TRAINING
HerzLeben Agentic Coding Training
AI writes; people check. Corporate training in AI-driven development — agentic coding with tools such as Claude Code — learned by building something that runs with medical data.
- Hands-on with medical data
- Nothing to install
- No programming experience needed
From vibe coding to AI-driven development you can explain
Code nobody can explain goes into your work
You can explain what the AI wrote
The goal: explain in your own words what the AI wrote — and check it.
Why we teach with medical data
Medical data calls for judgment that generic sample data never raises — and only people who know what is inside the data can teach it.
No clear denominator
How to set the patient population
Codes change
Absorbing yearly revisions
Small cells stay hidden
How to suppress small numbers
Can it go in the report?
The call before a number is published
Start with a free 2-hour session
We run a 2-hour version of HerzLeben Agentic Coding Training at no charge. Participants try the AI-writes, people-check way of working hands-on, with medical data.
- 2 hours (can run about 30 minutes over for Q&A)
- Up to 3 people, any role
- Online or at your office
- Nothing to install
SeeCo-authorship network
Map researchers’ connections and central figures from PubMed papers
Ask for numbersAsk your data in plain language
The AI aggregates the data behind your question and answers with tables, charts and maps
Ask as a teamKnowledge search across past projects
Answers from internal documents with sources. Questions can also come from Slack
Choose one of the three topics. Screens show apps we built or mockups (public or synthetic data).
- 0:00–0:30Agentic coding basicsThe plan → execute → verify loop, and how to write instructions
- 0:30–1:30BuildAdd features to a prepared base with your own instructions
- 1:30–1:50VerifyCheck the results against the source data and get it to a usable state
- 1:50–2:00Your own workWhere this fits in your job
Two days building a workflow app
with AI and medical data
Day 1 builds the way of working; on Day 2 each team shapes one build drawn from its own work.
HerzLeben Agentic Coding Training — two-day program
Learn how to split the work with AI
- Have the AI plan, implement and verify
- Run the full loop once on a small task
Read and explain the code AI wrote
- Read the diff and explain it in your own words
- Write your business context into instructions
Designing for safe delegation
- Draw the data line and set permissions
- Organize requirements; have the AI plan
Shape one build as a team, and verify it
- Get it running and check the results are right
- Present it and plan how to keep going
Examples: a co-author network from published papers / an app that answers data questions asked in plain language / a search over past project documents (with Slack). The data is public medical data, or your own data under an NDA.
It runs in the browser: nothing to install on participants’ PCs, nothing to request from IT.
- All code written during the program
- Claude skill definitions, ready to use in-house
- Follow-up support after the program
Delegating safely in healthcare and pharma
The more you hand to AI, the more people must decide up front. The training builds on these four ground rules.
Separate data the AI may see from data it may not
Handling personal and unpublished information, and substituting public or dummy data
Narrow what the AI is allowed to do
Settings that keep what it can read, write and run to a minimum
Nothing goes through until a person has read it
How to read a diff, and how to tell changes to accept from changes to stop
Keep a record of what the AI was asked to do
Instructions and change history kept so the work can be verified and explained later
Track record
We have taught in programs delivered with training companies and in a university course for medical data scientists. Client names are withheld; only the theme and audience are listed.
FAQ
Is this a course on how to use Claude Code?
No. It teaches a way of working where AI writes and people check (AI-driven development, or agentic coding). Participants learn Claude Code along the way.
How is the free session different from the two-day program?
The free session is two hours: you experience the approach by adding features to a prepared base. In the two-day program you first learn the approach, then build one thing as a team around your own department’s work. You can also start with the free session to see whether it suits you.
Can we take it with Codex or other tools?
Exercises mainly use Claude Code, but the approach is the same whichever tool you use. Ask us about running it with the tools you already have in-house.
Do participants need programming experience?
No. The goal is not to write code but to read, explain and check what the AI wrote.
Can we use our own data?
Yes, under an NDA. If you have none to bring, we run the same format on public medical data.
We are concerned about information security.
The four SAFETY points above are built into the program. We provide the development environment, so nothing is installed on participants’ PCs. We can also adapt the approach to your internal policies.
Can the program be used with Japan’s human resource development subsidy?
Yes, we support programs planned around the subsidy. Hours, curriculum and schedule are set to meet the requirements of the training plan filing.
CONTACT
Tell us who the participants are and what you want them to be able to do afterwards, and we will propose a program.