Onboard
The New
Hire.
Oct 14, 2026
The Coordinator
Does The Managing.
Claude Projects Run Your Business (Exact Setup) argues a solo owner has a capacity problem, not a work ethic problem. Most AI tools keep you in the manager seat: ask, wait, check, ask again. A Claude Project adds a coordinator. You hand over one task, the coordinator splits the task into threads, runs them, and returns a finished result, flagging you only when a thread needs an answer. The setup runs in four steps: pick one job, load context, work the task with Claude, then schedule the task.
The biggest mistake in the video: a new project named "My business." A project needs a North Star goal. Ongoing projects never end, such as a weekly newsletter. Fixed projects end on a date, such as a course launch. He names the project, writes one goal line, and skips context at creation. He recommends the desktop app.
The coordinator on the left decides which threads to open. Threads on the right are individual agents doing the work. Settings hold a separate model and effort level for each: the coordinator runs lower, threads run higher. The feature rolls out in batches, so screens differ by account.
Without context, Claude acts like a smart hire who does not know the company name. The library holds up to 50,000 files. Two files go into every project: a business overview (offer, audience, team, operations, repeating work, audited tasks with yearly savings, front matter for review dates and status) and a 90-day priorities file.
For the newsletter project he adds a voice file, a life stories file, a copy email skill, email format references, and a Markdown export of recent campaigns with stats and body copy. His test: what would a human employee need to produce the result you already know works?
His first prompt asks Claude to read every file in parallel through subagents, read the business overview before any work, and write memory rules. Projects keep two layers: instructions you write, and an auto memory Claude writes during work and reads before every new chat.
Claude saved rules and returned three gaps: referenced files never uploaded, a voice file nobody needed, and an unexplained list drop from 45,000 to 9,000. The drop was a deliberate re-engagement test to avoid spam flags. A thread waiting on an answer shows as blocked.
One thread, four subtasks. The draft leads with Claude catching a conflict in his own files: 30,000 subscribers in one file, 45,000 in campaign stats, 58,000 in reality. He rewrites the preview line, picks a subject from alternatives, and sets a 48-hour reminder for the video link.
By hand, the newsletter takes him 60 to 90 minutes. Inside the project, about 15 minutes. At one or two issues a week, the gain covers the setup.
The coordinator turns the session into a routine: Saturdays at 8:47 a.m. ET. Each run reads his files, checks the week's videos, asks for a topic, pitches three angles, and drafts in his voice. Drafts only. He sends. The panel shows the trigger, next run, and full instructions.
He adds access as trust builds. A support inbox with an FAQ and past replies runs close to hands-off with review before send. The newsletter stays hands-on. The video closes on a paid course for building his context library.
The four steps, as taught. 01 Create a project with a clear goal. 02 Add context. 03 Complete a task with the project until the result matches what you want. 04 Put the task on a schedule so the process runs without you starting from zero.
Onboard
The New Hire.
You treat AI like a search box and get answers built for nobody. Treat the project like a new hire instead. Give the hire one job with a number attached, two onboarding files, an interview before the first task, supervised reps until the work ships, and a standing shift. Your weekly work then starts from a draft built on your business, not a blank page.
Early career, solo or junior, no assistant, explaining the same business to an AI from scratch every week.
If you hired someone Monday, what would you hand them before the first task? Why does your AI get less?
Every Chat Starts
On Day One.
You started the job, the business, the side practice. You do everything: the content, the research, the inbox, the follow-ups. You use AI every day, and every chat opens the same way. Explain who you are. Explain who you serve. Paste last week's draft. Correct the tone. Fix the mistakes you fixed last week.
By Friday you spent more minutes briefing the model than the model saved you. Better prompts and newer models leave the leak in place, because your context still lives in your head and dies with each chat. The model is a new hire on day one, every day.
Claude Projects Run Your Business (Exact Setup) shows a different order. Pick one job. Load the files a hire would read. Let the model ask questions before working. Run the task together until the result ships. Then schedule the walkthrough.
The setup takes an afternoon. The onboarding files carry the load every week after.
One Project.
One Job. One Number.
The first warning in the video: a project named "My business" has no finish line, so the coordinator has nothing to aim at. A hire gets a job title before a desk. Your project gets one job before a single file.
Pick the job you repeat most and resent most: the weekly newsletter, the client status update, the listing description, the monthly report. Decide the type. An ongoing job never ends. A fixed job ends on a date. Then write the goal line in one sentence with an output, a cadence, an audience, and one number to track.
Weak versus strong
Rule. If two goal lines share a project, split the project. One project per goal line.
Write The Two Files
Every Hire Reads.
The creator's top recommendation for every project: a business overview and a 90-day priorities file, read before any task. The overview tells the model who you are, who you serve, and how your week runs. The priorities file stops generic best practice and points every suggestion at this quarter's goals. Then add a job kit, so the model sees what good looks like.
The video shows a library with room for 50,000 files. Room is not a goal. His own files disagreed on his list size three different ways. Every file you add is one more place for a number to go stale.
Rule. Upload a file only if a new hire would need the file to do this job. Date every file. One clear page beats ten loose ones.
Let The Hire
Interview You.
Before the first task, the creator told Claude to read everything in parallel, save memory rules, and report back. Claude returned a gap list: referenced files never uploaded, a file nobody needed, and a number nobody explained. A good hire asks questions in week one. Your project should too, and you answer every question before assigning work.
What to do with the answers
Rule. No first task while the gap list holds an open item.
Work The Shift
Together First.
The creator scheduled nothing on day one. He ran the newsletter by hand inside the project, read the draft, rewrote the preview line, chose a subject from Claude's alternatives, and asked for a 48-hour reminder to add a missing link. His argument: you cannot trust a schedule for work you never watched. Run the job inside the project three times before you automate any part.
One supervised run
The log
Track three numbers per run: minutes from first prompt to finished output, edits you made, and whether the output shipped. The creator claims 60 to 90 minutes by hand down to about 15 inside the project. Treat his number as a target and your log as the proof.
Rule. Three runs with a falling edit count, then schedule. A rising count means a file needs fixing, not more runs.
Schedule The Walkthrough,
Not The Send.
After a clean run, the creator asked the coordinator to turn the whole session into a routine. Claude reviewed how the draft came together, checked his time zone, and set Saturdays at 8:47 a.m. ET. Each run reads his files, asks for a topic, pitches three angles, and drafts. He still sends. The routine is a standing appointment with the work, not autopilot.
The access ladder
Rule. Move up one level after four clean runs. Move down one level after any wrong send.
Andre's Thursday
Email, Onboarded.
Andre is a first-year real estate agent with 140 past clients and sphere contacts, no assistant, and a weekly market email he skips two weeks out of four. Before the project: about 75 minutes per email, three different tones, no record of what worked.
Planning estimate. About 75 minutes per email down to about 20 (10 of review, 10 of stat checks). Week four sample: 118 delivered, 9 replies, 2 valuation requests. Planning figures only. Log four Thursdays and replace each number with your own.
The files, written
The Friday check / ten minutes
Read The Fine Print.
Seven Days.
One Hire.
Name the job. Write the files. Answer the questions. Work the reps. Then hand over the shift.
Onboard
The New Hire.
Join The
AigentLab.
Build the onboarding files with people building the same way. Weekly field guides, working sessions, and the templates from every issue.
Join the AigentLab ↗