Mat Ferland, Founder of V2 Cloud, joins AI InterConnect to talk about what happens when AI agents stop chatting and start doing real work on computers. V2 Cloud has built cloud desktops and infrastructure for businesses, managed service providers and software companies since 2012, and is now extending that infrastructure to a new kind of worker: AI agents.
A school project that became a company
V2 Cloud began as Hazeware, Mat's university project: could a desktop computer be rendered reliably inside a web browser? At the same time he worked for a software company whose desktop applications ran slowly on customers' old, poorly maintained computers.
That gave him the idea of moving legacy client-server applications to the cloud, so software vendors could offer them through a browser without rewriting anything. When COVID sent people home, the product pivoted to give remote employees secure computers.
A secure computer in a browser
Mat starts with the problem: how do you give remote employees, in another state or country, the tools they need without putting company data on their personal devices? V2 Cloud provides a virtual computer, hosted in the cloud, with the applications, data and security settings already in place.
A business owner can sign up online and pay monthly. When an employee leaves, no data remains on their device, and access can be switched off instantly. Against Amazon and Microsoft, which offer what Mat calls an empty shell, V2 Cloud competes with a turnkey service and real people on the phone when something goes wrong.
The brain and the hands
Mat explains today's AI with a simple picture. The large language model, from OpenAI, Anthropic or Gemini, is the brain. The harness is the hands: the software that lets that brain take actions on a computer.
Vision models can now read a screen, much as a person does, and click, type and complete tasks. As a virtual desktop provider, V2 Cloud saw the fit and introduced a copilot that can navigate websites or run an account reconciliation on a virtual computer. Computer use is still expensive, slow and imprecise, he says, but improving.
Boundaries and audit trails
An agent on your computer, Mat warns, is like hiring a brilliant computer scientist with your level of access. If it is compromised through prompt injection, it can pursue a different goal entirely.
His two rules are boundaries and audit trails. V2 Cloud installs agents on isolated virtual computers that cannot reach anything they are not meant to, and records every action with screenshots and logs, so failures can be traced and fixed. On responsibility, he thinks both AI companies and the people deploying agents should be accountable.
Automating apps without APIs
Most automation tools need APIs, but many businesses still rely on applications that have none. Computer use lets an agent operate those applications the way a person would, and Mat would invest heavily in making it faster, cheaper and more deterministic.
He sees the gap between open source and commercial models shrinking, pushing model companies up into the app and harness layer. For V2 Cloud, success means agents that let SMBs, IT teams, MSPs and software vendors deliver more output with the same people.
Advice for builders
After building a bootstrapped company through several technology cycles, doing every job himself in the early years, Mat now spends most of his time on new products, the part he loves.
His advice: reputation takes a long time to build and very little time to destroy, so do not rush AI into production the wrong way. And with so many AI startups, choose a niche where you can lead.
Watch the full conversation
Watch Mat Ferland explain how AI agents can safely automate work inside secure virtual desktops.
AI InterConnect hosts conversations with founders, investors and operators building with AI. Explore more episodes on the AI InterConnect podcast page.