T
to
actions
Mission
Building the AI last mile
With three startups under our belt each and years of multitasking, finding a reliable way to delegate tasks we'd rather not handle has always been a challenge.
So, while GenAI apps primarily focus today on text-to-content use cases like content generation and knowledge search, we have decided to focus towards building the text-to-action capability of GenAI.
Our starting point is to leverage the latest advances in computer vision to develop a text-to-action system that can shadow us on our computers and learn how to use any type of apps, just like humans do.
Our goal is to build a reliable app that you can seamlessly delegate your time-consuming and repetitive tasks to, similar to delegating them to a smart intern.
Plan
Towards a reliable and hands on AI work companion
Over the last months, we delved into the daily tasks of 200+ knowledge workers, crafting a versatile system that autonomously handles common use cases reliably. It surpasses anything we've tested from open source repositories and research papers — reaching a level of reliability that can be trusted.
Yet, there's still much to be done.
twin-master-plan.md
Careers
Welcoming founding team members
We're seeking individuals enthusiastic about making AI truly practical in the real world to join our founding team. Together, let's transform our experimental prototype into a robust product for our initial customers. We have two exciting sets of challenges to tackle.
On the technology front — crafting reliable autonomous agents capable of executing tasks in a browser, learning from users and mistakes. Our aim is to make our system increasingly faster, cheaper, more capable and reliable. We're consistently exploring new techniques to boost model accuracy for relevant tasks, like incorporating self-learned paths or fine-tuning specialised language models.
On the product front — forging an interface distinct from what we’ve seen in the first wave of GenAI tools:
Proactive
Our goal is to minimize the time to completion as much as possible. Simplicity in setting up a task is crucial. We want Twin to take the lead in the conversation and notify users when necessary. Not the other way around.
Action Driven
We do not envision a world where we spend our entire day in a chat box. Natural language should be used sparingly, and Twin should focus on taking action rather than engaging in lengthy discussions.
Trustworthy
If Twin is uncertain or unfamiliar, it admits it, and should ask the user for validation. When Twin works with users’ apps, users should be notified. Transparency is key in building trust.
Learning
Interacting with Twin should feel personal, not generic. As users share information, Twin should remember, and asking the same question twice.
Design Partners
Launching our B2B partner program
One of the major challenges of GenAI today is its lack of steerability. While viewed by some as a feature, we believe that for widespread adoption in organizations, this needs to be addressed. That's why, early in our journey, we're onboarding a select group of partners to co-design custom task libraries. These are trained to align with their internal processes, work methods, and leverage their specific tools - making sure that when we ship the first version of Twin, it seamlessly integrates into their operations.
We've started training Twin on tasks from our community. If integrating GenAI into your organization is a 2024 priority and you're keen on more than content generation, contact us. We're soon onboarding the first batch of partner companies.
Investors
We have raised a $3m pre-seed round and are proud to be backed by some of the best AI investors, founders and operators.
Get updates
Secure your spot for occasional development updates and early access. No spam, just the good stuff.
