Everybody has heard the promise by now. AI is going to save time, reduce costs, and help businesses get more done. The problem is that most people still don’t know where to start.
That’s the challenge Will Ruben is trying to solve with WorkClaw, a new product from Workmate Labs that turns AI agents into something closer to digital employees.
Ruben describes WorkClaw as “an AI team for your team.” Instead of asking users to learn prompt engineering or build complicated workflows, the platform lets them create AI teammates with specific jobs. A florist could train an AI to process invoices. A marketer could create a content assistant. An engineer could build a coding partner. The goal is not to replace workers, but to give every company access to the sort of specialised support that was once available only to large organisations.
The idea grew naturally out of Workmate, Ruben’s first product. Workmate focuses on scheduling, one of the most common tasks handled by executive assistants. After building an AI that could manage meetings and calendars, the company began looking at what else an AI teammate might be able to do. Recent advances in large language models made that expansion possible.
One of the most interesting parts of our discussion centred on a problem many AI founders rarely talk about. Traditional software is predictable. AI is not.
Ask a database the same question twice and you get the same answer. Ask an AI system twice and you might get two different responses. That creates challenges for companies trying to build reliable products.
Ruben compares the situation to earlier machine learning systems, including the recommendation engines that power social media platforms. The answer, he argues, is measurement, testing, and designing systems that can recover gracefully when things go wrong. If an AI makes a mistake, users need a way to correct it, and the system needs to learn from that correction.
That uncertainty also creates cost concerns. During our conversation I joked about running OpenClaw on a Raspberry Pi and accidentally generating a large OpenAI bill because a poorly configured process kept checking my email. Ruben believes those problems will become less significant as companies gain access to cheaper open source models and more efficient infrastructure. His view is that most business tasks do not require the most advanced models available today.
Perhaps the biggest challenge facing AI startups now is not technology but distribution.
Building software has become dramatically easier. Getting people to use it remains difficult.
Ruben said Workmate Labs relies on a mix of product-led growth, advertising, traditional sales, and good old-fashioned conversations with users. One tactic that has worked particularly well is identifying companies that visit the website, understanding who they are, and following up before interest disappears.
Looking ahead, Ruben says WorkClaw’s next step is reducing the friction involved in getting started. While the current product removes much of the technical complexity, users still have to decide what kind of AI teammates they want and how those teammates should behave. Future versions will offer ready-made AI roles, including executive assistants, marketers, engineers, salespeople, and operations staff, making it easier for businesses to start seeing value immediately.
The broader question is whether people want another tool or whether they want something that feels more like a co-worker.
Ruben is betting on the second option.
If he’s right, the future of software may not be a collection of apps sitting on a desktop. It may be a collection of digital colleagues working quietly in the background, each trained to do a specific job and each getting a little better over time.
That future is still taking shape. But products like WorkClaw suggest it may arrive sooner than many people expect.
