Jozef Soja: AI agents and the future of work

Jozef Soja, a research analyst at ARK Invest focusing on the next generation internet, presented key insights from ARK's Big Ideas 2026 report. The discussion centered on the accelerating capabilities of AI agents, the collapsing cost of AI intelligence, and the evolving global race in AI development. Soja's analysis highlights how AI is poised to fundamentally reshape productivity and economic growth.

AI Agents Becoming More Proficient

Advances in reasoning, tool use, and extended context are driving an exponential increase in the capability of AI agents. Soja presented data showing that AI agents are becoming significantly faster at completing tasks. In early 2025, AI agents could reliably complete tasks in 6 minutes that previously took humans 31 minutes. By the end of that year, this time was projected to further reduce to tasks that take humans over 30 minutes. This improvement reflects an increasing value unlocked by AI, with a reported daily time saving of 50 minutes per user, which translates to significant value even at a modest subscription cost.

The Collapsing Cost of Intelligence

A critical trend highlighted is the dramatic decrease in the cost of AI intelligence. Software development costs for AI models scoring above 50% on benchmarks have fallen 81% from $3.50 to $0.32 per million tokens in just eight months. This cost reduction is making sophisticated AI more accessible to a wider range of users and businesses. For instance, the cost of achieving human-level performance on tasks like reasoning or coding is plummeting, suggesting a future where advanced AI capabilities are not a luxury but a standard offering.

US Leads AI Race, China Rapidly Gains

The analysis also delved into the competitive global landscape of AI development, particularly between the US and China. While US-based AI models currently demonstrate superior performance, with many of the top open-source models originating from US labs, China is making rapid strides. The performance index shows a significant improvement in Chinese AI models over the past year, closing the gap with their US counterparts. However, China's semiconductor manufacturing capabilities, exemplified by SMIC, lag behind Taiwan's TSMC by a substantial margin, producing 38 times less compute. This dependency on external chip manufacturing could present a bottleneck for China's AI ambitions. The report projects that to maintain its lead, China will need to increase its access to advanced compute capabilities.

AI-Native Revenue Growth

The revenue generated by AI-native companies is growing at an unprecedented rate. General-purpose AI providers like OpenAI and Anthropic are now rivaling large public software companies in revenue. Furthermore, specialized AI startups are demonstrating remarkable growth. Companies like Cursor, Harvey, OpenEvidence, and Sierra, all founded within the last three years, have already achieved annualized recurring revenue (ARR) milestones of $100 million or even $1 billion. This rapid growth indicates strong market demand and successful product-market fit for AI-driven solutions across various sectors.

Accelerating Worker Productivity

The adoption of AI is expected to significantly accelerate worker productivity, presenting a multi-trillion dollar opportunity. As businesses increasingly adopt AI, global software spend is predicted to grow at an annual rate of 14% over the next five years. This growth is driven by increased automation and efficiency gains. Soja presented scenarios showing that in a rapid mass adoption scenario, automation could account for 61% of tasks, leading to a value unlock of $117 trillion by 2030. Even with more moderate investment, AI is projected to contribute trillions in productivity gains. This acceleration in AI capabilities and adoption suggests a transformative period ahead for the global economy.