The article says 95% of AI pilots miss desired results and ROI, to the extent implementation can be measured. At The Next Endeavor in Mountain View, panelists urged clear goals, agent governance and production planning, while noting workflow changes are difficult.
Panelists included Paul Baier of GAI Insights, Andrew Lau of Jellyfish, Pankaj Thakkar of Kloudfuse and Rahul Todkar of Nasik. Todkar said agents are easy to build but difficult to run in production, where sandbox setups may not reflect live systems and debugging can take time. Thakkar raised the question of who is responsible when a production system fails.
Lau said engineering may be one place to see productivity gains, but token costs are easier to track than returns; Todkar said usage patterns vary among functions such as engineering, finance and HR. On open-source models, Thakkar said they can be adapted to specific needs but the industry is not yet there, while Lau expected wide differences in cost and utility once returns are clearer.
