Every massive consumer technology business is selling the same thing: the removal of thought. In the end, convenience always wins.

Dishwashers automated hours of domestic chores. GPS externalized our spatial memory. Calculators offloaded routine arithmetic. The pattern accelerates with each technological wave. Google’s search box outsources memory. Amazon’s "Buy Now" outsources patience. TikTok’s infinite scroll outsources choice—a drip feed of dopamine-calibrated algorithms that know your desires better than you.

We no longer have to think about what we want. Technology has decoded our curiosities into anticipatory streams, unleashing our attention for the pursuits that define us: deep work, creative exploration, and genuine connection. People are allergic to optionality. Most iPhone and Gmail users have never altered a single default setting. In countries where you have to actively opt-out of organ donation, consent rates are 98%, whereas in countries where you have to actively opt-in to organ donation these rates are 15%. We do not want to try new things and especially don’t want to change what is already working.

The ultimate evolution is the invisible AI operating system. An omnipresent companion, attuned to the rhythm of your life, orchestrating thousands of micro-decisions in the background. A tool to alleviate the consumption of your cognitive bandwidth. A tool that augments human capability in the pursuit of the best work. A tool to serve convenience in the form of removed thought.

It won’t ask about your schedule. It would have already briefed you on your day ahead, primed you on the important objectives for each meeting, and distilled transcripts into notes that feel unmistakably yours. It won’t wait idly for you to act on customer feedback; it will have already drafted structured feature requests, ready for your approval and instant incorporation. Without the friction of explicit requests, it anticipates what you need by analyzing your data in real-time and taking action flawlessly on your behalf.

While the merits of this OS are tantalizing, we’re far from close. Today’s assistants are still glorified command-line interfaces waiting for the perfect prompt. Instead of eliminating cognitive friction, AI assistants today demand more from us—more effort, more interfaces, more behaviors to learn.

The next wave will embody ambient intelligence: No extra commands. No expected self-education. No frothy promises about enhanced workflows.

You don’t win by asking users to go somewhere new. You win by being where they already are.

History unequivocally shows that the most powerful technologies emerge where multiple breakthroughs intersect.

The PC arrived when microprocessors met affordable displays. The internet exploded when browsers met widespread connectivity. Mobile surged when touchscreens merged with apps. What becomes possible when the world’s most sophisticated generative models can fit in our pockets?

Today, we’re witnessing three critical barriers fall simultaneously: computational (through efficient processing), intelligence (through breakthrough models), and interface (through natural interaction). What we could achieve today would not have been possible at any prior moment. The dishwasher, "Buy Now," and soon, the augmentation of human cognition.

The opportunity to make humans more productive is an enormous prize.

Owning the agents that plan, predict, and supplement the lives of knowledge workers will dominate and define the future of work. Just like the assembly line empowered a single worker to outperform entire teams, these agents will empower individuals to do the work of ten, achieving superhuman level efficiency. When elite individuals harness this power, they will set new benchmarks for innovation and impact, cascading their influence through teams, networks, and entire industries, and simultaneously capturing an unprecedented share of human attention and economic value.

What was once a vision only dreamed of by science fiction—the elusive "magic" black box that commands our machines—is finally possible to be created. We are witnessing the dawn of truly personal AI companions and the question is not whether it will happen, it is who will lead it.

Personal computing hinges on two primitives: retrieval and prediction. Each amplifies the other; neither thrives in isolation. Each problem could be a full company on its own.

Retrieval

Data underpins intelligence. Modern data streams are inherently multi-tenant: fragmented, unstructured, and boundaryless. Creating a unified index from this chaos—including emails, texts, calls, locations, browser histories, keystrokes, clicks and microinteractions—requires total capture. The system must ingest and process massive volumes of data in real-time, detecting subtle deviations that suggest meaningful shifts in user behavior or intent.

Prediction

Prediction without context is just sophisticated guesswork. Prediction grounded in context, however, is informed foresight. True breakthroughs emerge when machines learn to recognize patterns and build an intuition of intent. This alignment enables anticipatory execution—completing the text, dispatching the preemptive invitation, presenting the next action before the previous one finishes. Profoundly opinionated, yet continuously refined by algorithmic adjustment.

