Fireside Chat, Interview
AI Is Only as Fast as Your Slowest Process | Jyoti Bansal, Harness | RAISE Summit 2026
- Harness is projected to become a mainstream concept in the AI conference space, with adoption levels exceeding initial expectations at the company's inception.
- Current AI-generated code output is estimated at two to three times previous volumes, yet consumer shipments increase by only 10% due to downstream bottlenecks.
- Organizations implementing automated AI platforms like Harness can expect to ship at least two times more code, aligning delivery speeds with generation rates and avoiding the 10% output ceiling faced without automation.
- Harness has internally doubled its software shipping capability following the implementation of automated downstream processes.
- Software adoption lags have compressed from three to four years for Europe and seven to eight years for the Middle East relative to the US to a maximum of two years.
- Future innovations such as open-source models and regional data centers are anticipated to emerge in response to government restraints on specific models or data centers.
- There is a 10-year risk that AI may surpass human creativity, shifting the required educational focus for children toward creativity and out-of-the-box thinking rather than strict programming or reliance on AI suggestions.
- Internal tools tracking token usage and tying metrics to specific outcomes, ROI, and unit costs are planned for release to address the transition of AI from an experimental cost to a major expense item.
- Excess power, hardware, and compute resources are expected in the future, potentially rendering them cheaper and more abundant than current needs.
- Organizations must rapidly establish AI governance and manageability protocols because they cannot prevent employees, developers, or users from utilizing the technology.
- European organizations are expected to move slower than American counterparts due to a need for more consensus building, despite sharing similar urgency levels.
- Simulating human judgment remains the most difficult bottleneck in AI processes, particularly for high-stakes environments like banking where single errors can cause significant issues.