Panel, Fireside Chat
Pilot to Production: The Valley of Death and How to Cross It | RAISE Summit 2026
RAISE SummitBarak Kaufman, Jonathan Corbin, Euro Beinat, Ivan Burazin, Ariel Shulman, Venkat Pullela, Yvonne, Juro
- Real-time web data access is deemed essential to prevent AI agents from remaining contextually stuck, as reliance on static data or unverified web environments risks providing misinformation that leads to catastrophic, radical decisions during production scaling.
- Deployment is expected to evolve from initial "co-pilot" phases in 2024 to "function-oriented work" in 2025, characterized by a shift from technology hurdles to organizational restructuring required to break silos and drive value creation.
- To ensure safety and compliance at scale, companies are building internal guardrails and utilizing Windows Sandboxes for computer use, while future operations aim to reach an "headless via API" end state with a focus on "consistency and reproducibility" in data to offset the lack of human intuition.
- Full computer access, including accounts and financial instruments, is anticipated to enable end-to-end task execution, while strategies for model selection increasingly favor "model agnostic architecture" and "on-prem deployment" to mitigate lock-in risks and data security concerns.
- Enterprises are predicted to move from "pilot to production" within two years, leading to a rethinking of workflows where legacy system renewals are no longer automatic and the number of deployed systems decreases as redundant databases are eliminated.
- Financial projections indicate a 20-30 percent annual growth rate for companies treating AI as a growth tool, with specific implementations targeting 50-60 percent reductions in operational inefficiencies and CSAT scores between 96-97 percent.
- Value extraction is expected to persist for the next five to ten years even without model breakthroughs, though ROI justification will increasingly depend on revenue gains rather than cost savings, requiring a "non-consensus" belief system and enduring first-principle strategies to survive infinite competition.
- Organizations are expected to leverage "continual learning" and post-training or RL on open-source models to achieve specific state-of-the-art capabilities, while startups hold a distinct advantage by avoiding legacy baggage to build agents from scratch.
- Future success relies on avoiding catastrophic errors from inconsistent data, such as mass incorrect transactions, necessitating reliable data access and "forward deploy teams" to drive adoption and workflow rethinking alongside client teams.