How This Ex-Meta L8 Engineer Ships 40 PRs a Day with AI Agents
Kun Chen quit big tech and now ships more code in a day than most engineers ship in a month — by building three tools that move him almost entirely out of the loop.
June 7thA complete 45-minute course walking from zero to a fully personalized, Google-integrated, cron-powered Hermes agent running on your phone and Mac.
An always-on AI agent becomes genuinely useful not when you first install it, but when you connect it to your actual calendar, email, and messaging apps and give it scheduled routines that reach out to you proactively.
Hermes is an AI agent that lives in your messaging apps and gets smarter over time through persistent memory and auto-built skills. The course walks through every setup step: download the desktop app, connect GPT (the recommended model for cost and rate limits), configure a Telegram bot via BotFather, authenticate Google Workspace through a manual but one-time OAuth flow, enable voice replies with a single chat command, and finally create cron job routines that proactively send briefings and reviews. The key mental model is build the skill, test it, then schedule the cron — applied to everything from weekend activity planners to weekly business reviews pulling data from Gmail, YouTube, Granola, and a fitness app.
Sign in and you get 23 free chat messages on us — ask for the hook, quote a framework, find the exact transcript moment, generate a markdown action plan. Bring your own key when you want unlimited.
Create a free account →
Opens with the promise (complete beginner-friendly course), explains why Hermes beats competitors (reliability, lives in messaging apps, persistent memory, proactive cron jobs), and walks pros/cons of Mac mini vs. main laptop vs. VPS. Recommends Mac mini.

Download the desktop app, drag to Applications, hit Install. Walks through model selection: GPT at $20/month recommended (cheap, good rate limits), Claude API too expensive, Hermes subscription overkill. Connects GPT-4.5 and verifies it's actually using the right model in the picker.

Creates a Telegram bot via BotFather (/newbot → bot token), gets own user ID via @userinfobot, pastes both into Hermes desktop, restarts gateway, sets /sethome so Hermes knows where to send cron job messages.

Explains the five-stage loop: user message → gateway routes → model plans → tools do the work → memory + skills improve. Distinguishes Hermes from coding tools like Codex/Claude Code — Hermes is for on-the-go chief-of-staff tasks, not software builds.

Have Hermes interview you to populate user.md with your name, goals, and preferences. Then update soul.md with a custom personality prompt (sourced from behindthecraft.com) so it has opinions, avoids filler, and earns trust through competence rather than performance.

Two default fixes: run 'hermes doctor --fix' in terminal (self-repairs most issues), then 'hermes gateway restart'. If still broken, drop the Hermes folder into Codex or Claude Code and ask it to diagnose.

Enable voice with one chat command. Hermes uses Edge TTS by default with 300+ voice options. Tips: turn off streaming (responses appear all at once), suppress system messages for a more human feel, use 'only reply via voice until I say otherwise' for hands-free walks.

The most complex section. Steps: Google Cloud Console → create project → enable Gmail/Calendar/Drive/Docs/Sheets APIs one by one → OAuth consent screen setup → download JSON → drag JSON into Hermes chat → authenticate via Google sign-in → copy the broken localhost URL back into Hermes. Live demo: reads Gmail inbox, schedules a calendar event, updates a vacation Google Doc with Mexico City itinerary.

WhatsApp needs a second phone number (use Google Voice). Slack enables threaded multi-command conversations. Granola links with a single MCP line for meeting notes. GitHub for backing up your Hermes folder. YouTube/yt-dlp for transcript extraction in research.

Three-step pattern: build the skill in plain English, test manually until output is good, then schedule the cron. Demos: weekend family activity planner (Bay Area, every Friday 8am), morning briefing (calendar + recent emails → top 3 action items), business review (Gmail + YouTube outliers + Substack + Granola + Google Docs), health review (Withings scale API + custom MCP fitness app → weekly email).
An AI agent's usefulness is gated entirely by the tools you connect it to — raw chat capability is table stakes; the lever is integration depth plus scheduled proactivity.
“My goal is to make this the most beginner friendly and comprehensive Hermes course on YouTube.”
“I used to be on OpenClaw and it broke with every update, but ever since I switched to Hermes over a month ago, it hasn't broken a single time.”
“It truly becomes a useful personal assistant.”
“The sky's really the limit with the automations and routines that you can set up.”
See every word as it's spoken — crank it to 2× and still catch all of it. The same dual-channel trick behind Amazon's Kindle + Audible.
One afternoon and a few tedious Google Cloud Console screens stand between you and an AI that reads your email, schedules your meetings, and sends you a weekly business review while you sleep. This full-course breakdown walks every painful step — and flags the three places most people get stuck.
The three-step pattern for creating any reliable cron job routine in Hermes. Never schedule something you haven't verified produces good output first.
The internal cycle that makes Hermes more useful over time — each interaction updates both memory and auto-built skills.
Give Hermes its own Gmail account and Mac username (separate from yours), with read-only access to your calendar/email and limited write access to specific Drive folders. Limits blast radius if something goes wrong.
“you can also find the full course on behindthecraft.com — it has a bunch of personal skills that I've set up that you can easily copy and paste and AI tool discounts”
Soft close after the content ends. Mentioned twice (mid-video and end). No hard pitch; the newsletter is positioned as the extended guide with copy-paste prompts and discounts on tools used in the video.
00:01
00:50
01:23
01:57
02:30
03:00
03:25
04:23
04:30
05:18
05:51
06:10
06:58
07:36
08:02
08:39
09:18
09:46
10:02
10:39
11:26
11:59
12:33
13:06
13:40
14:18
14:52
15:37
15:52
16:27
17:17
17:19
18:14
18:41
19:20
19:48
20:10
20:55
21:29
22:02
22:46
23:09
23:28
24:19
24:50
25:23
25:57
26:30
27:04
27:37
28:11
28:44
29:18
29:51
30:25
30:58
31:44
31:53
32:22
33:23
33:45
34:19
34:52
35:26
35:59
36:33
37:06
37:54
38:10
38:47
39:20
39:54
40:41
41:01
41:34
42:08
42:33
43:22
43:48
44:22Kun Chen quit big tech and now ships more code in a day than most engineers ship in a month — by building three tools that move him almost entirely out of the loop.
June 7thA 13-minute walkthrough of the four plain-text files that give an AI enough context to provide genuinely useful life and career advice.
June 17thA 47-minute conversation with Matt Van Horn, who ships viral open source projects without reading code or even the plans his agents write.
June 14thA 19-minute live build showing how to make Claude Code skills that grade their own output, remember past sessions, and get better every time you run them.
June 3rdJosh Pigford built and sold Baremetrics, now runs five AI products solo — and his Claude Code skill stack is the most systematic one on record.
May 31stThe VP of Product Engineering and lead PMM at HeyGen walk through the complete workflow for generating polished AI launch videos using HTML, CSS, and JavaScript — at zero cost.
June 21st