Evidence for the Agentic Shift
The field guide states propositions. This page holds the observable ground beneath them: what AI systems demonstrably do today, what has been announced and is arriving, and what credible sources project. The two are kept deliberately separate, because a discipline that cannot tell its claims from its evidence is not a discipline.
Sources are primary wherever possible: official product documentation and announcements first, institutional research second, high-quality journalism third. Every entry below was verified against the live source. Each carries an honest status: observable in shipped products, emerging, or a forecast attributed to whoever made it.
01 · AI mediates information consumption
Machines increasingly read the web so people do not have to: summarising, synthesising, and collapsing many sources into one answer.
- AI in Search: Going beyond information to intelligence · Google, May 20, 2025. [Observable now] AI Mode rolled out to all US users; Google reports AI Overviews driving over 10% more usage on the query types where they appear.
- Google users are less likely to click on links when an AI summary appears in the results · Pew Research Center, July 22, 2025. [Observable now] With an AI summary present, users clicked a result on 8% of visits versus 15% without; they clicked a source inside the summary on 1% of such visits.
- Introducing ChatGPT search · OpenAI, October 31, 2024. [Observable now] Web-sourced, linked answers inside the assistant for queries that previously required a search engine; opened to everyone in February 2025.
- Gartner Predicts Search Engine Volume Will Drop 25% by 2026, Due to AI Chatbots and Other Virtual Agents · Gartner, February 19, 2024. [Forecast] Gartner's projection that generative assistants become substitute answer engines, cutting traditional search volume by a quarter.
02 · AI infers, compares, and recommends
The systems do not stop at answering. They compare options, infer needs, and produce recommendation sets: the mechanics beneath the Shortlist Effect.
- How customers are making more informed shopping decisions with Rufus, Amazon's generative AI-powered shopping assistant · Amazon, July 12, 2024. [Observable now] Live for all US customers; tens of millions of questions asked, explicitly for comparing options and getting recommendations.
- Shop like a pro: Perplexity's new AI-powered shopping assistant · Perplexity, November 18, 2024. [Observable now] AI-selected product cards with one-click checkout; a merchant programme in which sharing structured product data raises the chance of being recommended. The shortlist mechanics, stated by the platform itself.
- Shop with AI Mode, use AI to buy and try clothes on yourself virtually · Google, May 20, 2025. [Emerging] Gemini narrowing 50-billion-plus product listings into recommendation sets, with agentic checkout announced to buy on the shopper’s behalf.
- Introducing shopping research in ChatGPT · OpenAI, November 24, 2025. [Observable now] A model trained specifically for shopping tasks researches across the web and delivers a personalised buyer’s guide with comparisons and trade-offs, on all plans.
03 · AI acts on delegated tasks
From reading to doing: the origin points of agents that use browsers and computers, complete forms, book, and buy, under human supervision. The Copenhagen errand is this section, lived. Where it has gone since is the next two sections.
- Introducing computer use, a new Claude 3.5 Sonnet, and Claude 3.5 Haiku · Anthropic, October 22, 2024. [Observable now] A frontier model operating a computer the way people do: looking at the screen, moving a cursor, clicking, typing.
- Introducing Operator · OpenAI, January 23, 2025. [Observable now] An agent with its own browser that fills forms, orders groceries, and books, handing control back to the person for logins, payments, and confirmations.
- Introducing ChatGPT agent: bridging research and action · OpenAI, July 17, 2025. [Observable now] ChatGPT doing multi-step work on its own computer, with explicit confirmation before actions of consequence: supervised autonomy, shipped.
- Piloting Claude in Chrome · Anthropic, August 25, 2025. [Observable now] An agent acting inside the person’s own browser, with site-level permissions and action confirmations, extended to all paid plans by December 2025.
04 · The personal agent arrives
The 2026 state of the art: personal agents that hold a continuous voice conversation, run on your machine or their own, spawn further agents, and carry your tasks across every app you use. From frontier labs and from open source at once.
- Introducing GPT-Live · OpenAI, July 8, 2026. [Observable now] Full-duplex voice as the interaction layer over agentic work: GPT-Live delegates tasks needing search, reasoning, or agentic capability to other models mid-conversation. Over 150 million people already talk to ChatGPT weekly.
- OpenAI’s new voice mode makes it to the ChatGPT desktop app · TechCrunch, quoting OpenAI, July 24, 2026. [Observable now] OpenAI’s own line, quoted: “Control your computer and direct multiple agents running in ChatGPT Work or Codex, using just your voice.” Voice, computer operation, and multi-agent direction in one consumer surface.
- Claude Cowork is coming to mobile and web · Anthropic, July 7, 2026. [Observable now] Agentic delegation as mainstream knowledge work: hand Claude a task and it works across files, calendar, email, messaging, and the web. Anthropic reports more than 90% of Cowork use is not software development, and scheduled tasks now run with no device online.
