A YouTuber just automated my exact job.
Brett from The Brett Way took the role of a podcast producer (find guests, DM them, schedule them, prep research for the host) and replaced the whole thing with one Claude workflow that runs 24/7.
I run a podcast. I watched this video twice.
The big lesson has nothing to do with the tool. If you want to know how to automate your job with AI, the skill is systems thinking. Brett's argument: the people making money with AI are not the ones using the most tools. They're the ones who can break a job into steps.
Here's the full breakdown, with the exact system he built.
AI Replaces Tasks, Not Jobs
Everyone says AI is coming for jobs. Brett's framing is sharper: AI replaces the tasks inside jobs.
Every job is a stack of tasks. A YouTuber doesn't "make videos". They define a channel strategy, brainstorm ideas, write a script, shoot, edit, make a thumbnail, post.
And each task is itself a process. A thumbnail = research winning thumbnails → pick a concept → take a photo → edit it.
Once you see work like this, the AI opportunity gets obvious. AI is very good at completing structured processes. It's bad at "start me a business".
Most people prompt the second way. Systems thinkers ask: what processes create the outcome? Then they design the system around those processes and hand the system to the AI.
That's the entire skill. Map first, automate second.
The 6 Tasks of a Podcast Producer
Brett picked a real job to automate end-to-end: podcast producer. Every big show has one. (I don't. Which is exactly why I'm writing this.)
He mapped the job into 6 tasks, in order:
- 1. Podcast strategy. What's the show about, what topics, what vibe. An NBA show and a business show need completely different systems.
- 2. Ideal guest profile. Not "entrepreneurs". Brett's actual profile: 18-to-22-year-olds who have made $1M online. Scored on following size, revenue, age, and whether they've been on before.
- 3. Prospecting. Build a list of everyone who fits: name, business model, estimated revenue, follower count, and which platform to reach them on.
- 4. Outreach. Personalized DMs on X inviting them on, with a proposed shoot date.
- 5. Scheduling. Pick a time, drop a Google Meet link, put the guest on the calendar.
- 6. Research report. Background on the guest, what they've been asked on other shows, and suggested questions so the host walks in prepared.
Six tasks. Each one a structured process. Together, a system.
Notice he did this part by hand. You can ask AI to draft the map for you, but Brett is clear: the more specific you are about what you care about, the better the AI performs. The mapping is the work.
5 Claude Skills Built in Under 3 Minutes
Then the execution, and this part is fast.
Brett plays out a scenario: he runs a podcast producing agency and just signed a client who already has a show. He tells Claude Cowork: "I want to create an automated workflow to run my podcast producer agency. Build the system for my new client." Then he pastes in the client's YouTube channel and the 6-task system he mapped.
The YouTube link is the trigger. New client = paste channel, everything else runs.
Claude built 5 custom skills in under 3 minutes:
- Podcast strategy (custom to that specific show)
- Guest scouting (the ideal guest profile)
- X outreach (how to write the DMs)
- Scheduling
- Research and question prep for the host
Skills are basically saved prompts with judgment baked in. Brett calls it the new era of prompting: instead of rewriting a monster prompt every time, the agent writes its own detailed instructions once and pulls from them forever.
He runs all of this through Claude Cowork with the Chrome extension, so the agent controls his actual browser. No APIs to wire up, nothing to connect.
Then he types "run the podcast workflow" and watches.
The agent scouted 10 candidates, scored and ranked against his criteria. The top three: Zach Yadegari, Danny Postma, and Liam Ottley. It also surfaced Shane Copeland (the Polymarket founder Brett had already tried to book) and Pieter Levels.
That list is legit. Zach is the Cal AI founder who sold for millions at 18 (I broke down his full story here, from Brett's own interview with him).
Then the agent opened X in the browser, found each guest, and sent a personalized DM inviting them on the show. It handles replies and creates the calendar event too. And when a guest says yes, it produces the research doc: full background, what they've said on other podcasts, suggested questions.
