2025 AI solutions reviewed, 2026 trends identified

Summary

In the 88th episode of IBM's Mixture of Experts podcast, four panelists come together to review the key AI milestones of 2025. The discussion covers NVIDIA's chip dominance, open-source breakthroughs like Kimi K2 Thinking, the rise of agentic tools (ChatGPT Deep Research, Claude Code), and multimodal AI. The panelists also offer bold predictions for 2026.

Why it matters

The episode distills a noisy year of AI development into four focused lenses: hardware, open source, agents, and multimodality. These perspectives map where corporate investment and R&D are actually heading. For readers tracking the gap between hype and deployment, the panel's framing of 2025 as the year 'super agents' delivered concrete value serves as a useful checkpoint. The explicit emphasis on NVIDIA's supply chain dominance suggests that the defining constraint through 2026 may be compute capacity rather than models.

Highlights

• NVIDIA's dominance in chip supply was identified as the central hardware bottleneck for 2026
• Open-source models (with Kimi K2 Thinking cited as an example) played a leading role in making 2025 a genuine year of breakthroughs
• Agentic AI was validated as the year's 'super agent' proof points through ChatGPT Deep Research and Claude Code

Critical take

The episode features a panel composed of voices affiliated with IBM. Consequently, the 'bold predictions' carry a corporate bias toward enterprise adoption narratives and naturally downplay consumer-focused or regulatory risks. The framework of 'separating AI noise from reality' is itself a marketing tool. There is no independent benchmarking or dissenting opinion at the table. This makes the consensus feel more like a carefully curated summary than a genuine debate.