In this 22-minute walkthrough, Chase AI shares the exact Claude Code "personal assistant" setup he uses to run his agency, his YouTube channel, and his research, claiming it saves him five to ten hours every week. Unlike his previous videos, which focused on system design (skill architecture, Obsidian vaults), this one shows the specific skills and automations he actually runs day to day.
The core idea: identify the boring, redundant, low-decision "grunt work" in your life, turn each piece into a Claude Code skill, verify it works, and automate it. Do that a few times and you reclaim hours. He also stresses that most of this translates to non-creators, because nearly everyone deals with inbound email, proposals, and daily research.
Chase organizes his automations into three areas of drudgery:
He notes the productivity and research buckets translate to almost anyone; content is niche to creators.
This is the piece Chase says gives most people the fastest payoff. Every morning at 8 a.m., an email triage skill:
Claude auto-drafts a reply with a boilerplate response and a link to his media kit (pricing, etc.).
His website form captures budget, needs, and timeline. Claude Code picks that up in the triage and, if it judges the query is real (email makes sense, effort shown, budget aligns, not random nonsense):
Urgent emails get a drafted reply; other buckets are just marked. Claude writes a full markdown report into his Obsidian vault summarizing the morning, including the noise he doesn't care about.
Adapt it yourself: What emails do you get? What buckets fit? What should Claude do with each once sorted?
After a discovery call (booked via Calendly), Chase receives an automatic recap/summary of the key points. That recap is sent to Claude Code, which:
The dummy proposal shown breaks down the engagement, timeline, and investment, with a signature area at the bottom. He reviews everything, but it beats writing proposals manually. Follow-ups are tied to his AI audit service and are too technical/niche to cover here.
Research "spawns from knowing where your information originates." For AI, Chase's sources are essentially Twitter/X, GitHub, and YouTube. Find your niche's sources; the setup is the same. His main automation is a daily brief:
It also attempts content-opportunity analysis, but Chase admits that part is "hit or miss" — the real value is the raw aggregated data, especially GitHub, which would be miserable to gather manually. The daily brief and the email triage are combined into a single "morning automation," set up as a Claude Code routine that runs locally on his machine whenever it's open.
A custom app he built that lives on Railway, running 24/7. It sends him Telegram messages roughly every hour with the top trending AI tweet, tied to major creators (including people on the Claude and OpenAI teams), so he hears about big news immediately without doom-scrolling.
The built-in /deep-research command in Claude Code. Let loose, it spawns up to ~100 sub-agents doing adversarial information gathering — they find info and check it against each other, then synthesize. It's "standard web search on steroids," burns a lot of tokens, and is underused.
Searches YouTube for videos relevant to a topic, sends the URLs to NotebookLM (via the unofficial notebooklm-py CLI), and lets Google's servers do the transcript synthesis — so he gets NotebookLM's power without spending Claude tokens. The output tells him the throughline across, say, 10 videos, where they disagree, and why he should care, instead of watching them all.
Chase estimates the research setup alone saves five to ten hours. He argues productivity and research "should go hand in hand," since the emails you send and content you make are often based on research.
He spends the least time here (least relevant to most). Key theme: AI as a back-and-forth brainstorming partner, not a replacement for taste.
Getting the voice right takes a strong human-in-the-loop cycle. Chase's method: give Claude examples of your own writing, have it build a skill, let it produce an example, then "absolutely eviscerate it" — mark what's wrong and right — feed the corrected version back, have it update the skill, and repeat ~10+ times, then keep correcting live. There is no one-shot "self-improving" skill for voice; you must be in the loop.
Individual skills save hours, but a coherent system needs structure — a "brain" so Claude can reference past work and feed a single dashboard. Obsidian doesn't change how Claude Code works (the knowledge graph is just "a fancy visual"); its value is organizing everything the assistant produces into a map Claude can navigate. Chase describes the Karpathy method, imagining everything as folders:
Why it matters: without organization, Claude falls back on grep, which increases token cost and decreases accuracy as the file structure grows. The vault is essentially a filing cabinet for Claude Code.
The second half of the system is observability: a place to see everything in one spot. The terminal and Claude Code desktop app aren't built for that, so Chase uses custom command centers and a web-app dashboard, which must be 100% customized to be valuable. His dashboards show metrics, one-click runnable automations, an audience/morning-report breakdown, and document access to pull up reports — all connected to the same data as his Obsidian vault.
How the pieces fit: dashboards give observability, Obsidian keeps it organized, and the skills and automations do the actual time-saving work of a personal (or executive) assistant. Put together, Claude works for you nearly hands-off.
"Claude should be running your life, or at least all the boring parts."
"What is the boring redundant work that we can automate and turn into skills?"
"I'm still the arbiter here, whether we go forward or not."
"You're going to absolutely eviscerate it... you do this over and over again for like 10 times... that's really the only way to do it."
"This is essentially a filing cabinet for Claude Code."