How a principal engineer at Adobe uses parallel agents and custom skills

In this livestream, Lars walks through how he went from juggling split terminal panes with agents running in skip-permissions mode to managing everything from a single Intent workspace with notifications and organized task switching. He demos a real workspace where 60 agents were delegated to merge two POC branches into a consolidated WebRTC-powered browser agent, shows his .claude.md configuration, and walks through his library of custom skills.

Topics covered: Intent workspace navigation, the before/after of terminal chaos vs organized agent orchestration, the coordinator/implementer/verifier workflow, ACP (Agent Client Protocol) and how Intent handles tool invocations differently from Zed, building AI-aligned git tooling that blocks agents from committing everything, Playwright CLI skills for browser automation, RTK (Rust Token Killer) for token compression, and practical advice for developers getting started with agentic coding.

Lars also shares strong opinions on when to intervene vs let agents struggle, why code formatting and documentation hygiene matter more with agents, and his "ask me once, ask me twice, give it back to you" approach to human-agent collaboration.

Try Intent: https://augmentcode.com/intent
Augment Code Docs: https://docs.augmentcode.com
Subscribe: https://youtube.com/@augmentcode

00:00 - Intro and welcome
00:11 - Lars Trieloff introduction
00:25 - What we'll cover today
00:38 - Pre-Intent terminal workflow
02:46 - Running agents in YOLO mode
03:25 - Homegrown agent orchestrator
03:30 - Terminal tabs accumulate over time
04:19 - Intent as a single workspace
04:42 - Notification system and inbox
05:04 - Agents working while you wait
05:25 - Orchestrating across different tasks
05:44 - Live bug fix attempt
06:10 - Slick browser agent project
07:05 - Chat shoutouts
08:22 - Lars' multi-monitor setup
08:40 - Split keyboard and macro board
10:10 - Switching to a new project
12:19 - AI-aligned Git tool explained
13:13 - Telling agents to read node_modules
13:51 - Can agents discover themselves
14:12 - ACP and how Intent uses it
14:43 - Zed invented the protocol
15:05 - Zed as former daily driver
15:30 - Intent in light mode, why
15:57 - Process tree inspection
18:57 - Auggie vs Zed tool delegation
19:25 - Bring your own agent support
21:03 - Compaction loops and story points
21:45 - The PR that's almost too big
21:55 - 60 agents delegated
22:06 - Agent tree visualization
22:23 - Multiple coordinators explained
#AugmentCode #Intent #AICoding #AIAgents #MultiAgent #CodingWithAI #ClaudeCode #Codex #OpenCode #AgentOrchestration #Adobe #AdobeExperienceManager #ParallelAgents #DevTools #ACP #AgentClientProtocol #PlaywrightCLI #RTK #RustTokenKiller #Skills #AgenticCoding #LarsTrieloff #BrowserAutomation #WebRTC #GitWorktrees #BYOA #Zed #SpecDriven #SoftwareEngineering #2026

Видео How a principal engineer at Adobe uses parallel agents and custom skills канала Augment Code