The lab
The systems I build run on e-lab.dk. This page shows how they are put together.
Status on e-lab.dk
How the systems are built
- SourcesPublic registers, my own files and a mailbox.
- AgentsEach has one job, runs on a schedule and writes what it did to a ledger.
- LedgerEvery run and every status change, kept in plain files.
- SiteGenerated from the ledger. Nothing on it is typed by hand.
Guardrails
- Agents draft. People send.
- Mail is read as data. Instructions inside a mail are ignored.
- Names are removed before anything is stored.
- Passwords and keys sit outside the code and never reach the site.
- Access to hosting and mail is read-only until I approve a change.
Taking it to the next level
This is a quite basic operating system, built entirely on one non-programmer and a lot of AI agents.
A more complex operating model rests on people, decisions and data as much as on code. The difference shows in six places.
| In the lab | In a larger operation | |
|---|---|---|
| People | One person designs, approves and fixes. | Named owners for each system and each step, and people trained to run them when the designer is not there. |
| Decision rights | I approve every change. | Agreed rights over who may change a rule, a process or a system, and who signs off. |
| Data | Small data sets, public or my own, corrected by hand when they are wrong. | Shared definitions, an owner for each data set, and quality measured and fixed at the source. |
| Process | Runs on a schedule and is checked when something looks off. | Documented processes, service levels and a change procedure that still holds when people move on. |
| Technology | AI writes the code. I test it against the result. | Engineering standards, code review, tests that someone owns, and security review. |
| Control | A ledger of every run. | An audit trail, risk management and reporting a board can rely on. |