Summary
This second-place winner from DeepLearningAI’s 7-day Voice AI Builder Challenge demonstrates an AI coding assistant named Echo — built on Claude Code and Vocal Bridge — that works autonomously through a full project brief and places a single outbound call to the developer when it reaches a decision it cannot make alone. The user deliberately steps away from the computer, triggering an “AFK skill” that gives the agent a decision matrix for escalating only non-reversible, judgment-required choices.
The project involves positioning and launch page copy for a product called Tempo. Echo processes all tasks it can complete independently, constructs a structured decision card, and calls the developer mid-walk to ask whether the launch campaign should lead on value or lead on price. The agent doesn’t just present options — it offers a recommendation (lead on value), articulates the risk of leading on price against a better-funded incumbent, and fields follow-up questions before locking in the developer’s final call.
After the call, Echo records the decision directly in the project, creates a follow-up task to reassess the strategy in seven days, and delivers a full session summary when the user returns with an “AFK back” command. The demo illustrates a practical asynchronous human-in-the-loop architecture where developers can be genuinely offline while complex, multi-step work continues and escalates gracefully.
📺 Source: DeepLearningAI · Published August 05, 2026
🏷️ Format: Showcase
