newsfilter.io
Conference Presentation, Panel

Welcome to the New Frontier of Agentic AI | Wiz, Merge, You.com & More | RAISE Summit 2026

  • Organizations will likely transition from cost-insensitivity to cost-optimization as the industry matures and usage scales, potentially resulting in rate limiting for premium users.
  • Companies are expected to adopt multi-model routing strategies that supplement proprietary frontier models with open-source alternatives to optimize costs and specific workloads.
  • Productivity gains from AI are projected to correlate with the ability to reduce cycle time rather than token spend.
  • The industry may reach a point where most models are "good enough" for most tasks, shifting the primary competitive focus from model quality to cost.
  • Security and privacy will remain the primary decision-making factors for private AI deployments, particularly where data ownership and inference security are paramount.
  • Future deployments are expected to heavily favor American open-source models and fully US-deployed infrastructure, including air-gapped GPUs, while disfavoring direct access to Chinese open-source providers.
  • AI model selection trends are expected to mirror the cloud industry's shift from single-cloud to multi-cloud, requiring architectures that allow flexible switching between models to maintain confidence and scalability.
  • Software infrastructure for agents is currently in an early stage with significant gaps in observability and security, presenting an opportunity for engineering-focused solutions.
  • Organizations will likely require robust control planes and governance frameworks to handle regulatory constraints and customer mandates regarding model and regional usage for data.
  • Benchmarks measuring model performance will increasingly need to quantify variance stemming from agent inconsistency and task difficulty rather than relying on single static numbers.
  • Agent inconsistency in multi-turn, multi-tool environments is expected to be a significant reliability risk that applications must be designed to withstand.
  • For code-specific workloads, model routing may simplify to a selection among a small set of models rather than a complex router managing hundreds of options.
  • Step-by-step model selection using different models for each step of an agent's reasoning is expected to yield poor results compared to a mixed-phase approach distinguishing planning from execution.
  • The strategic use of specialized tools and data partnerships can allow standard or lower-cost models to outperform frontier models on specific benchmarks.
  • Future architectures must be designed to withstand massive shifts in model providers and capabilities over time due to the pace of change in AI.
  • The fundamental "mode" of AI systems will increasingly reside in the underlying data model or security graph rather than in the specific inference model used.
  • A trend is expected where frontier model providers integrate tool-calling directly into their model training, potentially creating closed agentic frameworks that perform poorly outside their native ecosystem.
  • The industry has not yet experienced a singular "mythos" event that has fundamentally disrupted the competitive landscape, though continuous evolution is ongoing.
  • Organizational trust is identified as the primary barrier preventing the move toward end-to-end automation of day-to-day workflows, with automation levels expected to rise as trust grows.
  • Open-weight models are predicted to become smart enough in the near future to accomplish many complex tasks currently requiring frontier models.
  • Future AI development is expected to shift from a scientific problem to an engineering problem, focusing on infrastructure and harnesses to ensure stable outcomes regardless of model changes.