The Case Against Platform Lock-in

New AI agent platforms emerge rapidly, making it risky to commit to a single ecosystem. Instead of choosing one, adopt a multi-platform strategy. By separating your business logic (the "skills" or instructions) from the execution platform, you ensure your workflows remain resilient even as individual tools fall behind or change their pricing models.

Categorizing Agent Workloads

To optimize efficiency and cost, categorize your agent tasks into two distinct buckets:

  • Routine Jobs (Category 1): These are recurring, low-stakes tasks such as daily reports, SEO health checks, content ingestion, and GitHub PR monitoring. These are best suited for platforms like Hermes, which offers unlimited scheduled tasks and a superior chat interface (via Discord) for monitoring and interaction.
  • Creative/High-Stakes Jobs (Category 2): These include writing, design, and coding tasks where output quality is paramount. These should be routed to frontier models like Claude Opus. Due to Anthropic’s June 15, 2026, policy change, running these models via third-party platforms now incurs pay-as-you-go API costs, making it more economical to run these tasks directly within Anthropic’s native tools like Claude Cowork.

Building a Portable Infrastructure

Regardless of the platform, the underlying pattern remains the same: an agent follows a specific "skill" (a set of instructions) on a recurring schedule.

  • Hardware Setup: Use a dedicated machine (e.g., a Mac Mini) to host background agents. This separates automated "team" work from your daily driver machine, ensuring tasks run reliably without interruption.
  • Messaging Interface: Use Discord for agent communication. It provides superior markdown support, threading, and multi-channel organization compared to Telegram or Slack, making it feel like interacting with a human team member.
  • Skill Portability: Store your agent instructions (skills) in a centralized file system. This allows you to easily port processes from one platform to another—as evidenced by the author's migration from OpenClaw to Hermes—without re-engineering your business logic.

The "Night Shift" Workflow

Automate the "intake" of your business by having agents ingest transcripts, tweets, and journal entries into a centralized markdown-based file system. Use scheduled cron jobs to trigger agents at off-hours (e.g., 3:00 AM) to synthesize this raw data into daily summaries. This creates a knowledge base that future agents can draw from, turning raw noise into actionable business intelligence.