Product Talk · Free post · Building AI products · Execution, roadmaps & process

My Team of Agents: How I Get Claude to Do Tasks While I'm Away from the Computer

Teresa TorresMay 20, 20264 min
SourceProduct Talk
KindFree post
PublishedMay 20, 2026
Originalproducttalk.org ↗
N:

Teresa Torres describes a personal 'team of agents' built on Claude that works for her while she is away from the computer: a podcast manager, a sales admin, and a coding manager. Each agent prepares tasks, reports, or next steps in her to-do list, so she starts each day with work already organized. She was inspired by OpenClaw but wanted a safer, cheaper way to get similar benefits, avoiding broad machine access and runaway usage bills. The architecture rests on four parts: an agent identity file, a macOS scheduler, markdown task files, and per-agent scripts, all stored in Obsidian rather than as Claude Code skills. The approach makes the system portable across devices and across AI coding tools. The article is a partial preview, with the full setup instructions behind a paywall.

01Key takeaways

  • Give each agent a clear identity file that states its role, task folder, and context so its outputs stay focused.
  • Use your OS's built-in scheduler to run agents under your own user permissions and existing AI subscriptions.
  • Store agent instructions, tasks, and scripts in plain markdown files so you can swap AI tools without rebuilding the system.
  • Automate recurring prep and follow-up work, such as meeting briefs and next-step tasks, before automating complex decisions.
  • Have an agent review past work weekly to surface recurring mistakes and dead ends you can fix.
  • Weigh the security and cost risks of broad agent access before granting machine, browser, or credential permissions.

02Key sections

A daily routine run by agents
Torres shows how agents add prep tasks before podcast recordings and sales calls, and how a weekly coding report offers retrospective tips. These outputs appear directly in her to-do list.
Why she built it
The OpenClaw agent harness inspired her, but its broad permissions, third-party skills, and high usage costs pushed her to look for a safer alternative.
Four core components
The system uses an agent identity file, a macOS LaunchAgents scheduler, markdown task files with frontmatter, and per-agent scripts folders.
Storage and portability
Identity, tasks, and scripts live in Obsidian and run on an always-on Mac Mini, so the setup works across devices and can switch between Claude Code, Codex, or other CLIs.
Disclosure and paywall
She notes she has no compensation from Anthropic for the series, and the detailed step-by-step setup is reserved for paid subscribers.

03From the post

“Every morning I wake up and Claude has already added several tasks to my to-do list. If I have a podcast recording on the calendar, my podcast-manager agent (powered by Claude) prepares a podcast-interview-prep task with a summary of who I'm interviewing and what they”

“Every morning I wake up and Claude has already added several tasks to my to-do list.”Teresa Torres · Product Talk
“I wanted a safer way to explore the advantages of OpenClaw while also managing the costs.”Teresa Torres · Product Talk
“Agent identity, tasks, and scripts are saved in Obsidian—not Claude Code skills or agents.”Teresa Torres · Product Talk

04Frameworks mentioned

Summary and takeaways written by PM Atlas; quotes are short excerpts. © the original author.