How do you diffuse AI into the real world? — Varun Shenoy, Long Lake
Real-world agent adoption depends on workflow redesign, operational traces, and hands-on enablement. Code-agent patterns help, but service work has messier exceptions and triggers.
Long Lake says it has acquired **35 services businesses** after raising **over $3 billion**. Its autonomy ladder runs from copilots through synchronous, asynchronous, and long-running agents toward proactive AI coworkers.
For non-coding work, borrow the sandboxed asynchronous pattern but adapt its triggers and interface to each industry. Capture task traces and real outcomes, turn weekly gains into regression tests, and embed tools in the systems employees already use.
Long Lake says it has acquired **35 services businesses** after raising **over $3 billion**. Its autonomy ladder runs from copilots through synchronous, asynchronous, and long-running agents toward proactive AI coworkers. For non-coding work, borrow the sandboxed asynchronous pattern but adapt its triggers and interface to each industry. Capture task traces and real outcomes, turn weekly gains into regression tests, and embed tools in the systems employees already use. Service workflows are more serial, customized, and exception-heavy than coding. Better agents do not create initial usage on their own; adoption still requires process redesign, close observation, training, and often in-person work with operators.
This narrows the enterprise shift from assistance to execution for service businesses: coding’s sandboxed asynchronous pattern is reusable, but its triggers, interfaces, and evaluation must be rebuilt around serial, exception-heavy workflows. It confirms that task traces and regression tests matter while adding process redesign, training, and field observation as prerequisites for actual usage.