Three months, one person, and an army called Claude Code: How I used AI to build the foundation of a season-long career

Looking through the system records, the first time I used Claude Code as my main tool was April 6, 2026. It has been just over three months since I wrote this.

The objective figures during this period are as follows: more than thirty active working days, more than two hundred work sessions, and nearly fifty projects. On average, almost every day I actually sit down to work, I open six to seven sessions, and one session can often complete a task.

I want to keep a good record of what I have done in the past three months. Not to show off the quantity, but because when I spread out the things scattered in various folders, various domains, and various records, I realized for the first time: It turns out that one person, with the right set of working methods, can build so many foundations in one season.

Through this article, I want to record my digital footprints and the stories behind these processes.

Let’s look at the timeline first: four months, four rhythms

First of all, these three months are not evenly applied, it has obvious tides.

**In April, the foundation was laid. ** The main focus of this month is to institutionalize and automate the daily operation and release process. A set of workflow skills for a one-person company (opens in a new tab) was put into its first official version and became the matrix for all subsequent content products; the structure of a virtual engineering team was also formed, establishing my core working method: The commander does not leave the field, and the heavy work is assigned to the clone. At the end of the month, I completely tore down one of the sites and rebuilt it, changed the technology stack and deployment platform, and built a complete safety net for cross-site deployment: a check tool, a stack of troubleshooting manuals, a gatekeeper clone, and an automatic monitoring mechanism after deployment, all stacked up in four layers.

**In May, add what can be sold. ** Now that the foundation is in place, the product begins to grow. A set of skills for Academic Writing (opens in a new tab) has been launched one after another, and the tool chain for visuals and presentations has been completed, including a framework for converting web pages into video frames and a generator for converting papers into posters. This month I also did my first serious cost physical examination, taking stock of fixed expenses and cutting off redundancies.

**June, dive. ** From mid-May to late June, system activity was obviously sparse. At that time, I spent more time teaching and running business outside. It returned intensively at the end of June, and the theme turned to content quality and material management: establishing the principles of film editing, repairing a batch of pitfalls in the publishing process, and establishing a workflow for material libraries.

**In July, it broke out. ** This is the most dense month, with almost a new system every day. Content production has been upgraded from scattered single-point tools to managed production lines; the business command team has grown an AI team that will review each other; advertising has gone from zero to systematic, and the data connection, delivery system, and material pipeline were all built within a month, and I actually invested in my first Facebook Advertisement; three dashboards were launched one after another; the first practical battle of an evolving lesson preparation system, all ten quality levels were passed.

Work inventory: What have been built in the past three months?

Classify things according to their properties, roughly into these groups.

**Self-operated website group. ** From the main website blog, the content course store of the one-person company, the AI ​​content engine brand site, AI Builder Academy (opens in a new tab) for non-engineers, to the operation dashboard, investment stock selection panel, AI laboratory (opens in a new tab). These sites connect a complete funnel: the media attracts traffic, the list accumulates relationships, the store completes the transaction, and the consultant accepts high prices, and most of them are actually online.

**Commission and customer site. ** In the past three months, I have also received several external website building projects: a lecturer’s personal brand website, a functional textile brand’s B2B official website, a personal brand operator’s image website, and a traditional production company’s official website redesign prototype. They prove one thing: using this method, website building has changed from a self-use skill to a service that can be charged externally. One of the most unforgettable scenes was taking a group of local folks who had no written procedures at all and spending an afternoon making their own official website.

**Content Machine. ** The scale of new content added in the past three months is far greater than I imagined: the main site has added more than a hundred original articles, each of which is automatically generated in English and Japanese versions; the other three sites have accumulated hundreds and dozens of articles respectively. What I underestimate even more are the presentations: there are nearly a hundred web presentations, including a complete series of courses, and most of them first write the planning document and then produce it. It has become a habit to plan first and then start.

**Tool kit available for sale. ** Among the many gadgets I make, there is a clear path to productization: an academic research workbench, a workflow suite for a one-person company, a startup package that generalizes and packages your own system files, a card note-taking tool, and a video note-taking tool. They all follow the same model: first use them for your own use, make them easy to use, de-identify them, and then put them on the shelves at a price. I also made two plug-in services for AI. One is designed to intercept the drift of simplified and traditional terms and punctuation before the content is sent out, and the other turns the management of research documents into a workbench that AI can directly operate.

**Videos, Visuals and Dashboards. ** An open source framework for converting web pages into videos, which has led to several short video and audio projects; a set of typesetting tools that include PDFs, landing pages, and posters. In July, I also made three dashboards in one go, reflecting a consistent impulse: to consolidate scattered information into a panel that can be understood at a glance.

(opens in a new tab)

The “Academic Research Workbench” mentioned above is one of the ways I polished my own tools into courses. If you are a graduate student, teacher or knowledge worker and want to learn how to let AI truly take over the chores of documentation, writing and submission, welcome to take a look at my “AI Empowering Academic Research and Writing Practical Workshop”.

More important than the work: four things I learned

If you only keep numbers, this record will be wasted. What’s really interesting are a few recurring patterns.

**First, self-used workflow is the best product prototype. ** Almost everything that can be sold in the past three months was not designed out of thin air, but I used it every day. After using it smoothly, I generalized, de-identified, priced and put it on the shelves. Solving your own problems first is the most honest product research.

**Second, stepping on thunder means institutionalization. ** I have accumulated a large stack of notes, most of which are not knowledge but the result of accidents: a certain setting has an upper limit, a certain cache will cheat verification, a certain step will print a false success message. I didn’t keep these lessons in my mind, but wrote them down as files and stored them in the system, so I wouldn’t step into the same trap twice. This is how the moat was dug out with shovel after shovel.

**Third, one person can also have a team of mutual reviewers. ** From the virtual engineering team at the beginning, to the AI ​​team that will review each other, and then to the monthly audit to review the frequency of my intervention, my working style has evolved from “I direct the AI” to “the AI ​​team reviews each other and regularly goes back to review the number of times I intervene.” The one who does the work does not check his own work. This discipline makes the quality no longer rely solely on my attention. This also echoes something I have always believed: Stop defining your company by headcount.

**Fourth, infrastructure construction must be connected from points to lines and then to surfaces. ** The deployment of this matter was just a verbal rule in my mind in April, and later it became a four-layer system of tools, manuals, gatekeepers, and automatic monitoring. Every time there was a failure, I used the manual to feed it, and the more I used it, the smarter I became. The real leverage is not how fast you can do something one time, but the infrastructure that turns each experience into the next.

Write at the end

Three months ago, Claude Code was just an AI tool that could help me write programs. Looking back three months later, it is more like the army that has finally been replenished in the company that I built alone: ​​it builds websites, writes articles, makes presentations, edits videos, places ads, and manages data. And for the first time, I can free my hands and concentrate on making those judgments and choices that only I can make.

This list is not an end-point, it is an inventory. When you see the foundation clearly, you know where to build the next floor.

If you are also using AI to build your own business, I would like to give you the most important lesson from these three months: Don’t rush to pursue how fast you can do it in a single time, but find ways to turn each experience into the foundation and nutrients for the next time.

---

Want to build your first foundation with your own hands?

I have organized the introductory part of this “self-use first, then productization” method into AI Builder Academy (opens in a new tab). If you also want to never know how to write programs and practice all the way until you can use AI to help you build websites, make products, and grow your business, welcome to join me and turn every failure into steel for the next level.