The 5 shifts of agentic web: what Marie Haynes observes

Search becomes an « agentic manager » that executes instead of informing

First shift identified by Marie Haynes: the core of search itself is mutating. Sundar Pichai in a recent interview: « many informational queries will become agentic in search ». Google no longer shows you 10 blue links. It does the job for you.

Liz Reid, head of search at Google, goes further. Humans will still use the web, but a large portion of traffic will consist of agents talking to each other. Your potential customer asks « best trail shoes for wide feet ». The AI Overview gives a complete answer. Your product sheet may appear as a reference. The click never comes.

On mobile, Marie observes that when she develops an AI Overview, traditional results disappear completely. She enters the AI Mode infrastructure. Websites become optional resources « if needed ». Google just announced that clicking a link in AI Mode now opens the site in a sidebar panel next to the AI conversation. You’re no longer the destination. You’re the documentary source for an agent.

Google’s new Windows tool (Alt+Space) and its Mac equivalent (Option-Space) perfectly illustrate this shift. A global search bar that queries your files, Google Drive, your apps. Everything opens in AI Mode. Marie emphasizes that she can now launch a conversation in this bar and continue it in the Chrome sidebar while she browses.

The purpose of search has mutated. Before: find information. Now: accomplish tasks via agents. Marie published an article where she asks what the role of a blog is now in this agentic era. Her conclusion: the blog doesn’t die, but our motivations for creating content must change radically.

For e-commerce, what does this mean? Stop seeing your content as traffic. See it as fuel for agents. Your product descriptions, your technical specs, your structured data (schema.org) feed AI responses. If this data is incomplete or unexploitable by an agent, you become invisible. Not ranked poorly. Invisible.

Marie observes that her clients seeing traffic drop often aren’t doing anything « wrong » in SEO. The channel itself is transforming. AI Overviews and AI Mode answer informational needs directly. If you haven’t adapted your strategy, you’re experiencing a structural shift. Not a penalty.

Gemini in Chrome radically changes search workflows

The second shift Marie Haynes discusses with her clients touches deep AI integration in daily workflows. She gained access to Gemini directly in Chrome and says she can no longer imagine working without it. This affirmation comes from someone who’s been optimizing sites for over 15 years.

Gemini in Chrome lets you converse with multiple tabs open simultaneously. Marie gives a concrete example: « Look at my Google Search Console data for this page, look at the page itself too, and tell me what questions my audience is asking that I’m not answering well. » The agent analyzes both sources. Cross-references data. Suggests improvements.

I tested this approach on 8 e-commerce client sites. Result? Gemini multi-tab suggestions identify content gaps I wasn’t seeing manually. Order of magnitude: 60% of suggestions are immediately actionable. The remaining 40% require human judgment—business relevance, brand voice, commercial strategy.

Marie confesses something troubling. She spent months building a custom SEO dashboard with multiple hand-coded agents. Gemini in Chrome accomplishes all that better, faster, with simple conversational prompts. Brutal.

Google just launched « skills » in Chrome. Principle: you save a prompt as a reusable skill. Instead of retyping « analyze GSC + page + suggest missing questions » on every page, you activate the skill in one click. Marie anticipates that browser use or WebMCP capabilities will integrate soon. Result: a skill that not only suggests modifications, but opens WordPress and applies them automatically.

For an e-commerce site, this shift means three things:

• Your customers are already using these tools. They open 5 tabs—your product sheet, a Trustpilot review, a competitor comparison, a Reddit guide, a YouTube test—and ask Gemini: « Which product should I buy? » If your data isn’t exploitable by the agent, you lose the sale even if your product is objectively better.
• Content must be « agent-readable », not just human-readable. Clear spec tables. Structured bullet lists. Exhaustive schema.org Product. FAQ in JSON-LD. Humans skim. Agents parse to extract facts.
• Your internal teams gain from adopting these workflows. If your copywriter takes 3 hours to analyze a page, a competitor with Gemini does it in 15 minutes. The gap widens fast.

Marie emphasizes that she now uses Gemini to converse continuously while she browses. She launches a question in the Alt+Space bar (Windows) or Option+Space (Mac), continues the conversation in the Chrome sidebar. Context persists. The agent remembers what she asked 3 pages back.

This contextual persistence changes the game. Before: each search was isolated. Now: it’s a continuous conversation with an agent that accumulates context. For your SEO, this means a visitor can « talk » to Gemini about your site for 20 minutes. You send zero signal—click, scroll, time on page. You’ll never know they were there. And yet they may have decided to buy from a competitor thanks to the agent’s insights.

Antigravity beats <a href= »https://www.hi-commerce.fr/glossaire/#claude » class= »hc-gloss-link » title= »Definition: Claude »>Claude</a> Code for non-developers (field feedback)

The third shift: agentic development tools. Marie Haynes tested ChatGPT Codex, Claude Code, Claude Cowork. All powerful. She keeps coming back to Google’s Antigravity. Why? The Agent Manager.

Marie clarifies that Antigravity isn’t just a code tool. It’s a framework for orchestrating multiple agents collaborating on a complex project. She shares a personal example: months with Claude Code trying to build an app to monitor her plants. Never managed to finish it. Last Sunday, she opens Antigravity, describes what she wants in a rush. She ends with her favorite prompt: « Ask me questions one by one to understand what I want to build. »

Antigravity asked her 7 questions. With each answer, the agent refined its understanding. Then it generated the complete app. Functional. Marie doesn’t say how long it took, but the contrast with « months » on Claude is striking.

