Get unstuck from the sprawl-or-stall loop
When I meet with Canadian business leaders, I’ve noticed two themes related to their AI challenges. The first one is frustration about moving too quickly, as they know employees are using unapproved AI to support day-to-day work. Perhaps someone in finance is running company data through a rogue tool, or someone from sales is feeding client information into an ungoverned platform. Regardless of where it’s happening, the genie is out of the bottle
When this leader’s head hits the pillow, the nagging worry is pointed: What sensitive company, client, or employee data is already being exposed through unapproved AI tools, and what regulatory obligations could that trigger in our industry?
The second concern is the opposite problem: moving too slowly. These leaders have clearly identified AI use cases that are ready to be turned into agents. They may have completed a successful pilot to prove the concept, but when it is time to launch, someone gets spooked about security or compliance, and the agent gets axed. In other cases, someone from the C-suite demands hard numbers about costs and ROI, and no one has trustworthy answers.
These leaders keep asking: What does it take to get our AI and agentic initiatives off the ground?
Despite the prevalence of these challenges, most of the noise in the market doesn’t address either one. Instead, the hype centers on two words — “at scale” — as if the answer is just to think bigger.
When it comes to agents, ‘boring’ beats bigger
In reality, both scenarios have the same root issue: a lack of capability.
Canadian organizations have ample AI ambition. Most have acquired tools. Many even have a substantial budget for AI. What they lack is a dedicated AI and data team with the capacity to turn business ideas into working agents, connect them to the right systems and data, govern how they are used, and measure whether they are delivering real value.
Without that capability, companies get stuck in a sprawl-or-stall loop.
Getting value from AI doesn’t require you to reinvent what matters to your business. In fact, I would assert that it’s better to be “boring” by starting with one repeatable, high-volume process that you already understand and measure.
Once you’ve stacked hands on a process, define success before you build anything. This step is crucial to answering the inevitable questions from your C-suite. If you can describe the process and articulate a positive outcome, you’ve already done most of the hard work.
Then — and only then — focus on the unglamorous parts:
- Is the use case specific enough to build against and measure?
- Is the data trusted, governed, and available to the agent?
- Can the agent connect to the right business systems, with clear controls over how it acts?
For many Canadian organizations, the answer must be “yes”: PIPEDA requires clear accountability for how personal information is handled, and some sectors may also face data sovereignty expectations around where data is stored, processed, and accessed.
That’s where the true complexity will always live. You can’t govern what you can’t see, so the same foundation that lets you deploy an agent safely also lets you watch it while it runs.
How OnX is helping organizations with AI agents
OnX has developed Forge Agents to help organizations tap into the capability required to build, deploy, and run AI agents. In doing so, we’re following two key principles:
- Your agents should live in your existing environment — in your own cloud, next to your own data — and should reach into your CRM, ERP, service desk, finance, and other business systems. Establishing these connections cleanly and securely is what separates a demo from something durable.
- Your business needs flexibility. You could hand the work to a single AI platform and hope they stay ahead. But today’s top model may not be the cheapest or best for the job six months from now, and the last thing you want is a digital worker you can’t move. The OnX approach enables you to stay agnostic: Pick the right model for each task. Send expensive reasoning to a capable model, high-volume grunt work to a cheaper one. Keep the freedom to make changes.
Unfortunately, you can’t skip the prep and expect that payoff. By doing the unglamorous work first, you go from needing outside help for every new idea to running the play at speed and scale. And because you set it up right, you know what you’re paying for (and what it should return) before you sign anything.
Turn agentic ambitions into business value
Whether you’re challenged by sprawl or stall, the path forward is the same: Pick one thing that matters, decide what winning looks like, and turn pressure into progress. Get in touch if you’d like to explore how OnX can help you choose your use case, build and implement an agent, and then govern and secure it.