ML Interfaces Built for Non-Experts
Oscar Beijbom, co-founder and CTO of Nyckel, on the opportunites in the AI/ML tooling market
The winning AI product for this part of the market is not the smartest model, it is the one that hides the model behind a dead simple job to be done. Nyckel is describing software where a user brings examples of inputs and outputs, then the product handles labeling, training, deployment, and monitoring, which turns machine learning from an engineering project into something closer to setting up Stripe or Twilio.
-
This shifts the abstraction layer from model settings to the data itself. A non expert user is not tuning architectures or pipelines. They are uploading text, images, or tabular rows, checking predictions, and correcting mistakes so the system learns what good output looks like.
-
That is a different product from older AI platforms like DataRobot and Scale. Those products grew up around ML teams, data prep, model choice, and labeling operations. Nyckel is betting that foundation models and smaller sample requirements let a vendor package most of that complexity into one workflow.
-
The closest analogy is the OpenAI style interface. Instead of hiring specialists to train a custom model, the customer starts with a capable base system and adds lightweight guidance through prompts, examples, and review. Ease of use becomes the wedge that expands AI adoption beyond technical teams.
This is where AI tooling is heading, toward products that look less like developer infrastructure and more like business software with an AI engine inside. As models keep improving and the amount of example data needed keeps falling, the companies that win will be the ones that make prediction, classification, and extraction feel routine for ordinary product and operations teams.