The keynote began by outlining six key themes for the AI engineering conference, emphasizing that AI is evolving beyond just models to encompass services, data, branding, and products. This broadening scope benefits AI engineers by expanding their roles beyond model development. A significant trend highlighted was the increasing proportion of code generated by AI, which is projected to rise from around 10% currently to nearly half of all code by the end of the year. The emergence of AI agents capable of self-hosting and operating in complex, non-verifiable domains was also noted, alongside AI’s growing impact on advanced scientific and technological fields, reinforcing the speaker’s optimistic view on the future job security and influence of AI engineers.
George Cameron from Artificial Analysis then took the stage to provide an overview of the current AI landscape as of mid-2026. His company specializes in independent benchmarking of AI models, agents, inference providers, and hardware across various modalities including text, image, video, speech, and music. Cameron presented a detailed chart showing a rapid increase in AI model releases and intelligence improvements, countering claims that AI progress had slowed. He introduced the Artificial Analysis Intelligence Index, a composite metric derived from ten benchmarks, which currently ranks Claude Opus 4.8 as the leading language model, surpassing GPT 5.5.
Cameron discussed the geographic distribution of leading AI labs, noting dominance by the US and China, with contributions from France, South Korea, and the UAE, but highlighted Australia’s absence from the frontier AI model development scene. He compared proprietary models with open-weight (open-source) models, showing that while open models lag behind by three to nine months in intelligence, they remain competitive and valuable due to their flexibility and cost advantages. This dynamic is expected to continue, providing users with multiple viable options depending on their needs.
The talk also covered the economic aspects of AI model usage, explaining that although accessing GPT-4 level intelligence has become cheaper, overall spending on AI within companies is increasing. Cameron identified six key drivers affecting AI costs: smaller yet more capable models, software and hardware efficiency improvements, the demand for larger and more intelligent models, increased reasoning token usage, and the rise of multi-turn agent interactions that multiply inference costs. He illustrated how inference prices have dropped dramatically over 6 to 18 months, often by factors of 10 to 100, enabling more cost-effective deployment of AI for various tasks.
Finally, Cameron emphasized the importance of benchmarking aligned with real-world use cases to ensure meaningful progress in AI capabilities. He showcased examples of agentic benchmarks demonstrating significant improvements in knowledge work and creative tasks over the past year. The presentation concluded with a reflection on the affordability of running comprehensive intelligence benchmarks today compared to the past, underscoring the rapid advancements and cost reductions in AI technology that are shaping the industry’s future.
