Most AI discussions revolve around model capability — better reasoning, more context, higher benchmarks. But AI adoption rarely follows capability. It follows how naturally the technology fits into the workflow people are already in.
This is part of the Hidden Variables in AI Strategy series — examining the factors that determine whether AI initiatives deliver real organisational value or remain stuck in perpetual pilot.
The Bolt-On Problem
Right now, much of enterprise AI is being added as an adjacent feature. Side-pane copilots. Separate chat interfaces. Copy-paste prompts into external tools. AI becomes a wingman sitting next to the system rather than part of it.
The problem is not the model — it is the product design. Humans are fundamentally habit-driven. We rarely change our tools unless the experience itself changes. Asking someone to open a new pane, switch context, and formulate a query during a high-pressure workflow is asking them to override muscle memory. Most of the time, they will not. The feature will exist. It will not be used.
A Pattern Technology Has Seen Before
This is not unique to AI. Every transformative technology went through the same inflection: from bolt-on to built-in.
- Security adoption accelerated when it became built into operating systems — not when it remained as separate antivirus software users had to install and remember.
- Accessibility improved dramatically when it became part of core UX design standards — not when it was an afterthought patched in post-launch.
- Cloud infrastructure scaled when APIs were designed into platforms from the start — not bolted onto legacy systems after the fact.
Technology spreads fastest when it becomes native to the product architecture. Not when it becomes another tool in the toolbar.
From the Field: The Incident Resolution Experiment
I experienced this pattern directly while building an AI solution to accelerate live-site incident resolution.
Our first approach was straightforward: a conversational AI interface. On-call engineers could query system telemetry, retrieve diagnostic insights, and run analysis through a chat pane attached to the incident management tool. Technically, it worked well. The model was accurate. The latency was acceptable.
Adoption was negligible.
The Redesign
We stopped asking engineers to consult the AI. Instead, we made the AI run automatically when a ticket was created. By the time the on-call engineer opened the incident, the system had already generated context, identified impacted services, and produced a preliminary root-cause hypothesis — surfaced within the ticket interface itself. Engineers could dig deeper if they chose, but the first layer of analysis was already there. Engagement increased multiple-fold. Not because the model improved. Because the product design changed.
The engineers did not adopt the AI because it got smarter. They adopted it because it stopped requiring them to do anything differently. It was simply there when they opened the ticket.
Capability attracts attention. Native integration within product design drives adoption.
Three Questions Every AI Initiative Should Answer
Before asking whether a model is good enough, ask whether the integration is designed for adoption. Here are the three tests:
1. Is AI embedded natively in the product?
Does the user have to leave their current interface or open a new tool to access AI? If yes, you have a bolt-on. Bolt-ons rarely scale. The goal is to make AI invisible — already running within the flow the user is already in.
2. Does it preserve existing user habits?
Any friction introduced in a new workflow is a reason not to adopt. The most successful product integrations do not require users to learn new patterns — they make existing patterns easier. If your AI integration requires behaviour change, expect resistance proportional to the change required.
3. Does it remove steps rather than introduce them?
This is the clearest signal of whether an AI feature is additive or transformative. Count the steps in the workflow before and after. If the number goes up, the AI is adding cognitive load, not reducing it. The best integrations make users faster at what they were already doing — not better at a new skill they must acquire.
The Bottom Line
The real shift in AI adoption will not come from smarter models. It will come when products are designed AI-native from the ground up — where intelligence is a property of the workflow, not a feature attached to it.
Capability attracts attention. But it is product design that converts attention into adoption. And adoption is the only thing that determines whether AI delivers real organisational value.
Disclaimer: The views expressed are those of the author and are for informational purposes only. They do not constitute financial, legal, or investment advice.
