100 hours of Hermes Agent lessons in 19 minutes
TL;DR
Claude Opus is worth the premium: Finn pays about $40/day for Opus because it's the only model that reliably completes every task, while other models give up when they hit obstacles.
Run at least two agents simultaneously: Multiple Hermes agents can monitor and fix each other, providing crucial failovers when one breaks or loses connection.
Skip the separate accounts paranoia: You don't need a dedicated Mac Mini or Gmail for your agent; security concerns are overblown and adding friction just slows you down.
Match your platform to the task: Use Hermes Desktop at your computer, iMessage for quick prompts on the go, and Telegram for deep mobile work with rich formatting.
Do a morning interview with your agent: A reverse-prompting technique where your agent asks about your priorities uncovers 2-3 new automation opportunities every day.
The Breakdown
The Model Hierarchy: Why Opus Commands a Premium
Finn's official recommendation is Claude Opus, which he calls the best agentic model ever made. He spent $1,400 last month on Opus credits alone, averaging $40 per day, because Opus is the only model that guarantees task completion. His analogy: Opus will crawl to the finish line even if it loses a leg mid-marathon, while other models roll over and quit when they stub a toe.
Budget Alternatives That Actually Work
For those unwilling to spend thousands monthly, Finn offers two alternatives. ChatGPT 5.5 and later is now usable with Hermes (anything earlier was horrible), while GLM 5.2 is the ultra-budget option at a fraction of the price, though more robotic since it's distilled from the bigger models.
The Multi-Agent Safety Net
Finn insists you need at least two Hermes agents running at all times. His main agent runs on Opus, while his backup GPTI runs on ChatGPT. When GPTI's token expired recently, he simply screenshotted the error to Hermes, which fixed the problem automatically. This failover system means you're never stranded when one agent goes down.
Security Theater Is Wasting Your Time
Finn calls out users who buy separate Mac Minis and create dedicated Gmail accounts for their agents as making a massive mistake. Hermes only does exactly what you prompt it to do. He challenges viewers to name a single person who has had a legitimate security incident with Hermes, arguing that most fears are overblown and the added friction is unnecessary.
The Platform Playbook
Finn uses three platforms strategically. Hermes Desktop is the best user experience for computer work with easy profile switching and session pinning. Telegram handles deep mobile work with rich formatting. iMessage, a new addition, is perfect for quick prompts since it's the app he's already in all day.
Cron Job Cleanup for Better Performance
The number one culprit behind sluggish agents is accumulated cron jobs running in the background. Finn recommends opening the desktop app's cron list weekly and pausing old jobs you no longer need. This instantly boosts performance and saves money on token usage.
Tailscale Unlocks Your Device Fleet
Finn uses the free app Tailscale to create a private network across his six computers, allowing Hermes to move between devices seamlessly. He can ask Hermes to SSH into his DGX Spark, check running models, or grab a presentation from another computer and send it to his laptop. His other computers don't even need monitors connected.
The Morning Interview That Finds Hidden Automation
Finn's reverse-prompting technique has Hermes interview him each morning about priorities, stressors, and tasks. The agent then identifies what it can take off his plate, what it can automate, and what needs permission. This five-minute routine consistently surfaces two to three new jobs for Hermes to handle, which he then tracks in the built-in Kanban board.
Was This Useful?
Share
Keep Reading
Make Alcreon Yours
Tune your feedFive quick questions, and the feed ranks what matters to you first.Or just get notified
The weekly Echo. Signal worth keeping in your inbox.
Every new piece, announced on X.
Read Next
See all
Playbook
The Retirement Email Isn't a Warning
Model retirements now arrive every few weeks; the config-eval-rehearsal loop turns each deprecation email from a fire drill into an afternoon swap.
Playbook
The Cheapest Model That Passes
OpenRouter lists 400 models behind one API. The fix for choosing isn't a better leaderboard, it's a four-step protocol that ends in a real eval.
Playbook
Cheap Models, Hard Tasks
Most agent workflows route every step to the frontier model by default. The bill scales with how chatty the agent gets, even when most steps don't need that brain.