Shenoy compares AI adoption with the slow diffusion of electricity: a powerful general-purpose technology does not transform an organization until equipment, processes and skills are redesigned around it. Long Lake approaches that problem as an owner-operator, acquiring or partnering with services businesses and taking responsibility when deployed AI fails.
He describes an autonomy ladder from copilots to synchronous agents, asynchronous agents, long-running agents and eventually proactive AI coworkers. Teams must earn each increase in autonomy because model capability and user trust vary by task, while interaction patterns that work for software engineering may not transfer directly to property management, architecture or finance.
The company captures rich traces as agents collaborate with employees on real work, including tool calls, exceptions, corrections and final business outcomes. Those traces become automatic evaluations and regression tests, support post-training on private operational data, and expose the messy edge cases that are absent from public training corpora.
Shenoy links continual learning and enablement into one flywheel: usage supplies feedback that improves the agent, and a better agent encourages more usage. Initial adoption still requires software-service co-design, embedding tools in existing systems and working directly with employees to understand how their day-to-day processes actually function.
Watch on YouTube