One typed sentence. The producer job, done.
The Step Everyone Skips: QA Before Autopilot
This is the part of the video I'd tattoo on every founder's forearm.
The first runs had back and forth. Brett stayed in the loop on purpose:
→ Is the strategy doc actually right?
→ Are the prospected guests people you'd genuinely book?
→ Do the outreach messages sound like you, or are they cringe?
You babysit the system until each answer is yes. Then, and only then, you take yourself out of it.
Most people do this backwards. They automate a process they've never run manually, get garbage, and conclude AI doesn't work. The system was never designed. Garbage in, garbage at scale out.
Scheduled Tasks: The Job Runs Without You
Once the outputs were clean, Brett asked the agent to automate the workflow so it runs with no approvals needed.
Claude created 5 scheduled tasks. The week now looks like this:
Monday 9:00 a.m.: pipeline review.
Tuesday 10:00 a.m.: guest prospecting. The agent searches for 10 to 15 candidates and scores them against a 30-point rubric.
Wednesday: outreach to the top 3 candidates on X, plus replies and booking.
That's an entire role inside a company running on a calendar, hands-free. And because the system is templated, the next client is one sentence: "run the podcast producer workflow for my new client" plus a YouTube link.
If you're wondering what else deserves this treatment in your own company, start with the boring repeatable stuff: lead follow-up, onboarding emails, reporting. This guide to business automations is a solid map of which processes to systemize first.
Why This Matters: You Stop Selling Time
Brett's closing point is the one with money in it.
Once a system is built, you run it repeatedly. You're no longer selling hours, you're selling the output of the system. One person operating like a team.
That's why Y Combinator said they want to fund AI-native agencies: agencies where AI workflows do the work instead of large teams. Investors don't get excited about "I use ChatGPT". They get excited about a business where the org chart is mostly cron jobs.
Brett followed this video up with 7 concrete businesses built on exactly these systems. I covered that one too: 7 AI businesses you can start with Claude agents.
How I'd Steal This as a SaaS Founder
You probably don't run a podcast. Doesn't matter. Run Brett's exercise on your own week:
→ Write down every task you did in the last 7 days. All of them.
→ For each task, write the process as steps, like you're training a new hire.
→ Circle the ones that are structured and repeat weekly. That's your automation queue.
→ Build one system. Babysit it until the output is something you'd ship under your own name. Then schedule it.
The founders I interview who are past $50K MRR mostly did some version of this before agents existed. They wrote SOPs and hired humans. The playbook didn't change. The worker did.
And the mapping skill compounds. Every process you document is one more thing your business does without you touching it.
FAQ
What is systems thinking in AI automation?
Systems thinking means breaking a job into its individual tasks, mapping each task as a step-by-step process, and then handing those processes to an AI agent to execute. AI performs well on structured processes and poorly on vague goals, so the mapping is what makes automation actually work.
What tools did Brett use to automate the podcast producer job?
Claude Cowork with the Claude Chrome extension. The agent controls the browser directly, so there are no API integrations to set up. Claude generated 5 custom skills (strategy, guest scouting, outreach, scheduling, research) and 5 scheduled tasks that run the workflow on a weekly calendar.
Can AI really run outreach without sounding like a bot?
Only if you QA it first. Brett manually reviewed the outreach messages until they sounded like something he'd actually send, then removed himself from the loop. Skip that step and you're mass-sending cringe. The system is only as good as the process you mapped and the standards you enforced during the manual phase.
Which job should I automate first in my business?
The one that's structured and repeats on a schedule. Lead follow-up, prospecting lists, reporting, onboarding sequences. Don't start with judgment-heavy work like closing deals or product decisions. Map the process, run it manually once, automate, verify, then schedule it.
I interview bootstrapped founders every week about exactly this: the systems behind businesses doing $100K to $10M a year. Real numbers, real playbooks, no fluff.
