Manish Dasaur is a Managing Director at PwC with over 20 years in data and AI, having helped 100+ clients navigate AI disruption and extract real business value from data, AI, and agentic AI initiatives. In this episode, he breaks down why most enterprise AI programs stall — and the playbook the winners are using instead.

Loop engineering is the shift from manually prompting AI agents turn-by-turn to designing autonomous systems that do the prompting for you. Instead of writing a prompt, reading output, then writing the next prompt, you build a small control system that discovers work, delegates tasks to agents (and sub-agents), verifies results, persists state, and decides what to do next on a schedule or until a goal is met. It sits one level above agent harness engineering: where a harness equips a single agent run, a loop keeps firing agents on a cadence, spawning helpers, and feeding itself. The framing is increasingly tool-agnostic, as both Claude Code and OpenAI Codex have converged on similar primitives. The people building at the frontier are already living it: Boris Cherny, head of Claude Code at Anthropic, has said his job is no longer writing prompts but writing the loops that do.

Letting coding agents search the web for documentation wastes tokens, increases latency, and risks importing outdated or incorrect API patterns. To resolve this, this article presents `agent-docs`, a system of local references stored directly within the repository. By instructing agents to search local Markdown documentation maps before attempting web queries, developers can ensure that agents always build on correct, version-aligned instructions.