For Complete Work Vs Formatting

I am married to an artist (he has pieces in museums, has done light projection, and currently works for an ornament company). Part of how we keep a happy relationship is that I don’t bother him for commissions because he was so undervalued early in his career. Instead I pay anywhere from $300-$800 for original pieces.  These pieces are born of partnership between me (the client with an idea) and the artist (who has a style I was drawn to). There’s a beautiful friendship that can develop between artist and client.

You can imagine I was not thrilled to see how heavily I/O leaned into taking the human out of art. From fully generated shorts from Veo 3, to the complete disregard for rights of artists (actors, musicians, and beyond), this feels like an apocalyptic breach of what makes a human human. Microsoft wasn’t much better with their ClipChamp tool (though it was positioned more as an editing assist).  

Yet at the same time, both Microsoft and Google launched a slew of incredibly helpful editing and formatting tools. Additionally, Google launched SynthID, which will help identify content created in part or in full with AI. 

If you’re going to use these tools for formatting and turning photos into videos, and other operational tasks, there’s less of a moral question than if you’re going to use fully generative content for commercial purposes. 

I’m excited about the ways both Google and Microsoft are enabling marketers to better reach their people with easier access to video. GML’s announcements highlighted how easy it will be for video creation (especially for shorts) as well as content creator hubs. Microsoft Build and Activate highlighted improved content formatting tools and video resources for Audience Ads and PMax. 

Much like with the agentic web, a lot of the great insights got hidden on the Microsoft side. Here are the creative tools I found interesting: 

1. Microsoft 365 Copilot “Create”: Instant Image Generation

What it is: Tucked inside the Microsoft 365 Copilot experience (in apps like Word, PowerPoint, and the web), the new “Create” function allows users to generate custom images simply by describing them. Powered by OpenAI’s vision tech (GPT-4), you can now type “a logo of a dancing robot” and get a visual ready to drop into your pitch deck, blog, or social post—no designer needed.

Where it lives: If you have Microsoft 365 Copilot, this feature is now part of the ecosystem—accessible within Word, PowerPoint, and the Copilot hub.

Why it matters: Small business owners, in-house marketers, and solo operators often don’t have the time or budget to spin up original graphics. This tool lets them skip the stock photo scavenger hunt and get on-brand visuals instantly. It also supports faster iteration during brainstorms and client reviews.

While the image quality is impressive for quick content, it’s not a replacement for strategic creative thinking. Use it to prototype, not to autopilot your entire brand presence.

2. Clipchamp Copilot: Taking the Stress Out of Creative Editing

What it is: Copilot is now embedded in Clipchamp, Microsoft’s video editing tool, and it’s capable of turning a simple prompt into a full video. Think: script writing, AI voiceover, background music, stock footage, transitions...everything tied to the editorial process..

Where to find it: This lives in the Clipchamp app for commercial Microsoft 365 users with a Copilot license. Access it through the Copilot panel or via the “Visual Creator” in Microsoft 365.

Why it matters: For marketers producing short-form video content (ads, explainers, social clips), this can significantly speed up the production process. Instead of sourcing assets and scripting from scratch, you can prompt Copilot with something like “Create a 30-second video about our summer sale” and get a first draft in minutes.

Let’s be clear, though: Fully AI-generated videos are not a creative solution in and of themselves. They’re a starting point. I don’t condone publishing AI-generated video content without a human shaping the narrative, making final edits, and ensuring everything aligns with brand voice, values, and legal requirements.

Use this tool for what it’s good at: jumpstarting production and unblocking ideation.Never hand over the full reins. Your audience deserves better than content that feels generic, disconnected, or off-brand.

3. GPT-Image-1: Microsoft + OpenAI's Enterprise-Grade Image Generator

What it is: This new model, GPT-Image-1, brings high-quality image generation into the enterprise world via Azure. It supports advanced capabilities like inpainting (editing parts of an image via text), image-to-image transformations, and embedded text (ideal for ad banners, posters, or social graphics).

Where to access it: Through Azure OpenAI Services (currently in gated preview). Developers can use the Azure Playground to test it, or integrate it into apps via API.

Why it matters: This isn’t just about making one-off visuals. If you're building scalable content ops (ecomm images on demand, dynamic ad creatives, or personalized visual experiences) GPT-Image-1 gives you API-level control with enterprise security baked in.

This is where the real creative automation potential kicks in: integrating AI generation into your CMS, CRM, or ad workflow, enabling content to scale without sacrificing relevance or brand standards.

4. Sora: Text-to-Video for Developers, With Real Marketing Impact

What it is: OpenAI’s Sora model, now available via Azure, is pushing generative video into new territory. Unlike earlier models that produced jittery 4-second clips, Sora creates realistic, coherent short videos from detailed prompts (and optionally images or clips).

Where to access it: Coming soon via Azure OpenAI Services (preview mode), with a Video Playground interface for experimentation.

Why it matters: While this is still early-access and developer-focused, Sora’s potential is massive. Imagine dynamic ad videos that respond to user context or seasonality, or product demo clips created from a text brief instead of a shoot.

It’s also a game-changer for education, training, and B2B content. Instead of spending weeks on explainer videos, teams can generate first drafts in minutes, and then refine them.

Strategic takeaway: Keep an eye on this for long-tail content use cases. It’s not about replacing video teams, it’s about letting them move faster and focus on storytelling.