Compounding Intelligence

The system must evolve dynamically. Each interaction fuels a flywheel: daily data capture paired with real-world feedback produces exponentially better predictions. This creates a tight feedback loop between user intent and system responses.

The core metric is predictive accuracy: the delta between forecasted action and real output. By training a reward model that tracks this loss and iteratively minimizes that loss, the system becomes increasingly anticipatory. It learns to refine every prediction with increasing precision. The result feels telepathic: a system attuned to both present desires and the ones still crystallizing in consciousness.

Implementation

In practice, this requires deep integration at the operating system layer—intercepting and interpreting system events at the lowest possible layer. The granularity of observing and controlling UI elements within every application far surpasses anything achievable through screenshot capture, offering a comprehensive understanding of application behavior and structure. By traversing element hierarchies, detecting text fields, and interfacing with control mechanisms, the system gains context to a microscopic degree. Through capturing global events, managing windows, and crossing process boundaries, it gains the control needed to predict your next action.

Invisible Automation

The interface must be omnipresent yet invisible, seamlessly inhabiting your existing tools and workflows.

As interaction patterns accumulate, the system continues to speculate on low-entropy (highly predictable) sequences of actions—auto-populating text fields, pre-selecting UI elements, and hovering over likely-to-be-clicked buttons. Discrete mechanisms today, like a simple "tab" confirmation, offer an effective way to approve and execute workflows, but they represent only a stepping stone to a world of new interfaces.

The model scales from augmenting small micro-actions with subtle automations to running indispensable workflows entirely, where each positive reinforcement—accepting the prediction—further cements the pathway until it becomes an intrinsic extension of you.

This is precisely where the next wave of software should position: building into the systems users already love. Form a thin and unobtrusive yet indispensable layer that becomes a part of routine. Then, enabling data aggregation and providing automation of increasing value on top. By focusing initially on reimagining universal, high-stakes workflows, like messaging, a natural wedge is created that compounds in value over the long term.

The power is not in building the most sophisticated prediction engine. The power lies in becoming the unconscious default.

Counter Positioning

New companies hold a noticeable advantage: the freedom to reimagine the operating system from scratch, unbothered by the need to protect revenue streams, maintain backward compatibility, or cater to existing user bases.

Why are we on the sixteenth generation of iPhone instead of pioneering brain-computer interfaces? Apple strives to be everything to everyone. Apple is incentivized to not change—not touch what already works for the company and its stakeholders, not cannibalize its decades of success.

This leaves a gap for startups to aggressively embrace opinionated software: Ones that obsessively seek the right answer no matter how conflicting it may be. Ones that commit to a fast cadence of release, failing fast, iterating even faster. Ones that thrive on the liberty to be wrong, until they’re right.

Lastly, this AI OS is not for booking hotels or planning travel—charming as they may be. It is for high stakes decisions, where there is no room for ambiguity: sending the right calendar invite, flawlessly migrating unstructured data from one home to the next, or prioritizing the right client to close a critical deal. It owns workflows that drive growth and define success.

Network Economies

The best part of the intelligence layer is that it becomes increasingly valuable as more people join, sharing common workflows and collective context. Each new user contributes unique insights—integrating niche tools or solving edge cases—that enrich the system’s intelligence for everyone. This compounding feedback loop transforms the AI OS into a "treasure trove" of exponentially growing value, scaling personal utility with every interaction.

Switching Costs

Once integrated, the AI becomes an extension of your thought process, holding the context of years of decisions, interactions, and behaviors. Its value lies in the intelligence it accumulates, making switching not just inconvenient but unthinkable—like losing the best coworker you’ve ever had.

Regulation

Regulatory shifts level the playing field. They force incumbents to change while opening doors for startups to seize the moment. Take GDPR and CCPA: These regulations made user data control a legal imperative, creating massive headaches for incumbents entrenched in old systems. For startups, though, these shifts are tailwinds. They can build with the new rules baked in from day one, while incumbents scramble to retrofit their systems.

Although we are playing the long game here, the playbook is clear: Anticipate the shift, align with the future, and execute faster than anyone else.

iMessage is the new email. Read my inbox and you’ll gather some context about me. But read my texts and you’ll know everything.

Inside Apple’s walled gardens is a stealth missile into the world’s most valuable and contextual conversations.

