Skip to content

The CCS Interview

Sina Yamani: AI that clicks like a person can reach the 99% of the web APIs miss

Sina Yamani is founder and CEO of Action Model, a community-owned project building AI that carries out tasks on a computer rather than just answering questions. In this conversation, recorded in February 2026, he explains large action models, why he thinks AI employees threaten white collar jobs, and how fractional ownership is meant to give users a stake. At the time, he said the platform had launched two days earlier.

Interview by Ashton Addison Recorded February 2026 5 min read

Key takeaways

  • Large Action Models differ from LLMs by using mouse and keyboard to navigate the internet like humans, accessing 99% of the web unavailable through APIs.
  • Action Model is a community-owned AI initiative where users contribute training data through a browser extension in exchange for fractional ownership.
  • Centralized AI automation threatens approximately one billion white-collar workers globally, with value capture concentrated among a handful of trillion-dollar tech companies.
  • Action Model's decentralized approach aims to distribute AI ownership across millions of people as an alternative to centralized corporate control of automation technology.
  • The platform is currently in beta with multiple use cases across marketing, sales, operations, and HR, available as a browser extension and desktop application.

From chatbots to AI employees

Ashton Addison: What is the difference between an LLM and a large action model?▶ 00:49

Sina Yamani: LLMs are the chat models. You speak to them and they give back an answer. We've been in this advice era for the last few years: you tell it what you want, it writes the structure out, and then you have to go and do the work. The next stage is that the AI actually does things for you. It won't just draft a contract or an email. It will read through your emails, make decisions because it understands your business, respond, follow up and do whatever work is required. That's very disruptive, because it pushes past 'a cool tool that speeds up a person's work' to 'does the employer need the person at all?'

Ashton: Why is it a problem if the big AI companies control this space?▶ 03:23

Sina: About a billion people around the world are employed to work behind a computer screen in some kind of white collar job. A lot of it is repetitive work like data entry that can easily be automated, and AI can do much of this today. Over the next few years, AI will learn to understand the business and its needs, and then do all of that work. Employers all want the same thing: to make more money and be more efficient. When the tool can do 95% of the work for 10% of the price, it's a no-brainer for these businesses. Unless governments act now with deep, strong regulation, this is inevitable. And almost all of that value will go to a handful of trillion-dollar tech companies. Anthropic's CEO said a few weeks ago that half of white collar jobs will be gone within five years, and I think that's an understatement. So this either needs to be regulated or it needs to be partly owned by the people. That is what Action Model is focused on.

When the tool can do 95% of the work for 10% of the price, it's a no-brainer for these businesses.

Using the web the way a person does

Ashton: LLMs are adding actions too. What does Action Model do differently?▶ 06:02

Sina: LLMs output text, and they connect to applications through APIs. Only around 1% of the internet has APIs available, which means 99% of it is only accessible to people using a mouse and keyboard. The next generation of AI actually uses the mouse and keyboard, so it can use a computer just like a person does. It doesn't need to integrate with anything or get access, because platforms and websites have no idea it isn't a person.

Ashton: The big labs release stronger models every month. How can a community-built effort keep up?▶ 24:01

Sina: I think the gap between the top LLMs will narrow. The main difference may end up being speed to first token and cost, because companies like Google own their own infrastructure and chips. But for what we're building, there's a significant difference. A traditional LLM struggles to navigate a browser. Its training data is about a year old, so if the UI has changed, it doesn't know the right sequence of steps. It also isn't trained on website interfaces, so it doesn't know where the buttons are or what they look like, and it hallucinates much more on actions than on text. Text hallucination is already around 20 to 30%, depending on the topic. A business needs close to a 100% success rate. We get live data. Tens of thousands of users downloaded the browser extension within a couple of days, and it learns from the platforms our community uses: the functionality, the sequence of steps, what the buttons look like, the mouse coordinates. That's live and accurate, not a year old, and it drastically increases accuracy.

Community ownership and early traction

Ashton: How does the community-owned part of Action Model work?▶ 07:49

Sina: This disruption is happening, and individually there's nothing we can do about it, but collectively, in the millions, we can definitely do something about it. We can own a stake together. It's fractional ownership distributed across millions of people, and it's all about contributing to the ecosystem. The easiest way is to download the browser extension, turn it on, and you're helping to train the action model. That training improves an AI that knows how to use every website.

Ashton: Will early adopters mainly be crypto natives?▶ 10:44

Sina: I don't think so. I think 95% of our community will be normal Web2 people who may have no involvement with crypto. In my last company, a UK fintech, we brought QR code payments to the UK and connected to banks directly, without Visa and Mastercard. That grew to millions of users within a year through word of mouth, because people like to support the resistance. Here, a billion people are at risk of losing their jobs to AI, and you also own a fractional share, so there's an incentive to participate. We launched two days ago and hit 60,000 users in the first 48 hours. The platform is invite only, and we've done no real marketing in Web2.

collectively, in the millions, we can definitely do something about it.

Making AI employees usable for ordinary businesses

Ashton: AI agents still require code. Does Action Model remove that barrier for non-technical workers?▶ 14:37

Sina: Yes. There will always be a thriving open-source community, but look at the user numbers. I'd say 99% of AI users pay for it through big tech, and only skilled technical people bother testing open-source models because they're complicated. The biggest challenge in this next stage is onboarding. You'd spend one to two months teaching a new employee their platforms, policies and processes, and an AI employee needs a similar investment. So we built an optional tool called the Onboarder. You turn it on and it shadows you and learns from what you do. You can leave it on for five minutes to show it one specific task, or for a month. By the end, you have an AI employee with the context: where your files are, what platforms you use, which workflows and schedules.

Ashton: How do workflows fit into the community-owned model?▶ 21:30

Sina: We have a marketplace within the ecosystem, and it's one of the ways to earn a larger fractional stake. You create workflows for yourself, and if you want, you click 'submit to marketplace'. Once it passes review, it's published, and anyone can attach it to their own AI employee in one click. The marketplace is split by platform, such as Gmail, Figma or Salesforce, and you don't need integrations or APIs. Tools like Zapier and n8n are still technical, with a learning curve of at least a month. In ours, if you can read and write, you can create workflows, because you're mostly telling the computer what to do in natural language: step one, open Chrome; step two, click the URL bar.

This interview has been edited for length and clarity. Recorded February 2026.

Sina Yamani
Sina YamaniSina Yamani is the Founder and CEO of Action Model, a company building a community-owned Large Action Model, an AI system trained to carry out digital tasks by learning from real user interactions. He previously founded the Welsh fintech Yoello, which provided QR order-and-pay and… Full profile →
Share this interviewXLinkedIn
▶ Watching the interview