The Story: You.com’s co-founders Richard Socher and Bryan McCann — among the most cited AI researchers globally — have published a whitepaper outlining 35 predictions for 2026, mapping where AI research, capital, and labor markets may shift next.

1. The LLM Gold Rush Is “Mined Out”
The report argues that the explosive phase of large language model scaling is slowing, with diminishing returns at the frontier. Capital, they suggest, will increasingly flow back toward foundational research, infrastructure, and new architectures beyond traditional transformer scaling.

2. “Reward Engineering” Becomes a Discipline
As agentic systems grow more autonomous, the authors predict prompts alone will no longer suffice. Instead, designing robust reward functions, evaluation loops, and behavioral constraints — dubbed “reward engineering” — could become a formal job category inside AI teams.

3. Traditional Coding Fades Into Oversight
Perhaps the boldest claim: by December 2026, most production code will be written by AI systems, with humans primarily managing architecture, constraints, and review. Developers transition from line-by-line coding to supervising agentic workflows and validating outputs.

Broader Themes
Across the full set of 35 predictions, the whitepaper highlights shifts in labor structures, enterprise AI adoption, agent coordination systems, and the economic compression caused by rapidly falling model inference costs.

Why It Matters: Whether or not all 35 predictions land, the document signals a growing consensus among AI insiders: the next phase of competition may center less on bigger models and more on orchestration, evaluation, and autonomy. If reward systems become more important than prompts — and AI writes the majority of code — the skills valued in engineering teams could change dramatically within a single year.

Source: You.com 2026 AI Predictions Whitepaper