AI Camp: Interfaces
A Betaworks Accelerator Program
To think that a text chat is the ultimate method for interacting with AI is akin to thinking that dial-up is the ultimate way to interact with the web, or that a punch card (or even the CLI) is the ultimate method of interacting with a computer. After all, the very first Camp program (all the way back in 2016, which included HuggingFace) was focused on natural language processing and chatbots. It’s been nearly 10 years, and in that time the interface of AI has barely changed.
We at Betaworks, for this Camp, sought out founders who were exploring the bleeding edge of interface in an AI-powered world. And boy did we find them!
timeline
August 18, 2025
November 14, 2025
November 4, 2025
PARTNERS
.png)
sponsored by
.png){kind=logo}
Cohort Demos
Feather
Feather is an OS-level client that uses the richest personal data set you have (your email) to create agentic workflows for life’s stickiest and most annoying tasks. Things like hunting for an apartment, planning a wedding, buying a house, shopping for a car, dealing with a health issue, etc. are difficult, task-oriented, and (unfortunately) highly specific and personalized challenges. With Feather, any of those ‘hair-on-fire’ challenges can be handled by AI, with minimal interaction from the user, because they have the context that flows naturally through the user’s email.
FOUNDERs
Marco Yu
connect
Intension
Intension is building the Oura for your attention on the desktop. The software tracks your attention, a combination of your focus, intention and capacity, to first help you visualize your tendencies (similar to how Oura gives you a score around various health metrics) and then obfuscates away the parts of your desktop that are interrupting you. Everything else is ‘removed’ from your sight, and interruptions are limited (or AI proactively responds on your behalf), so that you can get things done without constantly paying the taxes of context switching.
connect
Nora
Nora is a browser extension (across Chrome and Safari/Desktop and iOS) that tracks your shopping, helps you remember everything, price checks, suggests, and generally organizes what has become a stressful (nearly antagonistic) shopping experience into something enjoyable and productive. In the future, Sid imagines that your Nora will know enough about you, and be loyally aligned to your wallet and preferences, such that it can negotiate with brands on your behalf and even earn you the money that is currently going into the massive cyclone of advertising, referrals, click tracking, cookies, etc.
FOUNDERs
connect
Nubrain
Nubrain is creating the world’s first and largest proprietary data set of EEG data, using that data to train models that decode human thoughts into text, images and speech. Their first model, OB1, translates raw EEG signals into images with 20x higher accuracy than chance. This comes at a time when the hardware around EEG capture is becoming cheaper and more accessible to consumers.
FOUNDERs
connect
Orange
Orange is a technology built on three pillars: a simulation creator, a behavioral data engine, and a dynamic interface generator that makes realistic recreations of human interactions (writing emails, responding to texts, messaging apps). Together, these pillars form a platform that is called MyPlace. On MyPlace, users can play a daily micro-game that looks and feels more like a simulation from their everyday lives, the lives they want, or scenarios that reveal things about their character. MyPlace then takes that data and uses it to help you understand yourself, and that could take the shape of frameworks like Enneagram, Meyers Briggs, etc based on user preference. The insight is that play brings out the truth of who we are, which is not what we say about ourselves but how we make decisions and behave.
FOUNDERs
connect
Patina
Patina is like Sora for scent and flavor. The team have combined protein folding on the 389 known olfactory receptors and QSAR through deep learning to create the first ever ‘scent photograph.’ This foundation model, and the techniques behind it, unlock opportunities for safer and more captivating blockbuster fragrance molecules (current blockbusters generate some ~$100m/year per molecule). The Patina model can also help generate new molecules that can create ultra-realistic recreations of natural essential oils (like rose oil) which can be costly and carbon intensive. Patina is also at the forefront of the latest research around olfactory receptors in the skin, developing molecules that, in silico, have been found to imbue therapeutic properties like anti-aging and wound healing.
FOUNDERs
connect
Presq
Presq is a platform that ingests data from influencers and creators (such as athletes and musicians, who are coincidentally sick of shilling products that don’t mean anything to them) to develop a prompt interface and model ‘trained’ on the creators’ preferences. Creators can then iterate via prompt on the product they’d like to build (for now, the company focuses on footwear but will eventually expand to eyewear, jewelry, home goods, etc.) and Presq ensures that it meets the requirements of manufacturability (durability, safety, etc.) to be 3D printed in days instead of months. Presq is building a future that is mass-bespoke.
connect
Primitive
Primitive is a web app that turns big goals into small subtasks, complete with all their dependencies. Founders tend to be the bottleneck for their teams, and the same is true as we incorporate agents into our workflows. Primitive lets people brain dump their goals into the system via voice and link in with other tools like Notion and calendars. It can then generate a sprawling list of inter-connected tasks and sub-tasks, some of which can be done agentically (and those agents can even escape containment and start working in other tools). The critical technology here is the context preservation around these goals and tasks (and the data that informs them) for both humans and agents trying to be hyperproductive in an AI world.
FOUNDERs
connect
Telepath
Telepath is building the first sensemaking computer. This computer has no apps, and is built natively with AI that has a deep understanding of the user and the data on the computer, without the siloes usually formed by those application sandboxes. This starts with a series of parallel agents that look at the user’s existing data – Granola notes, Obsidian files, etc. – and analyze for things like network connections, opinions and beliefs, etc. to help thought leaders quickly ingest new information, formulate and update their beliefs, and make sense of those beliefs in a way that’s legible to others.
connect
cohort
Feather
Intension
Nora
Orange
Patina
Presq
Primitive
Telepath
nubrain
