Heidelberg AI Curriculum

18books at this level

Run One Assistant for a Whole Institute or Company, and Hand It Over

When 150 people use one assistant, its doors, budgets and records are yours to run. Put everyone behind the organisation's login, end a leaver's access with one tested action, give each group a budget, keep lean records, and hand it all to a substitute.

Outcome: An access audit for the shared assistant: every account, role, key and emergency access with its owner and revocation path, a tested one-action leaver removal, group budgets with last month's cost per group, the records and retention table signed by the data protection officer, and the substitute's run. It passes when a test leaver loses a live session and a fresh login within the stated window after one action, no local password login works, and the substitute repeats the audit from the runbook alone.

Operator20 minLast edited: Sep 23, 2026

T03-L0502

Keep a Production Prompt Healthy for Months, and Hand It Over

A prompt many people depend on drifts: new kinds of input, a model swap, a failure nobody saw. Grow a regression set from real failures, gate every change on a rule written in advance, watch three signals, handle incidents, and hand it all to a substitute who has run it.

Outcome: An operations folder for one production prompt: the regression set with its critical group and pass rule, the gate results for one change, a monitoring page with three signals and named responders, and one incident or drill note. It passes when the substitute reproduces the gate decision and a rollback from the runbook alone, and a colleague answers three auditor questions from the folder.

Operator21 minLast edited: Sep 23, 2026

T04-L0503

Keep a Connected Knowledge Base Current, and Prove a Withdrawn Document Is Gone

Deleting the file doesn't delete the answer. Keep a document assistant that many people rely on in step with its sources: set a withdrawal window, make the sync remove what the source removed, watch four signals, handle a withdrawal incident, and hand it all to a substitute.

Outcome: A withdrawal drill record for one connected knowledge base: the stated windows, a current and a withdrawn marker checked before and after, the time from withdrawal to disappearance in every place checked, who received the withdrawn document and how they were told, and the substitute's run. It passes when the withdrawn marker is gone within the window, the current one still works, and the substitute ran it from the runbook alone.

Operator18 minLast edited: Sep 23, 2026

T05-L0504

Keep Your Automations Running for Months, and After You Leave

Run a set of n8n automations that many people depend on: one register, four signals to watch (silence, failures, waiting work, cost), a routine for when one breaks, a way to take updates and model retirements calmly, a folder an auditor can read, and deputies who have practised taking over.

Outcome: An operations folder for every automation you run (register, alert setup, incident routine, change log and handover checklist), which passes when every automation has a keeper and deputy other than you alone and a heartbeat that alerts a team channel, a deputy handles a staged incident from the folder alone, and a colleague answers three auditor questions from it in 15 minutes.

Operator21 minLast edited: Sep 23, 2026

T06-L0505

Run the App Others Depend On, and Hand It Over

Operate a team app that many people rely on, for years: every change through a staging copy, a rollback you have practised, outside checks on sign-in, backups and certificates, a yearly plan for updates, licences and costs, a quarterly access review, and a deputy who can release, roll back and restore without you.

Outcome: A handover folder for one app (release steps, rollback record, monitor list, update and cost plan, access review, restore record), which passes when a deputy, with the builder's accounts disabled for the day, releases a change through staging, rolls it back, and restores last night's backup to a test server from the folder alone.

Operator18 minLast edited: Sep 23, 2026

T07-L0506

Keep a Team of Agents Running for Months, With Memory a Person Approves

Run several agents that colleagues depend on, over months: keep the number of agents small, let a person approve every note they remember, log every run, review four numbers weekly, handle incidents with a written runbook, keep an audit folder, and hand it all to a deputy who has run it.

Outcome: An operations folder for agents colleagues depend on: four weeks of run log and weekly reviews, the memory file with a source and review date for every note, a runbook with one drill, an audit index, and a deputy's handover record. It passes when the deputy runs a review and a drill alone and an auditor's question is answered from the folder.

Operator19 minLast edited: Sep 23, 2026

T08-L0507

Run Automatic Checks, an AI Reviewer and Releases You Can Roll Back

Keep code that many colleagues depend on safe to change for months: checks that run on every pull request, an AI reviewer that can only comment, a rule for every new package, releases the server runs by version, a tested rollback, a weekly look at the numbers, and a deputy who can run it all.

Outcome: A repository with required checks, a comment-only AI reviewer, dependency notes, a release record and a runbook, plus a handover record. It passes when a pull request that breaks a known answer is blocked, a rollback drill restores the previous version in under 15 minutes, and the deputy runs both without help.

Operator34 minLast edited: Oct 18, 2026

T09-L0508

Keep Every Published AI Image, Voice and Video Labelled, Cleared and Traceable

Run AI media for your whole institute or company over months: one register of every asset with its tool, rights, consent, label and approver, label rules that meet the EU AI Act, a monthly check, a tested way to pull an asset back, and a deputy who can answer an auditor without you.

Outcome: A media register and a one-page runbook in a place the organisation owns; it passes when every AI asset published in the last quarter has a complete register row, one monthly check has been run and logged, a takedown drill has been done on one asset, and the deputy answers an auditor's question about a randomly chosen asset from the register alone within five minutes.

Operator18 minLast edited: Sep 23, 2026

T10-L0509

Keep a Data Pipeline Right for Years, and Trace Any Published Number to Its Source

Run a data pipeline that many people rely on: keep every run's input, code and output under one run ID, watch that it ran and that its data still look like last month's, approve before anything is published, and trace any number in a board pack or paper back to its source rows, even after you've left.