- OpenClaw: Personal AI Assistant · OpenClaw (open source), 2026, actively released. [Observable now] “The AI that really does things”: an open-source, self-hosted personal agent that organises the inbox, sends email, manages the calendar, and checks you in for flights, from WhatsApp, Telegram, Signal, or any chat app. Your machine, your models, your data.
- Hermes Agent: The Agent That Grows With You · Nous Research (open source), 2026, actively released. [Observable now] A second major open-source personal agent: persistent learned memory, self-created skills, isolated subagents, browser automation, and natural-language scheduling, self-hosted behind a messaging gateway.
- Introducing Grok Bot · xAI, August 11, 2026. [Observable now] “Your team of always-on agents. They have their own computer, work inside tools and apps like you do, and keep working 24/7”, including across platforms with no clean API. Each agent runs its own cloud computer; in beta.
05 · The agent enters the operating system
The platform vendors are building the personal agent into the device layer itself. When the operating system ships an agent, mediation stops being an app the person chooses and becomes the default surface of computing.
- The Gemini app becomes more agentic, delivering proactive, 24/7 help · Google, May 19, 2026 (I/O). [Emerging] Gemini Spark: “a 24/7 personal AI agent designed to proactively manage tasks… an active partner that does real work on your behalf and under your direction.” Cloud-based, working even when the laptop is closed, and extending to the Mac to operate the local machine; over 900 million monthly Gemini users, with MCP connections to consumer services launching.
- What’s new in Apple Intelligence (WWDC26 guide) · Apple Developer, June 2026 (WWDC26). [Emerging] Apps connect their content and capabilities to the system agent: App Intents contributes app content to a semantic index and lets people act on it in natural language through the upgraded Siri, with third-party foundation models admitted into the system layer.
- Composing a new platform for agent-first devices (Project Solara) · Microsoft, June 2, 2026 (Build). [Emerging] “The next platform shift is from apps to agents: from software you open to intelligence you invoke.” Agent-first reference devices in private pilots with major retailers and brands, and interface generation framed on a spectrum ending in fully generative UI. Early, and explicitly said to be.
06 · The transaction layer is being built
The rails of agentic commerce: open protocols, agent payments, and machine-mediated checkout, arriving from the platforms and the payment networks at once.
- Buy it in ChatGPT: Instant Checkout and the Agentic Commerce Protocol · OpenAI, with Stripe, September 29, 2025. [Observable now] Checkout completed inside the conversation; the open-sourced Agentic Commerce Protocol lets agents, people, and businesses complete purchases together.
- Powering AI commerce with the new Agent Payments Protocol (AP2) · Google Cloud, September 16, 2025. [Emerging] An open protocol with sixty-plus organisations using cryptographically signed mandates to prove a person authorised an agent’s purchase, including delegated buying with the human not present.
- Find and Buy with AI: Visa Unveils New Era of Commerce · Visa, April 30, 2025. [Emerging] “Soon people will have AI agents browse, select, purchase and manage on their behalf.” Visa’s network, opened to agent developers, with tokenised credentials and consumer-set limits.
- Mastercard unveils Agent Pay, pioneering agentic payments technology to power commerce in the age of AI · Mastercard, April 29, 2025. [Emerging] Agentic tokens that let registered, verified agents pay on a person’s behalf, with merchant interfaces to distinguish trusted agents.
07 · Interfaces are being generated, not just visited
The pre-built page gives ground to interfaces assembled around intent, inside the agent’s own surface. The mechanics beneath Instant Rendition and the Ephemeral Interface.
- Introducing apps in ChatGPT and the new Apps SDK · OpenAI, October 6, 2025. [Observable now] Brands shipping interactive interfaces inside the assistant’s surface, to hundreds of millions of users, on an open standard built on MCP.
- Introducing A2UI: An open project for agent-driven interfaces · Google, December 15, 2025. [Emerging] An open format in which the agent generates the interface best suited to the conversation, rendered through the host’s own components.
- MCP Apps: Extending servers with interactive user interfaces · Model Context Protocol, November 21, 2025. [Emerging] A joint Anthropic–OpenAI proposal standardising how machine services deliver interactive interfaces to agent hosts: cross-vendor interface delivery through the agent layer.
- Generative UI: A rich, custom, visual interactive user experience for any prompt · Google Research, November 18, 2025. [Observable now] In production in the Gemini app and AI Mode: “an AI model generates not only content but an entire user experience… web pages, games, tools, and applications… fully customized in response to any question.” The pre-authored interface, made optional.
08 · The models learn to render the world
Rendition needs engines. Production image models now generate brand-grade visuals instantly, and world models, the research frontier behind them, generate whole coherent environments. This is the capability curve beneath Instant Rendition.
