This article went from concept to published webpage on an afternoon walk. I gave my AI assistant the task from my phone, walked, stopped once on a park bench to review the draft and approve it, and pushed it live. Concept to published page, on a walk. I was not texting with my head down in the middle of the street — mostly I was just walking and thinking while the AI handled the administrative and technical work of getting it online. For me at least that is the waste part I don’t enjoy. The actual thinking came from pieces I previously authored on my computer in various forms and two conversations I had audio recorded with colleagues. What follows is what the AI assembled, formatted, and published to my website — with my direction and editing. 

This is not science fiction. This is a random Tuesday afternoon in 2026. 

How I Got Here

About two years ago I started experimenting with building AI-powered problem-solving applications using traditional software development tools on my desktop. My first application was 50,000 lines of code with authentication systems, special databases, and hundreds of hardcoded prompts. A few of my colleagues in the national laboratory systems loved it. Most everyone else not so much. It worked, but it was difficult to maintain. 

Then about 12 months ago the large language models (LLMs) got better. I rebuilt the same problem-solving coaching functionality with 5,000 lines of code. Instead of one monolithic application, I built about a dozen smaller web apps for different purposes. They worked just as well and were easier to maintain. 

Then suddenly over the most recent New Year’s holidays the models improved again. Today, instead of using web apps, I get similar results chatting with AI through a custom setup on my desktop via my cell phone — zero code. Reflect on that for a moment: 50,000 lines of code to 5,000 lines of code to zero, with virtually the same quality of output and far more flexibility and customization potential. I’d call that kaizen any day of the week. Just a different kind than I am used to. 

In software, knowledge work, and cognitive tasks, something interesting is happening. Work that seemed too difficult in the past to attempt is suddenly easier than before. Tasks I had put off for decades I suddenly completed as weekend projects. I did not see this coming. It only made sense in hindsight, after I tried something, reflected on it, and thought: what else could I do? 

What I Actually Have Now 

My primary computer setup includes a local note-taking application called Obsidian that stores everything I write as plain text files — research, project plans, consulting notes, article drafts. This all sits on a Linux server at home as that is more secure and inexpensive to operate. Remote access is also easier. 

In total the setup has a couple of years of accumulated thinking in a local database. Alongside that is my regular messy 6 TB hard drive with almost 40 years of accumulated files and notes. A personal AI assistant can search, read, and write to those files, search the web, and take multi-step actions when I give it permission. A cell phone connection through the messaging app Telegram lets me talk to the AI from anywhere including my walks in the park. 

Inside the vault I furthermore have special “skills files” that describe how to perform certain types of work. Think of these as standardized work files for knowledge topics: how to do something, what good looks like, what mistakes to avoid, how to structure the output. When the AI reads a skills file it behaves accordingly. To change how it behaves, I edit the text file. No code. No deployment. No special development required. 

The advantage of the whole setup is that everything is just plain text files I own. No expensive software subscriptions. No proprietary platform locking up my data. And because things will change in 12 months, it is portable and maintainable. I am not vendor-locked into some type of ERP, Salesforce, Slack, or fill-in-the-blank custom software solution. I can take it anywhere I want next time as needed. 

Decades of Procrastination, Done in Weekends

This setup has liberated me in a way I did not foresee in the beginning. Let me give you three examples of things I had put off for years — in some cases decades — that this setup made possible. I thought about these three projects for years, but never got around to tackling them due to the perceived difficulty involved: 

• I moved my website off of WordPress which I had been meaning to do for a decade. 

  I built a structured eight-step problem-solving report around the famous Taiichi Ohno 5 Why example. 

  I built a custom 110-term TPS (Toyota Production System) encyclopedia based off of material on my website and hard drive. 

Moving Off WordPress

 My website artoflean.com had been on WordPress for 20 years. Publishing anything required a long and painful process of logging in, fighting the editor, formatting content, entering various fields of information, managing plugins, and dealing with a system that was not designed to work with AI. I kept meaning to post things and rarely did anymore. 

Then one weekend I used AI to migrate the entire site. It exported the content from WordPress, converted it to a new format, and rebuilt the site on a modern platform that works natively with AI tools. I thought the migration would take two weeks. It was done in two days. AI typically did the first 80% for each step, and I did the finishing touches. Once I had a platform where publishing was easy and enjoyable again, a lot of other things became more feasible to attempt. 

The Taiichi Ohno Problem-Solving Report 

Everyone who subscribes to Lean.org’s content likely knows the famous Taiichi Ohno 5 Why example — the one where a machine broke down on the production floor and the root cause was cutting chips penetrating the lubrication system. It is one of the most referenced examples in lean thinking. But it has bothered me for years. The 5 Why sequence became larger than the actual problem-solving process it was a part of. And somehow that further lost the entire context of an actual problem being solved and the organizational development of Toyota in the 1960s. 

