A challenge at 65
I started as a salesman in 1986. Then 18 years at LG, two companies founded, close to 30 patents. Today I am trying something new - learning as I go, studying as I go, to see if I can run alone, with AI, the work that once took about twenty people.
40 years, only the tools changed
In 1989 to 1990 I bought a dBASE III book and taught myself, using the IBM XT class PC that came into our office. I also built management records with the Lotus 1-2-3 spreadsheet.
Back then I went to the telex room every morning to check messages from overseas buyers. Telex was the text message of its day, and it charged by the character, so everyone used short forms. Today's messenger abbreviations come from the same habit.
Fax spread in 1992 to 1993 and let us send documents on the spot. Email spread in 1996 to 1998 and changed how we worked again.
Telex, then fax, then email, and now AI agents. The tools kept changing, but what I do has stayed the same: decide what to do, give the instruction, and check the result.
Career
I spent 18 years in global sales and planning at LG.
Then 17 years in overseas joint ventures and global C-level roles. I lived overseas for more than 10 years, working across Italy, Spain, Norway and China.
Now I run AKSys and SHAKS, founded and operated alone. Close to 30 patents, about 20 government R&D projects and a few EU projects led. I bring that experience into an AI agent organization.
What I do now
Now I give instructions and check what comes back. Research, documents, code and publishing are handled by a number of agents.
I built a structure that splits instructions across six departments and puts the result through an audit team. I moved it over the same way a human organization worked.
I put work like government R&D, hardware productization and international partnerships into the same structure. Not all of it worked. What did not work, I wrote down as not working. Where judgment could only be done by a person, I left it with a person.
Handing work to a machine and changing how I give instructions turned out to be two different things. It worked better when I gave the machine the same instructions I gave people for 40 years.
What slowed things down first was not speed. It was the pass criteria. Without writing down what counts as done, the work kept coming back. This worked for me, under my conditions. It may be different at another scale, or in another setting.