How to automate a $1M/yr business with Claude AI

TL;DR

Auditing your work and building AI driven skills with Claude can automate most non customer facing daily tasks. Start by recording what you do, ask Claude questions to fill gaps, score outputs, then institutionalize high scoring tasks as scalable skills. Schedule tasks that truly run autonomously and keep human oversight where needed.

Transcript

So, I gave Claude Code access to a business of mine that does around a million dollars a year, and I asked it to automate everything. And uh it actually did a really good job. And this isn't a clickbait or one of those like hype beast videos where I just talk about how it's automated my whole life, and everything's amazing. Um I've tried this many,... Read More

Key Insights

• Auditing work requires capturing daily activity with a tracking system to create a data driven foundation.
• Claude is used to generate, test, and refine automated skills by running multiple iterations and scoring results.
• High performing outputs become reusable skills that can be shared across an organization.
• Automation focuses on non customer facing work that can run without the human in the loop.
• Batch testing multiple tasks at once speeds up the refinement process and reduces overall time.
• Some tasks should remain human in the loop due to the need for judgement and customization.
• Institutionalizing skills enables easier scaling and delegation within the organization.
• Autonomy is pursued for drafts and repetitive communications, while human oversight is preserved for complex deliverables.

Questions & Answers

Q: What is the first step to automate a knowledge work process with Claude?

The first step is auditing your work by tracking what you do, including capturing a record of your activities and context. This creates a foundation for Claude to understand the tasks, identify patterns, and determine what can be automated. The audit often includes a week of data gathering and analysis.

Q: How does the video suggest gathering context for automation?

The video recommends using a tracking tool to capture what you are working on every few seconds and to compress screenshots into a legible form. It also suggests asking Claude 10 to 15 questions during the day to clarify intent, requirements, and needed systems, ensuring Claude has sufficient context.

Q: What is an eval in Claude automation according to the video?

An eval is a test that repeatedly measures how well an AI model performs a specific task given a defined prompt. In the example, each task is tested 10 times and scored, with a goal of achieving an 8 out of 10 or higher to consider the skill reliable.

Q: Why is batching tests recommended when improving skills?

Batching tests allows parallelized changes to the skills, which saves time by evaluating several improvements at once rather than one at a time. This approach reduces the overall time to reach a high performing system and helps identify the best combination of prompts and templates.

Q: What happens after a skill scores well enough to be automated?

Once a skill scores above the threshold, it can be institutionalized as a reusable capability that can be shared with others in the organization. This enables non technical staff to benefit from AI driven automation through simpler tools and predefined processes.

Q: When should tasks be scheduled to run autonomously according to the video?

Tasks that are stable and safe to run without human intervention should be scheduled to run autonomously. However, tasks that involve drafting for clients or building complex deliverables may require human oversight or a human in the loop before execution.

Q: What is the creator’s recommended model choice for this automation workflow?

The creator recommends using the most frontier model available on max thinking, arguing that upstream planning greatly impacts downstream capabilities. Even if usage costs rise temporarily, it is worth using the strongest model to maximize long term automation potential.

Q: What is the end result of following this automation process in the video?

The end result is that virtually all non customer facing tasks can be automated, leaving the person to focus on client calls and high level system design. This leads to a higher level of productivity and a shift toward AGI like capabilities in daily operations.

Summary & Key Takeaways

• This video outlines a step by step method to automate a knowledge based agency using Claude, focusing on auditing daily tasks, collecting context, and creating automated skills that can be shared across a team.

• The creator describes a workflow that includes capturing activity through screenshots, asking targeted questions, scoring outputs, and iterating until 7 or 8 out of 10 is achieved for each task.

• The final phase emphasizes scheduling autonomous tasks, while leaving critical decisions under human review, to reach maximum productivity with AI assistance.