Beneath innocuous blue bubbles, nestled between family photos and emoji reactions lay multi-billion dollar deals, career-making introductions, and split-second decisions—all in their most intimate form.

Sitting on top of this stream of richly personal context, we capture the highest-leverage interactions in their purest forms. Each message creates a compounding feedback loop: receive, predict, optimize, repeat. Every interaction becomes a data point, every response a learning opportunity to understand you one step closer.

The Trojan Horse is building the iMessage assistant that texts like you. But the true power lies in what this position enables—an unparalleled aggregation of context that fuels proactive assistance. By embedding in this essential channel, we expand methodically, shipping more delightful products that hook into the narrative of convenience, and over time, becoming the default of productivity. It becomes the intelligence layer on top of the most valuable asset humans own: decision.

The Wedge

Build something disproportionately valuable to get your foot in the door. Become the default way people text; capture the highest-fidelity signal of human intent and decision velocity; create unprecedented context for future intelligence.

Workflow Osmosis

Extend to adjacent high-stakes workflows (scheduling, tasks, documents). Tasks emerge from message threads. Meetings auto-schedule and auto-adjust. Frictionless coordination and work delegation.

Proactive Intelligence

Preemptively surface relevant context from conversations, documents, history. Suggest next-step action: fire off the follow-up next steps email, send the calendar invite, forward the flight receipt for reimbursement.

Chain Reactions

High-signal events (meeting acceptances, customer complaints, deadline changes) trigger cascading automated workflows. Learn to recognize these trigger points and their downstream implications, automatically orchestrating complex sequences of actions across tools and platforms. IFTTT meets intelligence—instead of manually programming triggers, the system inductively learns them from behavioral patterns.

Universal Operating System

Become the invisible transparent fabric driving all decision—from micro-interactions to strategic moves, each passes through a unified intelligence layer that deeply understands context, simulates outcomes, and orchestrates the optimal path.

The next wave of software will do work without you asking. By starting with immediate value—making messaging effortless—we create the first bridge to true anticipatory intelligence.

Rewind.

In 2022, Rewind popularized the concept of "always recording" on your desktop—capturing all the things you’ve ever seen, said, or heard. At first glance, it presents itself as the ultimate unlock for human productivity: a searchable archive of our entire digital existence. But the thing is: most people don’t need to revisit the fourth time they said "banana" in January last year. Search, in isolation, isn’t a product—it’s a means to an end.

When Rewind first launched, the limitations of M1 computers made local processing for such a task nearly impossible. But times have changed. We’re on M4s and anything can happen. Especially when you go local :)

Highlight.

Embedded desktop assistant. Highlight allows you to instantly process any digital content—from meetings to web pages—through powerful language models. By combining screen capture, voice recognition, and deep integrations with tools like Notion and Linear, it makes you much faster at one-step action. The biggest hurdle here will be that it is reactive first rather than addressing a burning, first-order problem that drives immediate adoption.

Granola.

You might be surprised to see Granola here. Unlike most AI companies chasing the universal assistant dream, they are focused on one thing and one thing only: a new way of taking notes. Meetings are the epicenter of knowledge capture and control. To cement into a workflow with persistent meeting context will serve as a treasure trove of organizational intelligence, creating a compounding knowledge base that becomes the foundation for more sophisticated AI application.

The Browser Company.

Famed for their first browser, Arc, The Browser Company has reinvented themselves with Dia—a browser powered by AI-driven "self-driving" for your computer. Like many browser assistants, Dia leverages the transparency of browser DOMs, reading events as they happen and commanding machines with unmatched precision. The question is whether you can get one level lower. Why stop at the browser?

Platform Dependence

Apple could revoke essential MacOS APIs tomorrow, shattering our technical foundation. Building on another’s platform demands constant vigilance and adaptability.

Market Hypothesis Risk

We’re betting humans will continue preferring to think less. However, a subset of power users might prefer deeper engagement with AI—they may want active participation and agency in system training rather than passive benefit.

The Hardware Reality

To truly disrupt incumbents, software alone won’t suffice. Becoming a device company is inevitable on a billion-dollar scale. This transition timing is critical—too early limits growth, too late cedes advantage. The decision to focus on hardware should be delayed until market validation, but is core to the company’s endgame strategy.