There is a version of AI adoption that produces the opposite of its intended outcome. Too many tools introduced too quickly, with insufficient support and unclear purpose, doesn’t produce an AI-capable team — it produces a team that is actively resistant to future AI initiatives because they’ve learned that participating in them means effort without … Read more
The biggest productivity drag in most AI adoption initiatives isn’t the tools or the training — it’s the blank page problem. Every team member who wants to try AI for a new type of task has to figure out from scratch what prompt to use, what format works, what the AI tends to get wrong, … Read more
AI adoption rollouts are often evaluated at the point of launch — training sessions attended, tools licensed, enthusiastic testimonials collected. The more useful evaluation happens six months later, when the training energy has faded, the novelty has worn off, and you can see which changes were genuine and durable versus which were temporary spikes of … Read more
Rolling out AI tools means encountering objections — some rooted in misunderstanding, some rooted in legitimate concern, and some rooted in the entirely reasonable observation that not every AI tool is well-suited to every task or every person. How these objections are handled determines whether the AI adoption initiative builds trust or erodes it. The … Read more
The most effective AI adoption initiatives don’t rely primarily on top-down mandates or formal training programmes. They rely on the informal influence network that already exists in every organisation — the colleagues people actually ask when they’re stuck, the team members whose judgment is trusted because they’ve earned it, the people who make complex things … Read more
Most AI adoption initiatives are measured by the wrong things — training completion rates, tool licence usage, and enthusiastic testimonials at the monthly all-hands. These measure activity, not impact. They don’t tell you whether AI tools are actually changing how people work, which workflows have genuinely changed, or whether the investment is producing business value … Read more
Rolling out AI tools to a team and getting people to actually use them are two different problems. The technology is rarely the barrier — the tools are accessible, the interfaces are straightforward, and the potential productivity gains are real. The barrier is human: identity, trust, workflow inertia, and the legitimate concern that investing in … Read more
Most AI agents run the same task the same way every time. They have instructions, they have context, and they execute. What they don’t have is any memory of what happened last time they did something similar — whether an approach worked, what went wrong, what a particular supplier tends to do, or how a … Read more
Every message in a conversation costs tokens. In short interactions this is irrelevant — even a twenty-turn conversation fits easily in a standard context window at negligible cost. But AI assistants built for extended interactions — customer service agents handling long support threads, project management assistants tracking months of decisions, research assistants building on weeks … Read more
AI models are stateless by default — each new conversation starts from scratch, knowing nothing about the last one. For a one-off task this is fine. For a business AI assistant that interacts with the same people repeatedly, on the same projects, with the same preferences, it’s a significant limitation. The memory layer is the … Read more