2025 AI review: Why LLMs stalled and the outlook for 2026

Here it is! We review the year where scaling large AI models hit its ceiling, Google reclaimed momentum with efficient vertical integration, and the market shifted from hype to viability.

Join us as we talk about why human-in-the-loop is failing, why generative AI agents validating other agents compounds errors, and how small expert data quietly beat the big models. We also share our excitement for 2026 as we see a necessary return to research-backed systems design with an eye for interoperability and measured outcomes.

Chapters

  • Let's review our 2025 topics (0:00)
  • Google’s comeback and monetization (0:44)
  • Scaling limits and diminishing returns (4:49)
  • Human-in-the-loop meets its match (9:13)
  • Agents and the validation loop (12:21)
  • Small, high-quality data outperforms (16:32)
  • Expert systems and scientific rigor (20:10)
  • The human aspects of AI (23:05)
  • Special thanks to our guests (24:12)
  • Surprises of 2025: Google and scaling’s end (24:58)
  • Unexpected hot takes in 2025 (26:57)
  • Hopes for 2026: Research and interpretability (29:03)
  • Cautious optimism: AI profitability, ROI, and market reality (34:44)
  • And some skepticism for AI in 2026 (36:15)
  • Economic reflections and what’s next (40:00)

Our annual review of AI and our outlook on AI growth for 2026: Episode summary

Beyond the headlines, we reviewed our discussions from the past year about the operational realities most teams faced: human-in-the-loop that functioned like a checkbox, agents validating agents in a tidy but fragile loop, and the rising cost of skipping validation in the rush from POC to production.

Over the past year, we noted that the AI conversation shifted from hype to habits. We share results showing how small, high-quality human data outperformed massive synthetic labels on nuanced tasks and why expert systems—sometimes classic models, not LLMs—delivered higher accuracy, lower cost, and cleaner governance. From actuarial methods to causal modeling, we showcase how machine learning strengthens the scientific method rather than trying to replace it, and we explain how interpretability and provenance could make a key difference for enterprise trust. Imagine answers that come with receipts: where they came from, why they’re likely, and how confident the system is.

Looking ahead to 2026, we’re optimistic about a return to first principles: models built for purpose, governance that measures what matters, and research that facilitates knowledge creation over knowledge replication. We explore the market’s demand for a clear path to ROI, not just usage graphs, and outline how teams can right-size their stacks, align incentives, and ship durable value.

Resources

Do you have a question about AI governance practices in 2026?

Connect with us to comment on your favorite topics:

  • LinkedIn - Episode summaries, shares of cited articles, and more.
  • YouTube - Was it something that we said? Good. Share your favorite quotes.
  • Visit our page - see past episodes and submit your feedback! It continues to inspire future episodes.

Follow, share with a teammate who sets AI strategy, and leave a quick review telling us your biggest AI win—or miss—of 2025.