Modern computing has undergone three major shifts, each defined by a new interface and a new economic tollbooth. In the 1980s, the personal computer put microprocessors on desks, allowing Microsoft to tax the operating system. In the 1990s, the browser connected those machines to the web, allowing Google to tax human curiosity. By the late 2000s, smartphones turned computing into glass in your pocket, handing Apple and Google control over closed mobile app stores.
Each platform shift altered the fundamental interaction primitive: typing command-line syntax gave way to clicking graphical icons, which yielded to tapping capacitive glass touchscreens.
Each transition also displaced the economic tollbooths of the preceding era. The open web made proprietary desktop software distribution secondary, turning operating systems into background plumbing for the browser. Mobile computing bypassed desktop web portals into closed app stores where Apple and Google collected 30% platform rents.
The fourth transition is underway, and its core primitive is delegation.
Instead of operating software, you describe an outcome to an autonomous system that plans, reasons, and executes across the digital world on your behalf.
Every previous platform shift changed how humans navigated interfaces. The personal agent does something more disruptive: it removes the human from the interface altogether. This creates a fascinating commercial problem for the companies that built the modern internet, because almost every digital business model is designed around monetising human impatience.
For the past 4 years, technology commentators claimed that artificial intelligence had already triggered the next platform shift. Most users remained rightfully skeptical. Opening a browser tab to ask ChatGPT for a recipe or having an algorithm draft a polite rejection email felt like using an impressive digital encyclopedia, but it hardly felt like a computing revolution.
That skepticism was generally justified - ChatGPT changed how computers generate text, leaving the underlying computing platform untouched. The user remained the manual operator. You typed a prompt into a text box, waited for words to stream across a screen, copied the result, and manually pasted it into your actual software. The human remained the runtime, bridging the gap between the model and the real world.
However, in early 2025, Anthropic introduced Claude Code, demonstrating what happens when a reasoning model is given bash access, file system permissions, and an execution loop. It stopped acting like a conversational assistant and began behaving like a software engineer: reading directories, formulating multi-step plans, editing source files, running test suites, and fixing its own compilation errors.
OpenClaw went viral in January 2026 by taking that pattern to its logical conclusion, and turning the terminal loop into a persistent personal operating system. By storing identity and working memory in version-controlled markdown files, OpenClaw gave agents durable state, scheduled cron jobs, and the ability to coordinate across secondary tasks. An agent could live on a homelab server, watch an email inbox overnight, and execute background workflows while its owner was asleep.
OpenClaw’s new year virality kicked off a race at the frontier to commercialise this exact design.
When you look at OpenAI Dots, xAI’s Grok Bot, Meta’s Muse, and Instinct, the underlying pattern is remarkably consistent: each product pairs a frontier foundation model with an always-on cloud computer, a headless browser, and a credential vault.
The primary unit of personal computing is no longer the physical device in your hand. It is an always-on virtual machine running in a datacenter.
OpenAI Dots gives every subscriber a cloud computer powered by GPT-6 Astra, capable of browsing websites, executing multi-step tasks, and connecting to thousands of external applications. Grok Bot gives agents a persistent cloud desktop with a shared terminal and browser, signing into legacy web portals and navigating software like human colleagues. Instinct packages that cloud execution loop into iMessage and WhatsApp, allowing users to delegate reservations, calendar coordination, and inbox triage through simple text messages.
Your smartphone or smartwatch is reduced to a thin remote control, a convenient surface where you state intent and occasionally approve decisions requiring human judgment.
Whenever a new computing platform appears, commentators routinely predict the total extinction of existing form factors. They claimed tablets would eliminate laptops, and they now suggest conversational agents will replace screens with ambient audio chimes and smart lapel pins.
This misunderstands what screens are actually good at. Visual displays remain unbeatable for spatial and aesthetic consumption.
Nobody wants an audio agent to spend five minutes reading turn-by-turn navigation instructions when a glance at a digital map conveys the geography in two seconds. Nobody wants to consume video, photography, technical architecture diagrams, or complex financial charts through an earpiece. Visual displays are irreplaceable whenever the information being absorbed is inherently rich or spatial.
However, screens are simply an inefficient mechanism for administrative labor.
Much of what passes for modern computer usage is tedious data entry disguised as an interface. You spend forty minutes tapping through eighteen pages on a mobile browser to compare home insurance policies, re-entering your address, dodging cookie consent banners, declining roadside assistance upsells, and dismissing fake countdown timers designed by growth marketing teams. You have to spend fifteen minutes scrolling through sponsored listings and questionable reviews on Amazon just to order an extension lead.
The personal agent replaces that manual labor with delegation. You give an instruction from your phone or wrist: “Find a home insurance policy covering water damage, match my current deductibles, keep it under £600, and purchase it.”
The agent navigates the comparison sites on its own cloud browser, evaluates policy terms, ignores dark patterns, and executes the transaction via API. The screen remains what it always should have been: a high-resolution display for media and communication, while routine administration disappears into the background.
This transition creates an awkward commercial dilemma for the internet’s largest advertising platforms.
Take Amazon. Outside observers often view Amazon as an e-commerce logistics company that happens to sell some ads. Inside the income statement, the relationship is almost inverted. Amazon’s direct retail operations operate on paper-thin operating margins. The real profit engine of its retail marketplace is its $68+ billion advertising business, which charges third-party sellers fees to appear at the top of search results.
If you search for an iPhone charger on Amazon today, the first 4 slots are paid placements, followed by private-label clones, followed by sponsored brand banners. Amazon’s retail profitability relies on human shoppers scrolling past those placements and occasionally clicking on them.
A personal agent shopping on your behalf breaks this economic model entirely.
An agent does not scroll through sponsored listings. It queries the product catalog, inspects technical specifications, checks historical pricing data, filters out review fraud, and selects the optimal item. A software script is completely immune to sponsored badges, bright orange stickers, or urgency timers.
If consumers delegate purchasing to agents rather than browsing Amazon directly, the value of buying a sponsored product placement collapses.
Google faces an identical problem at four times the scale.
Google’s $200+ billion search business is an auction mechanism for human attention. For twenty-five years, Google has monetised the intent of people looking for information by placing paid links above organic results. Over time, that interface has deteriorated into an obstacle course of sponsored placements, affiliate marketing sites, and content farms engineered to game Google’s ranking algorithms.
When a personal agent handles research, it bypasses the search engine results page entirely. It fetches the underlying documentation, cross-references factual claims across primary sources, and delivers a concise synthesis to your device.
The user never visits Google, never sees an AdWords link, and never generates an impression. The entire architecture of search advertising assumes a human eyeball at the end of the query. When you replace the eyeball with a script, the auction has nobody left to manipulate.
When web pages become headless endpoints and mobile apps turn into background execution plumbing, economic value pools in the proxy.
In the smartphone era, value accrued to Apple and Google because they owned the operating system and the physical glass in your hand. In an agentic era, power shifts to whoever controls the software that manages your credentials, executes your transactions, and holds your context.
The vendor whose agent decides which flight you book, which insurer you choose, which grocery store delivers your food, and which services touch your bank account holds immense distribution leverage.
The defining technology competition of the next decade will not be about who designs the sleekest handset or who trains the largest language model. It will be about who builds the personal agent you trust to run your digital life.