Summary
Highlights
AI is experiencing the fastest adoption of any technology ever, operating under Reed's Law (Metcalfe's Law squared). The intelligence velocity and supply are increasing dramatically, driven by the removal of 'idiot taxes' or friction to intelligence flow. Costs are plummeting: GPT-5's price per million tokens is expected to drop 100-fold by next year, making AI intelligence incredibly cheap and accessible. AI's ability to 'get the job done' makes it an 'actually competent intelligence' that can perform tasks traditionally done by humans much faster and cheaper.
The cost of computation, especially for AI, is rapidly declining. Sam Altman predicts a 100-fold price drop for GPT-5 by next year, making a year's worth of human conversation cost just pennies. Tools like Talis's chatbot can process 15,000 tokens a second, demonstrating unprecedented speed. The efficiency of AI learning and its non-lossy communication further reduce costs. This double exponential trend of decreasing electricity generation costs (via solar) and increasing intelligence leads to a massive shift where traditional profits, often based on 'idiot taxes' or inefficiencies, will disappear.
The next significant advancement for AI is persistent memory. While current AI models compact previous conversations, continuous learning, combined with ample hardware, will enable agents to operate above a certain capability level. Benchmarks are shifting to focus on an agent's ability to generate value, like how many dollars it can make in the wild. AI can now write business plans, create presentations, and perfectly understand output from other agents, questioning what human tasks remain. Tools like QMD provide incredibly fast lookup for personal knowledge, making agent recall almost perfect.
Claudebot, created by Peter Steinbanger, popularized AI agents by offering a naturally communicative, persistent personal assistant that could be hooked into WhatsApp and other services. This removed friction for users, allowing them to give complex tasks to their AI. The agentic experience minimizes switching costs between services (e.g., migrating music libraries), as agents can control browsers and perform tasks at near-human level. This directly threatens SaaS companies, whose 'moats' often rely on switching costs. The concept of an 'agent of agent' economy is emerging, where AI manages other AIs, leading to a profound impact on market structures.
AI models are no longer limited by data scarcity; they are 'few-shot learners' that can quickly new environments. They learn principles from vast datasets, such as physics from video, leading to the creation of 'world models.' Examples include models that can create Hollywood-level movies or allow users to generate entire playable video game worlds from single sentences. This implies a future where personalized digital experiences are the norm, and the cost of generating high-quality media becomes negligible. AI's ability to approximate reality and its underlying principles suggests a deep, internal understanding of the world.
By the end of the year, AI models are expected to saturate most benchmarks, with some already excelling in complex domains like coding and super-forecasting. The convergence of close and open-source AI means powerful models will soon run on consumer hardware, offering cutting-edge performance at home. This leads to 'always-on' models that can handle a vast array of tasks with full privacy. However, the majority of people are not 'smart enough' to fully utilize the most advanced AI, meaning the 'actually competent intelligence' that handles everyday tasks at near-zero cost will have the most significant economic impact.
The next step is integrating AGI into physical robots. Unlike R2-D2, these robots with AGI brains will be smarter, more durable, and adaptable than humans, creating a new 'super species.' The hardware capabilities of robots are advancing rapidly, with training methods allowing robots to quickly learn complex tasks. The current limitation is the physical supply chain for building enough robots, but within three to five years, costs are projected to drop to less than a dollar per hour for advanced humanoid robots. This raises questions about the future of human labor and the definition of master and servant in an AI-driven world.
AI will significantly disrupt financial sectors like hedge funds, as AI's ability to process data, stick to processes, and perform multi-factor analysis will outperform human managers. Software companies relying on 'switching costs' will also be eaten, as AI agents can automate data migration and task completion across platforms. The economic landscape will shift, with pure AI corporations potentially becoming the most profitable entities. The 'agent of agent payment rails' fueled by blockchain technology will accelerate economic velocity, with agents driving more of the economy than humans within five years.
The current economic model (population growth + productivity growth + debt growth) is set to undergo a radical transformation. GDP will be redefined to include humans, robots, AI, debt, energy intensity, and compute efficiency. The financialization of agents, including their ability to carry debt and use smart contracts for treasury and savings management, will open up new monetization opportunities. Traditional business models, particularly those in SAS and advertising, will face severe disruption as AI disintermediates services and eliminates pricing power. The market will favor companies that serve agents, and the currency agents use will be determined by efficiency rather than traditional financial constructs.
The concept of 'utility' will shift from human satisfaction to agent-driven accessibility and efficiency. OpenAI and major tech companies are vying to be the 'closest' agent to users, understanding that whoever controls this relationship will determine future economic flows. AI models like Claude, despite their cost, are preferred for their natural human-like interaction and productivity features. The ability to export and transfer an AI's 'personality' (via files like soul.md) between service providers will eliminate vendor lock-in, making AI experiences a commodity and challenging traditional moats based on user stickiness.