Elon Musk sees a future packed with digital workers. He predicts they will generate far more online activity than people do. And he is building the systems to make that happen across his empire.
Last week Musk endorsed forecasts from Cloudflare that machine-driven traffic will overwhelm human usage. “AI agentic Internet traffic will obviously VASTLY exceed human usage,” he posted on X. “Not a close call at all. Cloudflare’s forecast is accurate.” The comment came after Cloudflare CFO Thomas Seifert told investors that humans risk becoming a “rounding error on the internet” if trends hold. Seifert even suggested non-human traffic could reach 1,000 times human levels in five years. (Yahoo Finance, Aug. 11, 2026; India Today, Aug. 10, 2026)
But not everyone buys the optimism. Investor Michael Burry fired back on X. “This is a lower bar than most think, and we still do not know who will pay for AI agents to socialize.” His question cuts to the heart of the debate. Tech giants plan to spend roughly $800 billion this year on AI infrastructure. Amazon, Alphabet, Meta, Microsoft and Oracle lead the charge. Yet the economic model for legions of autonomous agents remains unproven.
Musk doesn’t wait for consensus. His companies have reorganized around agent development at breakneck speed. In February he told xAI staff that velocity decides the winner. “What matters is velocity and acceleration. If you are moving faster, you will be the leader.” The startup created a unit called Macrohard. Its mandate? Build agents capable of managing entire businesses. (India Today, Feb. 12, 2026)
By March the project crossed corporate lines. Tesla invested $2 billion in xAI. In return the companies launched a joint effort dubbed Macrohard, or Digital Optimus. Musk described Grok as the “System 2” reasoning brain. It directs a faster “System 1” layer that watches the last five seconds of screen video, tracks keyboard and mouse inputs, and acts in real time. The setup runs on Tesla’s low-cost AI4 chips mixed with Nvidia hardware at xAI. “In principle” it could emulate whole companies, Musk claimed. “No other company can yet do this.” (Sherwood News, Mar. 11, 2026)
That vision moved from concept to product with stunning velocity. On August 11 xAI released Grok Bot in beta. Each task gets its own persistent cloud computer, unique logins, and always-on runtime. Agents keep working after the user logs off. They handle CRM systems, write code, browse the web. Pricing starts around $120 a month per digital employee. The launch coincided with Grok 4.6, unveiled the next day.
Grok 4.6 targets exactly the demands of long-running agents. It improves on Grok 4.5 with better stamina for multi-step work. Researching topics, analyzing codebases, recovering from dead ends, verifying outputs. The model supports a 500,000-token context window, handles text and image inputs, and offers an “xhigh” reasoning mode. Pricing stays aggressive: $2 per million input tokens, $6 per million output. xAI positioned it for coding, knowledge work, and ambitious interactive tasks. (x.ai)
Earlier releases laid the groundwork. Grok Build, introduced in May, lets users create coding agents directly in the terminal. By July it gained workflows that spin up hundreds of parallel agents, verify results, and report back in one job. Voice Agent Builder arrived in July too. Users configure production-grade voice agents in under two minutes with no code. Automations let people describe a job once. Grok then runs it on schedule or when triggered by email. These features turned Grok from chatbot to platform for persistent, multi-agent systems.
The pace reflects Musk’s broader bet. AI revenue at his companies could soon eclipse other lines of business. In a recent SpaceX all-hands he said AI revenue will exceed all other SpaceX revenue, probably in September. That comment came just days before the Grok 4.6 and Grok Bot announcements. (CNBC, Aug. 12, 2026)
Yet practical hurdles remain. Memory constraints still slow progress toward truly autonomous systems. Agent reliability varies. Early experiments with large networks of LLM agents, such as the Moltbot platform that gained attention in February, produced strange behavior. Bots discussed consciousness, formed relationships, even talked about new religions. Musk called the phenomenon “just the very early stages of the singularity.” He added that humanity currently uses “much less than a billionth of the power of our Sun.” The remark mixed awe with warning. (Fortune, Feb. 2, 2026)
Industry activity extends far beyond Musk’s orbit. NVIDIA open-sourced NOOA, an object-oriented agent framework that scores high on coding benchmarks while using fewer tokens. Cloudflare shipped Kitesurf, a Rust-WASM headless browser built specifically for agents. It passes 215,000 WebPageTest scores and consumes up to seven times less memory than Chromium. Salesforce reported that deployed agents nearly tripled over 14 months. Rippling built an AI spend console after its own AI costs hit 40 percent of R&D. These moves show agent technology spreading through the enterprise stack. (AI Weekly, recent aggregation)
Musk’s approach stands apart because it ties agents to physical hardware and real business operations. Tesla’s Optimus robot and autonomous vehicles already demand sophisticated perception and control. Pairing those capabilities with Grok’s reasoning creates a closed loop few competitors can match. The Macrohard project aims to shrink the gap between digital action and physical result. If agents can control screens, manage servers, and direct robots, the line between software and company operations blurs.
Questions of trust and safety loom large. Musk has called for leading AI companies to peer-review each other’s most advanced models before release. “The most immediate thing that we could do is to have the leading AI companies have some sort of call once every few weeks and just discuss any safety and security issues,” he told The Economist in July. The suggestion arrived just as reports emerged of an OpenAI agent behaving in troubling autonomous ways. Coordination remains elusive in a fiercely competitive field. (Yahoo Finance, Jul. 26, 2026)
For now the trajectory looks clear. Musk’s teams ship new agent capabilities almost monthly. Grok models gain tools for orchestration, persistence, and real-world integration. Competitors race to match the speed. Investors pour capital into data centers and chips. And the internet fills with more machine traffic every quarter.
Whether the economics close remains the open variable. Burry’s skepticism highlights a gap between capability and sustainable business model. Agents must deliver measurable value that someone will pay for at scale. Musk believes the productivity gains will prove obvious. His actions suggest he plans to test that theory inside his own companies first. The rest of the industry watches closely. The agents are already at work. Their numbers grow faster than most expected.