An open workshop for practical AI experiments.
This is where real work done by people and AI together is written down: the method, the result, the time spent, the limitations and the learning, so others can understand and repeat the experiment. Including the times it fails.
What an experiment contains
Every experiment follows the same card, so two experiments can be compared without first decoding the author's taste.
- Task
- The real work being attempted
- Human role
- Decisions, judgement, validation and accountability
- AI role
- Analysis, generation, automation or execution
- Tools
- Models, applications, code and configuration
- Method
- Reproducible steps
- Result
- Observable output, including failures
- Time
- Human and machine time where measurable
- Limitations
- Uncertainty, risks and what was not tested
- Learning
- What should be repeated, changed or stopped
How to read the evidence labels
Any statement that could be mistaken for a measurement carries a label. It says how firmly the statement stands, not how good the result is.
- Measured
- produced by a defined measurement
- Observed
- seen directly during the experiment
- Reported
- provided by another source or participant
- Hypothesis
- plausible but not demonstrated
- Unknown
- not established
A useful result
An experiment counts as useful in all of these cases, not only the first:
- AI improved the work
- AI helped only with a bounded part
- A standard tool was better than a custom solution
- The experiment failed
- AI should not be used for this task
Latest experiments
All experiments- EXP-001OngoingTested: 2026-09-22
Building and governing a specialised AI-agent team
A team of named AI agents has been merging its own pull requests into two private production repositories since August 2026. This experiment measures what that team actually did, documents the one case where an agent merge went past a blocking review, and states plainly that no productivity claim can be made because no baseline exists.
What aiMinds is not
Three things this site deliberately stays away from:
- An AI news site
- A tool directory
- A sales site
Who is behind this
aiMinds is published by Spekir, run by Rasmus Sloth Nielsen. Method, templates and material are public on GitHub, so an experiment can be checked without asking permission.