Smart Flows
Small prototypes to explore automation without losing human oversight.
This is my workbench. Here I sketch the flows I want to build and test with AI, automation and agents: which tasks can run on their own, where a review makes sense and, above all, at what point a person should decide.
They are still sketches. I will turn them into small prototypes, one by one, and share here what works… and what doesn’t.
Automating is not letting go of the wheel.
01 Human in the loopWhere to automate, and when should a person take a look? Sketch
- Welcome
- AI processes
- Human validation
- Continue · Correct · Stop
- What I want to test
- Which tasks can run on their own, and at what point a human review prevents costly mistakes later on.
02 AI Quality CheckHow do you avoid accepting an answer just because it sounds confident? Sketch
- AI generates an answer
- Second review layer: sources, contradictions, errors and biases
- Confidence level
- A person decides whether it holds up
- What I want to test
- That a well-written answer is not the same as a reliable one, and which checks help tell them apart.
03 AI vs AIWhat happens when one AI contradicts another? Sketch
- One agent proposes an interpretation
- Another agent challenges it: looks for flaws and alternative explanations
- Weighing the arguments
- The final decision is human
- What I want to test
- Whether disagreement between agents helps reveal blind spots… or just adds noise.
04 Research FlowWhich part of research can be automated without automating the interpretation? Sketch
- Customer comments
- Automatic classification
- Thematic grouping
- Pattern detection
- Hypotheses
- Human review
- What I want to test
- How far the machine can go in sorting things out, and where the work of understanding begins. A pattern is a clue, not necessarily the explanation.
05 Prompt ControlHow much does an answer change when the way you ask changes? Sketch
- A bare question
- + context
- + constraints and instructions
- + quality criteria and examples
- + controls
- I compare the answers
- What I want to test
- That working well with AI is not about asking a question, but about designing the conversation.
The moment a person decides. Any examples I use will always be fictitious: never real customer or company data.
What I learn here, I will think through out loud in AI & Me →