Work with AI,
and build the skills AI cannot do for you.
Working with AI starts at ArchCross.
Over ten years in HR,
now turning real work into micro-systems through AI collaboration.
A lot of work never made it onto IT's schedule and didn't fit any off-the-shelf software, so it ended up done by hand in Excel. After more than ten years on the HR front line and more than ten micro-systems built with AI, I hand that kind of work to AI and automation and keep the judgment calls for people. So I can help you break a real piece of work down clearly: what to hand to AI, what to keep, and how to check the work once it's handed off.
More than ten years in HR, from recruiting and training to system rollouts, and on to digital transformation, process automation, and AI applications. Also helped companies win several national awards in talent development and branding.
Work everyone knows matters, but nobody had solved
Getting this work wrong carries compliance risk, and everyone knows it has to be done right. But it never made IT's schedule, no off-the-shelf software fit, and people kept it going by hand. I redesigned the whole process, decided which steps go to code and AI and which must stay with a person, and then built it together with AI. Here are a few examples.
-
Mandatory training tracking
A few hundred people, each with a different deadline, is too much to watch in Excel
It used to mean rerunning Excel formulas maybe every two weeks and checking the list line by line for people who had left. I took the process apart and rebuilt it: the platform matches records, counts hours, and sends reminders every day, and people only handle the exceptions.
-
New-hire training checks
Every new hire has a different deadline, plus special cases like people who leave and come back
Each deadline is worked out from the start date, and the people running late or already overdue are listed directly; for people who return, the gap since they left decides whether they need to retake the courses.
-
Certificate review
Certificate photos arrive one by one, and every field has to be checked
The process now has AI read the fields first and rules check them once; the reviewer looks at the results and presses approve personally, and every decision is logged.
-
Certificate audit for departed staff
People have left, but their certificate registration is still active, which is a compliance risk
This used to be a manual comparison of two lists, slow and easy to miss things. Now the lists are compared automatically every week, producing the names that need to be deregistered.
-
Training-hour reporting
The hour rules are complicated, and the same attendance records have to be reported internally and externally
The splitting rules and reporting formats are handled in code; attendance is turned into an internal record and an external report automatically, and people only check the exceptions the system flags.
-
Org charts and succession planning
HR data contains dirty records, so org charts and planning sheets go wrong easily
Errors like missing managers and circular reporting lines are listed before the chart is drawn, and exports match what is on screen; succession planning is kept separately for each year, with a warning when one person is slotted into several positions.
Once it became a system
Less time goes into repetitive work, and progress and exceptions are visible at any time. When the person in charge changes, the rules stay in the system, so whoever takes over does not start from scratch. Errors no longer depend on someone checking line by line, so the risk comes down too.
There are also a performance management platform, batch talent card downloads, a document merge tool, and more.
The machine is the Maker,
the person is the Checker
Every micro-system follows this principle: let machines handle calculation and output, while people hold the final review.
- AIDrafting options, structural reasoning, research, and writing code
- AutomationFixed rules and right-or-wrong checks
- PersonExceptions, and the final check before anything is written into a production system
Four things I hold to when building these systems
-
Redesign the process first, then choose the tools
Swap the tool but leave the process alone, and work still gets stuck in the same places. For every micro-system I built, I took the process apart and rebuilt it first, then decided which steps go to code and AI.
-
Skills are built in real work
A method learned in class sticks only after you use it once on real work. That is why Studio has you try it yourself in a new scenario before you see feedback, and Kaico builds conversations from what you actually need to handle next week.
-
Put errors in the open, never skip them quietly
When the org chart tool hits dirty data, it lists every case in a diagnostics report instead of quietly fixing it. In Compass reports, an observation a second AI cannot find support for in the conversation is left out, and directions with too little evidence are labeled as such.
Read “Handing Routine Tasks to AI: I Run a Comparison First, Then Catch Exceptions” → -
Data is used only where it should be
In my micro-systems, the reports used for comparison and the temporary files used for packaging are cleared after use. Kaico's server receives only the recognized text, never the recording. Every observation in your development profile can be confirmed, disputed, or hidden.
The more AI does, the more people need these four skills
My work in HR focuses on learning and development (L&D). In an AI-enabled work environment, everyone needs four solid skills: collaborating with AI, alongside judgment, self-knowledge, and language expression practiced by people. These four products are previews built together with AI, putting this perspective into practice. AI writes the code; what each product asks, how you practice, and how feedback is given come from the judgment I have built up working in L&D.
- Coach Collaborate AI advisors 14 advisors from different fields. Each one asks about your background and constraints first, then lays out options with their reasons, and you make the call.
- Studio Judge Scenario learning 13 built-in courses with 7 to 10 minute levels. Each scenario is generated from your earlier answers, and you get feedback as soon as you finish.
- Compass Self-knowledge Fit assessment Three topics: career exploration, new-manager readiness, and AI collaboration. Every observation in the report comes with the words you actually said.
- Kaico Language Second-language speaking Tell Kaico one thing you need to get done in another language. It turns that into a story and has you talk with the people in it. Nine languages, in any pairing of the one you know and the one you are learning.
Stuck on a step? Let's look at it together
If there's a piece of work you want to hand to AI but aren't sure how to break down, write in.
Get in touch