Outcome: A written trace of one number that was published from the pipeline, from where it appears back to the source rows, which passes when every step names its run ID, fingerprint and person, the missed-run alert and the drift check have each fired once in a test, and a colleague repeats the trace from the runbook alone and reaches the same rows.

Operator18 minLast edited: Sep 23, 2026

T11-L0510

Keep a Shared Model Server Fast, Affordable and Running

Run a self-hosted AI server that many people rely on, month after month: watch five signals, measure how many people it serves at once, set a limit below the breaking point, compare its cost per 1,000 answers with an API, and hand it to a second operator with records an auditor can check.

Outcome: An operations pack for a shared model server: a capacity table from a load test with the limit you set, a cost sheet comparing your server and an API per 1,000 answers, a one-page runbook and an operations log. It passes when a second operator has handled a practice outage from the runbook alone and the service owner has signed the cost decision.

Operator21 minLast edited: Sep 23, 2026

T12-L0511

Keep One List of Every AI System Your Team Runs, and Make It Survive You

Find every AI system your team uses, including the unofficial ones, and keep one inventory with a risk tier, a responsible person and stand-in, the data touched and who is trained. Link the proof an auditor asks for, decide which uses shouldn't exist, and test the handover without you.

Outcome: An AI inventory for one team: every system with purpose, risk tier, responsible person and stand-in, data touched, training record, linked proof and next review date, plus a record of one 'author has left' test. It passes when a stand-in who didn't build any of the systems answers an auditor's five questions and handles a practice incident from the inventory alone.

Operator19 minLast edited: Sep 23, 2026

T13-L0512

Keep Everyone's AI Skills Current, and Show It When Asked

Run AI literacy for a whole institute or company: a map of what each role must be able to do, short checks on real tasks instead of attendance lists, one register that follows people as they join, change role and leave, refreshers triggered by changes, and a record you can show an auditor or the works council.

Outcome: A role map and a training register for one real department, which pass when every role's abilities are written as checkable tasks, every person with access to an AI tool has a current check result or a dated reason why not, one monthly reconciliation with the access lists has been run and logged, one change has gone through the refresh decision, and a deputy answers an auditor's three questions from the register alone.

Operator19 minLast edited: Sep 23, 2026

T14-L0513

Run Your Organisation's Connectors: One List, Per-Person Access, a Log and a Clean Exit

Operate the connectors a department or company relies on: one list of every route from an assistant into a system, per-person access, a log line for every call, a check of third-party connectors before install, a daily test, and a retirement you can prove.

Outcome: A connector list for one department or company team, plus one retirement record: it passes when every connector has what it reaches, whose access, version, log, responsible person, stand-in and stop; two test accounts show different access where they should; and a retired connector can't be called while its former users know the fallback.

Operator19 minLast edited: Sep 23, 2026

T15-L0514

Keep Your Group's AI Use Compliant and Running After You Leave

Turn a group's AI setup into something that survives an audit, a reviewer and your departure: a register of every AI use with the data-protection and AI Act questions answered by the right office, an archive of prompts, models and versions next to the data, a retention plan, and a successor who has run it.

Outcome: A compliance and continuity dossier: a register of the group's AI uses with who decided each open question, a retention plan per item, and a record of a named successor's walkthrough; it passes when the successor has run the main AI-assisted step and restored one archived result without help, and the PI has signed.

Operator21 minLast edited: Sep 23, 2026

T16-L0515

Keep Your Toolset Useful

Run your organisation's AI tools as one set: a register with admins, contracts, data terms and renewal dates, a quarterly review of use, cost and terms from the admin consoles, a routine for price, terms and model changes, deliberate consolidation and retirement, and a handover so the set survives you.

Outcome: A toolset folder (register, last quarterly review, change routine, one completed retirement record, handover checklist), which passes when every tool has two named admins and organisation billing, every contract with personal data has a data processing agreement on file, one tool has been retired with a deletion confirmation, and a colleague answers three auditor questions from the folder in 15 minutes.

Operator17 minLast edited: Sep 23, 2026

T17-L0516

Decide Which Problems Deserve Building

Put every AI project your lab or company runs or wants on one sheet, as problems. Score each by value, effort, risk and who keeps it running after you leave, retire what no longer earns its upkeep, and keep a decision log a successor can follow without you.

Outcome: A ranked portfolio of every running and proposed item with four scores and a decision each, plus a decision log, which passes when every kept or built item has a named keeper and backup other than you, and a colleague who wasn't involved can answer three questions about past decisions from the log alone.

Operator17 minLast edited: Sep 23, 2026

T19-L0517

Run Your Organisation's Gitea: Teams, Safe Sign-In and a Restore You Have Tested

Move the lab's repositories into an organisation with teams, add two-factor and institute sign-in, mirror to GitHub, back up the whole Gitea, prove the restore on a separate test server, and hand over with a one-page runbook.

Outcome: A colleague restores last night's backup on a test server using only the runbook and finds the organisation with its repositories, issues and pull requests; every password account has two-factor, a student account can read but not edit, and the mirror updated within the last day.

Operator23 minLast edited: Oct 18, 2026

T20-L0518

Run Coding Agents for a Team: Shared Rules, a Test Set, Nightly Jobs and No Lock-In

Turn single-person coding-agent set-ups into a team routine: one AGENTS.md every harness reads, a test set that defines done, a nightly CI job that fails loudly, and harness, model and gateway as separate settings.

Outcome: House rules with their checks, the four-row run table, links to one passing and one deliberately failed workflow run, and the operator log. Passes when every rule has a check, the job fails on a broken rule even if the agent reports success, harness/model/gateway are single settings, and the log names the next test-set date.

Operator17 minLast edited: Oct 18, 2026