A one-day intensive course on learning to build with agentic AI. Practical ingredients for AI transformation, for non-engineering knowledge workers.
Run by Toby White and Dino Repo of ArcticRex. We've led multiple transformations at scale, hands-on, and we believe that skill and capability lie at the center of every successful technology transformation.NPS 68 (Q1 2026, earlier versions of this course)
Co-founder and CEO of ArcticRex. Transformational leadership across multiple stock-listed companies, including CTO at F-Secure. Passionate about lifting organisational capability and unblocking team potential through hands-on training.
Co-founder and CTO of ArcticRex. Ran F-Secure’s AI transformation on the ground across 350 staff, engineers and non-engineers. Brings a builder’s instinct for what's needed to run change at scale, and how to teach practitioners to address it from day one.
Until a few months ago, building agents required engineering work. Claude Cowork and OpenAI Codex changed that. Non-engineers can now build agents on their own laptops, against their own work, without writing code — using the agentic capability that until recently was only available to developers.
Most people have not yet sat down and worked through what these tools can and cannot do. The day is built around finding out what the tools can do for you: hands-on exercises, learning to work agentically with your own data, surfacing where these tools work best, and what are the ingredients of success.
The day is built around the patterns that are key to taking best advantage of agentic AI in your work and the work of your colleagues.
Not every task will fit agentic automation, and the guidelines aren't always obvious. Through the exercises we work on, you'll develop the lens for looking at your own and your colleagues’ work agentically.
When to keep a human in the loop, when to let the agent run. The right answer varies by the work; we'll work through cases that pull each way.
Agent capability is limited by what the agent actually knows - which is much less than the humans operating it. Most of the input that makes an agent useful sits in the context, not the prompt.
An agent that helps one person is not yet an agent that helps a team. Sharing brings its own design problems — context, trust, discoverability — and the course brings them to the surface so they can be addressed in your own organisational scaling.
Giving a process agentic capability isn't just about speed - often it changes the shape of the process. We'll help you develop instincts for how agentic AI drives the reshaping of existing processes.
Sensitive data, audit trails, accountability. The new capability surface is also a new risk surface, and humans still need to be responsible for agentic decisions. The day works through where to look and what to weigh.
Alongside these patterns, we'll be demonstrating working habits with the tools as they exist now: how to write a brief an agent can act on; when to add more context and when to start fresh; how to detect hallucination before acting on it; how to manage model choice and token budget across a session. These are the small things that make the difference between an agent that helps and one that wastes the practitioner’s time.
The next 12 months will see as much new development as the last. We can't predict the future, but we can help you think about the way that likely future developments might affect what you should be building and how.
Many of these specifics will be obsolete within months as the tools change and new capabilities emerge. Probably the biggest takeaway we want to give you is how to live in this fast-moving landscape. This needs to be the start of your learning journey, not the end. We aim to leave each participant equipped to keep doing that learning and re-learning on their own.
You've been using AI-as-chat for a couple of years and it's changed how you work already. You've seen what's being unlocked with agentic AI over the last few months and you're excited. But you don't yet understand how to scale it for yourself or your organization.
Each participant must have access to either an individual high-capacity agent plan or a company-managed workspace with expandable usage. For Claude, use Max 5x, Team Premium, or Enterprise/API-backed usage. For Codex, use ChatGPT Pro 5x, or Business/Enterprise with extra Codex credits enabled. Lower tiers may work for light prework, but are not recommended for the full hands-on course.
This capacity is not just for the training day. Once participants are enabled with agentic workflows, they will need similar ongoing usage to keep building, iterating, evaluating, and delegating real work after the course.