- Introducing Nano Banana Pro · Google DeepMind, November 20, 2025. [Observable now] Production multimodal image generation with brand-relevant fidelity: legible in-image text, consistency across up to five people and fourteen blended inputs, real-time search grounding, 4K output, shipped across Gemini, Search, Ads, and Workspace.
- Marble: A Multimodal World Model · World Labs (Fei-Fei Li), November 12, 2025. [Observable now] A generally available generative world model: persistent 3D environments from text, images, or video, exportable as meshes and splats, with spatial intelligence framed as “the next frontier in AI.”
- Yann LeCun’s AMI Labs raises $1.03B to build world models · TechCrunch, March 10, 2026. [Forecast] The world-model bet made institutional: a $1.03 billion seed round for LeCun’s venture building AI that “learns from reality, not just from language.” Years from product, by the article’s own account; cited as direction, never as arrival.
09 · Agents remember the person they serve
Persistent memory and preference modelling are shipping across every major assistant. This is the second input to every rendition: the agent’s standing model of its person.
- Memory and new controls for ChatGPT · OpenAI, February 13, 2024. [Observable now] Memory that references all past conversations to tailor responses, rolled out to free users by mid-2025.
- Bringing memory to Claude · Anthropic, September 11, 2025. [Observable now] Persistent, user-editable memory of projects and preferences, so context never has to be re-explained.
- Gemini introduces Personal Intelligence · Google, January 14, 2026. [Observable now] “The best assistants don’t just know the world; they know you.” Gmail, Photos, and YouTube connected as persistent personal context.
- Introducing Copilot Memory: A More Productive and Personalized AI for the Way You Work · Microsoft, July 14, 2025. [Observable now] The workplace assistant remembering preferences, working style, and recurring topics to personalise its recommendations over time.
10 · Brands are becoming machine-readable
The infrastructure through which a Brand API materialises: open protocols, structured data at web scale, and commerce feeds built to be read by agents.
- Introducing the Model Context Protocol · Anthropic, November 25, 2024. [Observable now] A universal open standard connecting AI systems to data sources and tools, replacing fragmented integrations with one protocol.
- Join the WebMCP origin trial · Chrome for Developers, June 9, 2026. [Emerging] WebMCP moves from proposal to shipping experiment: an origin trial from Chrome 149 letting web applications declare their purposes and expose functions as tools to in-browser agents. The explainer lives with the W3C Web Machine Learning group; pre-standard, and running in a real browser.
- Stripe powers Instant Checkout in ChatGPT and releases Agentic Commerce Protocol codeveloped with OpenAI · Stripe, September 29, 2025. [Observable now] Businesses must expose products, pricing, and checkout in a form agents can use, while retaining control over what is sold and how the brand shows up.
- Agentic Commerce Protocol (merchant documentation) · OpenAI Developers, current documentation. [Observable now] The production feed spec: structured catalogue data ingested so relevant products surface in context. The documentation itself ships machine-readable.
- Schema.org · Schema.org, current; vocabulary v30.0, March 2026. [Observable now] The pre-agentic baseline at web scale: over 45 million domains carrying more than 450 billion structured-data objects.
11 · The human consequences are being argued in the open
What delegation gives back, what it takes, and what should stay human. External thinking that frames the lovable half of the Law; evidence of the debate, never sources defining this field.
- The turbulent AI era is here. The choices we make now are critical. · Bill Gates, GatesNotes, August 26, 2026. [Forecast] Gates’s “Human Reserved”: domains kept for people “because the loss would be too great,” the preservation of “productive struggle,” and AI giving people back “some of the time and attention now consumed by paperwork, bureaucracy, searching.” His frame, cited as his.
- Your Brain on ChatGPT: Accumulation of Cognitive Debt when Using an AI Assistant for Essay Writing Task · MIT Media Lab (preprint, 54 participants), June 10, 2025. [Emerging] Early empirical grounding for productive struggle: LLM-assisted writers showed the weakest neural connectivity and the lowest ownership of their own work.
- Anthropic Economic Index · Anthropic, ongoing; updated June 2026. [Observable now] A live measure of where people currently draw the delegation line: per-occupation data on whether they collaborate with or delegate to the model.
None of these sources defines Agentic Branding, and none needs to. They evidence the conditions the discipline responds to: machines reading, choosing, transacting, rendering, and remembering, between people and brands. What a brand should do about it is the field guide's work.
Related: Agentic Branding, the field guide · The Shortlist Effect · The Legible-Lovable Law. Maintained by Thomas Marzano. Every source verified against the live page, 2026-08-29.
Published: 2026-08-29. Last substantive update: 2026-08-30. Author: Thomas Marzano. Maintained record.