5. Copilot in Microsoft Advertising: Creative That Actually Gets Your Brand

What it is: Microsoft Advertising now includes a Copilot that helps advertisers create campaigns from start to finish: copy, image, video, and all. Copilot-enabled “Brand Kit” enforces your brand’s fonts, logos, and colors across all assets.

Where to use it: Directly in the Microsoft Ads platform (and through the API for partners/agencies). When building an ad, you can now chat with Copilot or click “Generate” to auto-create assets in-line.

Why it matters: This is where the rubber meets the road for PPC marketers. Generative tools are no longer “nice to have”—they're embedded in campaign creation. For under-resourced teams, this means you can go from idea to launch without bottlenecks. For brand stewards, the Brand Kit ensures that speed doesn’t come at the cost of consistency.

Copilot can:

• Generate on-brand images and banners
• Assemble short video ad variants from stock assets
• Remove or create backgrounds to fit brand mood boards

These tools are accessible even to non-designers. Small business owners or account managers can produce high-quality visuals without needing to outsource.

Both Google and Microsoft offer us generative creative temptation and debatably more responsible formatting options. There isn’t a clear winner in the tools announced, but I will say it seems Google’s sights are on Meta and TikTok’s creator base and ad revenue, while Microsoft seems to be more focused on reinforcing its enterprise level tools.

Consumer Vs Enterprise Bets: Will The Toys Get Played With?

It doesn’t matter how cool a thing might seem - if the adoption isn’t there, it will be tough to defend development resources. There was an over two hour wait to play with Google’s Asta powered glasses. “Try It On” became a viral sensation. 

Microsoft’s announcements just aren’t geared towards consumers without a translation. The unapologetically enterprise showing was incredibly cool and useful for developers and IT professionals who have been forced to walk a tight rope between innovation and privacy. 

Based on last year’s data, Microsoft’s Azure bets seem strong, but if the agentic web is the ultimate proving ground, Google might get the edge because they have more accessible tools than codeless coding and IT empowerment.  

The average consumer seems more scared than anything else of the upcoming AI innovations. Perhaps selling into enterprises who will take on the burden of convincing their respective audiences is the smarter play. 

Policy Questions: How Much Risk Tolerance Are We Comfortable With?

There’s a very large elephant in the room: the Department of Justice has ordered Google to divest from Chrome...right as Google launched a flurry of new AI tools within Chrome. That’s not a coincidence; it’s a strategy. Whether Google genuinely believes it won’t have to comply, or it’s planning to drag the process out just long enough to harvest meaningful user engagement data, the move signals confidence that regulatory friction won’t slow its product ambitions.

The question for advertisers and marketers is: are we willing to keep building sandcastles on high-tide beaches?

This tension (between short-term opportunity and long-term platform stability) has me thinking more deeply about why Google consistently excels in ecommerce, while Microsoft quietly dominates in lead gen. It comes down to two structural differences:

1. Google’s ecosystem naturally absorbs ecommerce data. Most organic ecommerce sites also run Google Ads and push product feeds, so the system has rich visibility into what matters.
2. Microsoft’s ad formats are more feed-friendly for lead gen. Think Automotive Ads, Professional Services, and vertical-based targeting, all encouraging lead gen advertisers to submit structured data (never mind LinkedIn data).

So how does this tie into policy?

The DOJ’s other key mandate is transparency: ie. why we got all the new reporting tools at GML. But here’s the kicker: Microsoft has been transparent from the jump. Search terms, placements…advertisers have had access without a court order forcing the issue.

That difference matters. Microsoft has built a platform that’s inherently more privacy-forward and lead gen friendly. Whether that finally shifts market perception and earns it more data from lead-focused brands remains to be seen.

On the flip side, Google is signaling a desire to serve lead gen better: leaning into SEO/PPC integrations and hinting at a future where non-ecommerce advertisers might start sharing more structured data. Until that data shows up and shapes better tools, it’s still just potential.

The TL;DR: We’re in a moment where platform behavior, policy mandates, and data mechanics are colliding. Advertisers need to be clear-eyed about who they’re sharing data with, how transparent those platforms are, and whether they’re building on solid ground, or just hoping the tide doesn’t roll in too fast.

Final Thoughts  

It's normal to have conflicting and complicated sentiment around AI because it represents a new era of human interaction. The best way to fight the uncertainty is with an action plan that's open to adjustments. After thinking about all of the events and both major brands' perspectives on AI, I'd suggest the following:

1. Audit your data boundaries: Decide what data you're willing to let AI agents access; personally and professionally. Once that's done, separate work and personal accounts to ensure agents return the right insights with appropriate context.
2. Test agentic workflows in low-risk areas: Use AI agents for calendar management, basic reporting, or research tasks before handing off customer-facing work. Monitor for improvements to workflows as well as unnecessary friction and loss of empathy.
3. Vet creative tools through the lens of assist, not replace: Use image and video generation tools for ideation, not finished output. Layer in brand and compliance review before publishing AI-assisted content.
4. Prioritize platforms that align with your risk tolerance: If transparency and control matter, Microsoft may offer more guardrails. If speed and integrated consumer touchpoints matter more, Google’s tools may offer more immediate wins, but with tradeoffs.
5. Start building internal policy now: Don’t wait for regulators. Define your own AI usage standards: what’s allowed, what needs review, and what stays human. Communicate this to stakeholders so expectations and accountability scale with your tooling.

I hope everyone will embrace AI to a certain extent because there are some real innovations and empowerment opportunities. However it's critical to own the real negative impact AI can have in a world where we don't all agree on what is true. Remember AI can hallucinate and the content you are exposed to will influence your thoughts which become words and actions.