The network for creativity
Join 1.25M professional creatives like you
Connect with clients, get discovered, and run your business 100% commission-free
Creatives on Contra have earned over $150M and we are just getting started
Back to feedPost
One thing I’ve learned while building AI automations: not every decision needs AI.
For example, in a lead qualification workflow, I don’t want an LLM deciding whether a lead qualifies on its own.
Instead, I use AI to extract useful evidence from messy data, things like company size, services offered, or information from a website.
Then deterministic rules handle the final decision.
This makes the workflow easier to test, debug, and trust.
AI handles the ambiguity. Rules handle the decision.
I’m curious: where do you draw the line between AI and deterministic logic in your automations?
The network for creativity
Join 1.25M professional creatives like you
Connect with clients, get discovered, and run your business 100% commission-free
Creatives on Contra have earned over $150M and we are just getting started
Related posts
Will you learn basics or use AI as the basics#BuildInPublic #LearnToCode ##AIENGINEER #founders
AI Taste App: Learning from Personal Preferences
Personal taste is hard to describe as a set of rules. I built AI Taste App to explore a personal dating question: could a small application learn which faces I find attractive from simple yes/no feedback?
I split the experience into Teach and Predict. In Teach, I upload a portrait and choose “Yay” or “Nay,” saving another training example. In Predict, I submit a new portrait and get a score based on the earlier feedback, without changing the training examples.
Teach mode saves positive and negative feedback for the personal classifier.
Predict mode scores a new portrait against earlier feedback. This is an illustrative result, not an accuracy measurement.
Laravel handles the web application, while a separate FastAPI service handles face detection, image processing, and machine learning. A pretrained vision model extracts visual features, and a small personal classifier learns from the saved ratings. The practical work was connecting those steps to a straightforward mobile interface.
This is a private personal prototype: its predictions reflect the feedback it learns from.
Read about the implementation and see illustrative screenshots on NikoCodes.
I run my company on an AI chief of staff I built. It is called Coulson.
Every morning it puts the day on one page before I open email: the three revenue priorities, every commitment with its date, and one list of what is waiting on my approval. Behind it, a team of AI agents drafts replies, prepares meetings, chases follow ups and sends an end of day report at 7 PM. Nothing goes out without a yes.
Now I build the same system for other companies. The case study, with real screens on sample data: hqlabz.com/work/coulson
