Summary
Hursh Agrawal, CTO and co-founder of The Browser Company (makers of Arc and Dia), presents at AI Engineer on how engineering leaders can reclaim hands-on building time by treating AI coding agents as the execution layer beneath a manager’s schedule. Despite 15+ weekly recurring meetings, seven direct reports, and a toddler at home, Agrawal says he now consistently ships two to ten pull requests per week — something he describes as simply not possible before current-generation coding agents.
The core workflow is detailed and practical: before a coding session, Agrawal uses a co-work agent (Dia, The Browser Company’s own product) to spend 20 minutes researching Slack, Notion, Jira, and the codebase, then synthesize that context into a structured prompt for Claude Code or a similar coding agent. He runs the agent overnight with explicit instructions: write tests first, verify end-to-end flows via computer use, split work into reviewer-friendly PRs, get CI green, and run an AI code review pass before surfacing results in the morning.
Agrawal argues that this pattern also gives leaders irreplaceable ground-truth intuition about new model families — critical when frontier models ship every three months and their capabilities, prompting behaviors, and product fit shift significantly with each release. He notes that leaders are especially well-suited for agent steering because their organizational context makes every token of direction more impactful than an individual contributor’s, and because delegation skills transfer directly to managing autonomous agents.
📺 Source: AI Engineer · Published August 20, 2026
🏷️ Format: Keynote Launch