I wanted to rebuild the example as a proper eight-step problem-solving report — a historical re-creation of the actual event with visual analysis, root-cause verification, and countermeasure logic laid out end to end. I also wanted to make it a web page with a button at each step so readers could click for further historical context about what was happening at Toyota at the time. That kind of interactive teaching tool would have taken me weeks to build in the past. With AI I had it 90% done in two hours.

I started the first draft on my phone on one of my walks. AI generated the initial structure based on my skills files — plain text documents that describe what a good problem-solving report looks like, what mistakes to avoid, and how each section should connect to the next. I edited the final version on my desktop, challenged the AI’s reasoning where it was weak, asked John Shook (LEI Senior Advisor and fellow Toyota veteran) to add some organizational context, and pushed the final version to my website. The report is far from perfect, but this would have been just too difficult to justify a year ago. 

A 110-Entry Toyota Encyclopedia 

Once publishing was easy, a bigger idea clicked in my mind. I have been collecting Toyota source material for over four decades — internal training documents, research papers, reference guides, and notes from my years working there. Some of the most valuable materials are in Japanese and have never been translated or published in English. That material sat in my files for years. The work of translating, organizing, cross-referencing, and formatting it into something publishable was simply too onerous to contemplate. 

So as a test I decided to build a Toyota “TPS encyclopedia” from those materials. AI processed source documents, organized content, and generated structured entries. I reviewed everything, corrected some entries, and shaped the final result based on what I know about how these systems actually work in practice. AI did the grunt work. I provided the judgment. The initial result is 110 entries, each with Japanese terminology, historical context, and practical application. I never would have done that before. 

"AI did the grunt work. I provided the judgment."

Two sample entries are quite personal. You don’t hear much about them in English, but in Japanese they are a lot more common. Their roots stem back to when I was working in the Overseas Engineering Support Group at Kamigo Engine Plant, where we structured fundamental TPS shop floor activities for work teams in production. Toyota’s FMDS (Floor Management Development System) and 3 Pillar Activity (Sanbon-bashira Katsudō) are both foundational to how Toyota manages the shop floor. The source material for both entries was entirely in Japanese material on my computer. I also had hand-scribbled notes from when I worked at Kamigo, where the roots of those systems were designed. I always had the content, but I never had the time or energy to translate and publish it. Now it is available to anyone. 

I have hundreds of other files and documents in my system. AI now makes it realistic to extract the content and put them into the world with minimal difficulty. And once the architecture exists, customization becomes easy. A version of that encyclopedia tailored to a specific company — their terminology, their context, their problems — is no longer a months-long project. Similar company repositories of specialized knowledge are going to flourish in the years to come. 

What Makes This Work 

The model provides general intelligence. The skills files provide domain expertise. The human provides the direction, refinement, and quality control. The system gets better through use, the same way a good standard does. When someone finds a better way to coach a particular step or structure a particular output, you update the file. In the past encoding this kind of knowledge into a tool was relatively hard to do. It just got a lot easier for any type of knowledge work — the work I wanted to accomplish for years and the unfeasible future projects that will soon be possible. 

Current Capabilities 

Beyond writing articles and reports, the practical range of this application is wider than most people realize. 

A cellphone photo of a production dashboard, a handwritten A3 report, or a defect on the shop floor can go straight to the AI and come back with guided observations within seconds. The current generation of vision models handles this well. 

"The model provides general intelligence, the skills files provide domain expertise, and the human provides the direction, refinement, and quality control.” 

A recorded meeting can be sent as an audio file and returned as a structured summary with key decisions, action items, and follow-up emails drafted. Tyson Heaton, Executive Director of LEI’s LeanTech/AI, was at Gene Kim’s IT Revolution Tech Summit this past week. He recorded speakers he knew I was interested in using Plaud, which generated a clean summary and sent me a link to a webpage where I could read it. I reviewed the summaries and filed them in my vault for future work ideas. 

Raw quality data can be analyzed for deviation or statistical insights. An SOP can be evaluated for whether it properly distinguishes major steps, key points, and reasons why — or whether it collapses them together the way most procedures do. 

None of this is a future capability. This is what current models can do when connected to the right context and the right knowledge files. We just have not yet figured out how to incorporate AI into our daily routines for coaching, instruction, reviewing, and reflecting. 

Fast-Forward to the Future 

There is a lot of AI fatigue right now, and I understand it. The hype is exhausting. The fear is real. The predictions change every quarter. I cannot even guess what is going to happen since concepts I thought were implausible just 12 months ago are now possible. The technology is moving fast enough that anything specific I predict today will very likely look naive in 12 months. 

But on a personal level, I can tell you what AI is doing for me right now. It is making me more productive every day in ways that are concrete and low stress. The Ohno historical problem-solving re-creation, the encyclopedia entries, my website upgrade, this article — none of them would exist without AI assistance. Not because the task was beyond me, but because the administrative work (think waste) was too high. AI is helping to remove that bottleneck. 

The connection between humans and AI is getting easier every quarter. The barrier to entry drops with every model generation. Early versions of these tools, like my cellphone setup, still require more technical comfort than most people have today, but that is changing fast. 

I plan to go on a lot more walks. 

Humans + AI > Problems.