# How do you diffuse AI into the real world? — Varun Shenoy, Long Lake

## Executive summary

The deployment of advanced AI agents into real-world service industries is not merely a technological challenge but an operational one. Drawing parallels to the adoption of electricity and Ford's assembly line, the speaker argues that technology diffusion takes generations. Long Lake addresses this by acquiring and operating services businesses (e.g., property management) rather than selling software. Their approach focuses on building AI agents that move beyond simple 'co-pilots' to become autonomous 'co-workers,' leveraging proprietary ground truth data collected from messy, real-world tasks—a process requiring deep, physical co-design with the industry.

## Key takeaways

- AI Diffusion Takes Generations: The adoption of general-purpose technologies (GPTs) is slow. Just as electricity took decades to fully integrate into industries like Ford's, AI requires massive operational shifts—ripping out old processes and retraining staff—to achieve full diffusion. [1:30]
- The Value of Owning the Outcome: Long Lake does not sell AI software; they acquire and operate services businesses (e.g., HOA, architecture). By being the operator/owner, they bear the risk when the AI fails, ensuring deep integration and accountability that external vendors cannot match. [2:36]
- The Progression from Co-pilot to Co-worker: AI agents must progress through stages of autonomy: Co-pilot (simple RAG chatbot) $ ightarrow$ Synchronous Agent (real-time, two-way interaction) $ ightarrow$ Asynchronous Agent (background work, external triggers) $ ightarrow$ Long-running Agent $ ightarrow$ AI Co-worker (proactive partner). Achieving the co-worker requires earning the right to do more through iterative field deployment. [6:02]

## Technical details

- Agent Autonomy Ladder: The progression of AI agents is defined by autonomy and interaction type. An Asynchronous Agent allows work to run in the background, triggered externally (e.g., a task completion queue). The ultimate goal is the AI Co-worker: a proactive partner that works alongside the user. [10:32]
- Representing Knowledge Work as Code: Since models are trained on code, the strategy is to leverage coding agents (like Codex or Claude Code) not just for software tasks but to model and automate traditional knowledge work processes. This involves wrapping a coding agent in a sandbox to allow it to build, test, and submit results as a Pull Request (PR). [12:30]
- Ground Truth Data Flywheel: To train effective agents, the focus shifts from public internet data to proprietary 'ground truth.' This is achieved by generating rich traces of data (tool calls, hiccups) as agents collaborate with employees on real tasks. These traces allow for building robust, measurable evaluations (evals), which in turn improve the agent and create a self-reinforcing learning loop. [11:42]
- Continual Learning & Enablement: AI improvement requires two linked initiatives: Continual Learning (improving the agent over time via feedback, whether in prompts or weights) and Enterprise Enablement (getting users to adopt it). Usage drives continual learning, which drives better agents, leading to more usage. [14:00]

## Practical implications

- For build engineers, the focus must shift from building models to building robust data pipelines that capture 'ground truth' (real-world outcomes) and generate rich traces for continuous evaluation.
- System design should plan for asynchronous and long-running agent workflows, rather than relying solely on synchronous, real-time interactions.
- Deployment strategy requires a physical, hands-on approach ('touching grass')—integrating products natively into existing enterprise systems (e.g., ERPs, Excel) and working directly with end-users to overcome process inertia.

## Topics

AI Deployment, Workflow Automation, Enterprise Architecture, Data Engineering, Agentic Systems, Long Lake, American Express Global Business Travel

Source: https://www.youtube.com/watch?v=B0fjR3yaZFU