Back to Podcast Digest
Alex Finn13m

Hermes Agent just got WAY more powerful

TL;DR

  • Hermes' new session recall fixes the memory problem Alex hears about most — he shows Hermes pulling a May 10 deep equity research session from two months back, including the original tweet with 40 tickers across 10 themes, without using token-based memory recall.

  • Background tasks turn Hermes into a real multitasker without spinning up a swarm of agents — Alex queues research on AI agent startups, his last 50 newsletters, and trending AI YouTube videos, then casually asks for the Spurs game time while all of it keeps running.

  • The Grok integration is doing double duty: cheaper access and real-time X research — if you're already paying for Grok through X, Alex says you can use that auth inside Hermes and route tweet-related queries to Grok 4.3 for live post and trend discovery.

  • Native Codex CLI gives Hermes a serious coding upgrade and can save money on expensive models like Opus — Alex has Hermes use Codex in the background to build a 3D first-person shooter in a single HTML file while the main agent keeps doing other work.

  • Computer use makes Hermes feel like an actual remote operator, even if it's still a little rough around the edges — Alex has it inspect his Notion calendar and create a new event hands-free, though it schedules the item for 6:45 instead of the requested 7:00 p.m.

  • Auto-kanban task generation is the most 'autonomous employee' feature in the bunch — Alex drops a high-level goal like 'script a Hermes master class and design a thumbnail' into triage, and Hermes instantly breaks it into subtasks and assigns them to sub-agents while he goes to make coffee.

The Breakdown

The update that made Alex say Hermes is now ahead of OpenClaw

Alex opens with zero hesitation: this is Hermes Agent's biggest update yet, and in his view it's the one that officially puts it ahead of OpenClaw. He frames the whole video around eight changes that have materially changed how he uses the tool, with the bigger promise being a true 24/7 autonomous employee.

Session recall finally gives Hermes a real memory

The first feature is session recall, which Alex clearly thinks is a game changer because memory is the complaint he hears most about AI agents. He asks Hermes what they were working on on May 10, and it returns a detailed summary of a deep equity research project, including a tweet with 40 tickers across 10 themes, all without relying on token-heavy memory tricks.

Background tasks make multitasking feel native

Next he shows off background tasks with the /background command, and this is where Hermes starts to feel less like a chatbot and more like an operator. He sends it to research top AI agent startups from the last 30 days, analyze his last 50 newsletters, and find trending AI YouTube videos, then interrupts it to ask when the Spurs game starts — Hermes answers immediately while the other work keeps running.

Grok 4.3 becomes both a budget play and a live X research tool

Alex groups two updates together here: full Grok 4.3 support and the ability to search real-time tweets with it. His pitch is practical: if you're already subscribed to X, you can reuse that auth inside Hermes, and even if Claude or ChatGPT is your orchestrator, you can still call Grok as a specialized 'muscle' for finding live posts, trends, and news.

Native Codex CLI turns Hermes into a stronger vibe coder

Then he gets to native Codex CLI support, which he calls the best vibe coding tool available right now. His example is very Alex Finn: he tells Hermes in the background to use Codex to build a 3D first-person shooter with Three.js in a single HTML file, while he keeps chatting and researching in parallel, emphasizing both better coding results and lower cost versus making Opus do all the heavy lifting.

Computer use lets Hermes click around your machine for you

This is the most visibly agentic section: Hermes can now control the computer itself. Alex asks it to inspect his Notion calendar, identify meetings with Ben, Angel, and Allison, and add a new event called 'film video' for 7:00 p.m.; it successfully creates the event without him touching the mouse, though it lands at 6:45, which he laughs off as a sign the feature is powerful but not perfect yet.

Native video generation keeps everything inside Telegram

Hermes can now generate video natively through chat using Grok/Gro Imagine, which Alex loves because it avoids bouncing between tools. He prompts it to create a short video of a dragon fighting a horse, and a red-dragon-versus-white-horse clip comes back right inside Telegram; his point is less the specific use case and more the convenience of having text-to-video and photo-to-video built straight into the workflow.

The kanban board now breaks big goals into sub-agent work automatically

The final update is auto-kanban task generation, and it's the one Alex seems to use most in daily life. He drops a high-level goal into triage — script a Hermes master class and design a thumbnail with Gro Imagine — and Hermes immediately explodes it into smaller tasks, assigns them to different sub-agents, and gets moving; his routine is to dump in his morning to-do list, make coffee, and come back to 40 subtasks already in motion.

Share