SUPERNICE

The Lab

Some ideas become products.
Some become companies.
Some teach us what to build next.

This is the work that hasn't earned a domain yet. Some of it will. Some of it exists only to tell us which direction not to go, which is worth knowing early.

  • Multi-Agent Orchestration

    building

    Getting a group of specialized agents to complete a real job together — with the handoffs, checks and recovery that turns output into finished work.

    A single agent doing one task well is a solved problem. A group of them carrying a job from request to delivered result is not. The interesting work is in the seams: who decides what runs, how context travels between steps, what happens when one part fails, and how the result gets held to a standard before anyone sees it.

  • Local Search Intelligence

    research

    Understanding how AI systems decide which local business to recommend — and what actually moves that decision.

    When an assistant returns one business instead of ten links, the selection criteria matter enormously and are largely undocumented. We test against real businesses in real markets to find out what the models are actually reading, which signals carry weight, and how quickly a change propagates into an answer.

  • Autonomous Marketing

    building

    Marketing work that continues without someone driving it — research, production, publishing and reporting running on their own schedule.

    Most marketing automation still needs a person to start it. We’re testing what happens when the system decides what needs doing, does it, checks the result and reports back — with a human reviewing decisions rather than performing tasks. The hard part is judgment, not execution.

  • Voice & Conversational AI

    experiment

    Voice interfaces for businesses whose customers would rather talk than type or fill in a form.

    Plenty of businesses lose customers at a contact form. Voice changes the shape of that first interaction — answering a question, qualifying an enquiry, booking something — without a person sitting on the phone. We’re testing where it genuinely helps and where it quietly annoys people.

  • Automated Website Generation

    research

    Producing a complete, fast, well-structured website from a business's own knowledge rather than from a template.

    Templates produce sites that look the same and say nothing specific. We’re testing generation that starts from what a business actually knows about itself, and holds the output to real standards: performance, accessibility, structure and content that a search or AI system can understand.