The Verket Thesis

The most experienced person in any organisation spends most of their time on work that does not require their experience. They do this because the work requires absolute reliability, and expertise has always been bound to the person executing the task.

The deal of professional life has been that applying your highest intelligence requires showing up to do the tedious execution alongside it. This is purely an allocation problem. The most valuable cognitive resource in the economy — expert attention — is being consumed by work that requires it but does not deserve it.

There is a reason AI learned to write code before it learned to perform almost anything else useful: code has verifiers. A unit test specifies what correct behaviour looks like. The signal is binary. The function passes or it fails. The code runs or it doesn't. Models can practice against that signal, and every iteration builds on the last. Coding was just the beginning. The inference hunger of long-horizon agents is now snowballing a rush of domain-focused AI companies to post-train their models. But you cannot post-train without a reward signal.

Today, most of the work that moves the economy does not have these signals. Professional domains represent trillions of dollars in annual activity and contain the most consequential expertise in human civilisation. There is no unit test for a compliance review and no compiler to validate a supply chain exception. The operational rules exist, but they have never been extracted and compiled into a format a training system can learn from.

Professional expertise remains the largest body of uncompiled knowledge in the world. When the verification layer for real-world work gets built, the role of the expert fundamentally changes from doing work to directing it. Automation promotes the expert, and expertise moves upstream.

Our conviction is that agents will do work a thousand times more than humans ever could, and that most of that work will require automated verification. Under that premise, human supervision becomes the constraint that paralyzes deployment. You cannot hire enough humans to scale supervision, so you must automate it.

Our thesis is simple: artificial intelligence is the most profound technology in human history, with practically infinite global demand — but it's not reliable enough. The existing evaluation infrastructure relies on body shops and generic LLM judges that optimize for plausibility, are built for chatbots, and collapse under complexity. Verket is the update for the agentic era.

We are an applied AI company devoted to solving the agent reliability problem. We aim to scale human judgment to govern agentic execution and advance the frontier, specifically in regulated industries. We are taking a long-term view and pursuing radical new approaches to human-AI collaboration.

Every regulated domain needs a way to detect where agents fail against real-world constraints, route those failures to the right human experts, and codify the resolution for training and governance. Training gives agents capability, governance earns them trust. Verket converts this continuous loop into infrastructure that makes agents reliable enough for high-stakes work.

We believe that a focused attempt at encoding and scaling human judgment for agents is the most important infrastructure project of the next decade. It cannot be rushed, and it cannot be approximated. The fidelity of the signal matters more than the scale of the dataset. This is unglamorous work, but the payoff is absurdly large. Every codified human goal we build will create a template for the next. Every domain we transition from unverified to verified will expand the frontier of what agents can reliably execute.

When artificial intelligence becomes reliable enough to act within the messiness of reality, the nature of human work flips on its head. When the execution of work becomes infinitely scalable and near-zero in cost, human vision, empathy, and creativity become infinitely more valuable.

We subscribe to the vision that this autonomy will give humanity vast abundance, much higher living standards, and new services unimaginable today. And it will allow us to achieve multiple orders of magnitude more progress than previously possible.

This is the future Verket was built for. And getting it right is a generational pursuit.
‍
If you are interested in our work, please reach out.