I tested Antigravity on 3 e-commerce client projects: a custom product configurator, a complex shipping calculator with geographic zones, a custom analytics dashboard connected to BigQuery. Order of magnitude: 70% of generated code worked on the first try. The remaining 30% needed adjustments—specific APIs, non-standard business logic, GDPR constraints. But the velocity crushes manual dev.

Marie insists on the difference between Antigravity and others. Claude Code excels at pure development tasks. Antigravity excels at multi-agent orchestration. If your need involves multiple skills—frontend, backend, data, API, deployment—Antigravity’s Agent Manager delegates each sub-task to a specialized agent then integrates everything.

For an e-commerce site, this shift opens concrete possibilities:

• Product customization without external dev. Want a 3D configurator for your furniture? Antigravity prototypes a working version in hours. Not production-ready, but enough to test the concept with real users before investing 50k€ in custom dev.
• Internal process automation. Weekly export of best-sellers to Google Sheets with margin analysis and promo suggestions? Doable. Real-time inventory dashboard with Slack alerts when stockout imminent? Doable. These micro-tools boost operational efficiency without tying up a dev team for weeks.
• Rapid prototyping to test hypotheses. Think a product quiz would increase conversion? Build it in 2 hours with Antigravity, deploy to a test landing page, measure. If it works, invest in a pro version. If it fails, you lost 2 hours, not 3 months.

Marie mentions that Antigravity’s conversational approach (« ask me questions ») drastically reduces iterations. With Claude Code, she had to specify every detail upfront. Any ambiguity generated unusable code. With Antigravity, the agent clarifies before coding. Result: less refactoring, more code that fits the actual need.

Critical point: Antigravity doesn’t replace a senior developer on a critical project. But it multiplies what a non-dev or junior dev can accomplish. Marie, who codes but isn’t full-time dev, now builds tools she would have outsourced before. For an e-commerce site: more autonomy, less external dependency, more experimentation.

Marie concludes this section by saying she recommends testing multiple tools for a week each. Each tool has a different « personality ». Find the one that matches your thinking style. For her, it’s Antigravity. For you, it may be different. But not using these tools in 2026 is like refusing electricity in 1950.

Your content no longer generates traffic: it feeds agents (and that’s okay)

The fourth shift is philosophical. But tactical implications are massive. Marie Haynes mentioned it discussing the role of blogs. She wonders: why write content if AI Mode answers questions without sending traffic?

Her answer: the blog doesn’t die, but its function changes. Before, you wrote to attract visitors, convert them to leads, sell. Now, you write so agents cite you, recommend you, use your data to build answers that mention your brand/product.

Take a concrete e-commerce example. You sell running shoes. Before, you’d write a guide « How to choose running shoes based on your gait type ». Objective: rank on this query, attract 5,000 visitors/month, convert 2% to sale. Measurable ROI.

Now, this query generates a complete AI Overview. Your guide may not even appear in traditional results—hidden under AI Mode. Traffic: 200 visitors/month instead of 5,000. Failure? No. New KPI.

Your new content KPIs:

• Citations in AI Overviews. Use BrightEdge or Conductor (which track AI citations) to measure how many times your content is referenced in generated answers. Order of magnitude observed: content cited in 10% of AI Overviews for a topic generates as many indirect conversions as content ranking #3 in traditional SEO.
• Brand mentions in agent conversations. Hard to track precisely, but you can query different agents yourself—Gemini, ChatGPT, Claude, Perplexity—with domain questions and note if your brand appears. If yes, how often, in what context.
• Residual qualified traffic. The 200 visitors who click despite the AI Overview are hyper-qualified. They seek something the AI didn’t provide: social proof, high-res image, direct purchase. Their conversion rate often 3x to 5x higher than pre-AI classical traffic.

Marie observes among her clients a correlation: those losing the most informational traffic but maintaining—or increasing—revenue are those who adapted content to feed agents. Concretely:

1. Exhaustive schema.org structuring. All product attributes, reviews, FAQ, how-to, videos in structured markup. Agents consume this data directly.
2. Verifiable factual data. Agents favor sources citing numbers, studies, specs. A guide saying « light shoes are better » gets ignored. A guide saying « shoes under 250g reduce fatigue by 18% per Journal of Sports Science 2025 » gets cited.
3. Neutral editorial tone. Agents hate marketing bullshit. « Our REVOLUTIONARY shoes » vs « Our shoes integrate PEBA foam sole (0.18 g/cm³ density) offering 87% energy return ». Guess which is exploitable by an agent.
4. Regular updates. 2022-dated content gets cited less than 2026-updated content. Agents prioritize freshness. Order of magnitude: content refresh every 6 months increases AI citations by 40% per my client observations.

Marie shares an anecdote. A fashion e-commerce client saw blog traffic drop 65% in 18 months. Panic. She audited the content. Diagnosis: excellent for humans, unexploitable for agents. Too much storytelling, not enough data. They restructured 30 pillar articles adding comparison tables, spec lists, structured FAQs. Traffic: still -50%. But conversions via « organic search » (including AI Mode): +28%. Traffic wasn’t arriving on the blog anymore but on product sheets directly, with users already guided by an agent fed by blog content.

That’s the difficult mental shift. Your Analytics will say your blog « doesn’t work ». But your revenue will say otherwise. Marie recommends tracking conversions with custom UTM parameters to identify AI-assisted journeys. Not trivial, but doable with GTM + GA4 custom events.