2025 was the year of agents, what's coming in 2026?

Daniel Wightnack and Chris Benson discuss the rapid advancements in AI, highlighting that 2026 is set to be a pivotal year. They reflect on the shift from AI models and assistants to AI agents in 2025, with varying reports on the success of these agents leading to both confusion and excitement among organizations. Chris emphasizes the importance of identifying suitable use cases and establishing a solid business case for AI agents. The conversation touches on the evolution of coding with AI models like Opus 4.5 and OpenAI's 5.2, which have significantly improved senior-level coding skills.

The impact of agentic AI extends beyond coding, influencing various fields once successful use cases are identified. The need for quick adaptation to powerful new tools is underscored, as is the importance of domain knowledge in leveraging agentic workflows. Gartner reports that only 11% of organizations have agentic AI in production, predicting a 40% failure rate for projects by 2027. Key success factors include expertise in prompting and configuring AI systems, knowledge of data sources, and integration into workflows.

The podcast discusses the evolution of AI from models to systems, predicting increased multimodality, with most customer interactions involving multimodal AI inputs. The emergence of the "reasoning era" in AI is introduced, where reasoning models simulate reasoning to aid in complex tasks. However, concerns about latency in business applications arise from the use of reasoning tokens.

The costs associated with AI models are noted, with significant computational costs for each token generated. The choice between "instant" and "thinking" modes reflects consumer demand for complex outputs, which are more expensive to produce. Local and geopolitical dynamics are also addressed, with nations focusing on securing energy resources to support AI growth, leading to rising political tensions.

The evolving role of AI in policy decisions is acknowledged, with challenges in maintaining AI infrastructure and energy consumption anticipated as 2026 approaches. Notable advancements in generative AI models are recognized, although they have plateaued while predictive models continue to advance rapidly. The conversation explains the differences between generative and discriminative models, emphasizing ongoing improvements in predictive models.

The concept of auto ML has evolved into augmented analytics, integrating various tools into generative AI models for enhanced functionality. The importance of orchestrator models that integrate various tools is emphasized, creating a multiplicative effect in AI capabilities. The limitations of transformers and generative AI lead to a shift towards world models for enhanced predictive capabilities.

Looking ahead to 2026, there is a strong emphasis on the need for practitioners skilled in building Model Control Protocol (MCP) servers and orchestrating various systems. The roles of data scientists and software developers are expected to evolve, focusing more on AI integration and tool development. The conversation also touches on the potential for niche applications across industries, where domain expertise will be crucial.

Advancements in GPUs and ASICs are making AI technology more affordable and embeddable in smaller devices, paving the way for an "AI maker era" at the consumer level. The commoditization of AI models is noted, with open-source models catching up and emphasizing the importance of flexibility. The AI ecosystem is becoming increasingly fragmented, and successful companies will likely be those that simplify this complexity and provide consolidated solutions for AI implementation.