Ross Dawson

<start of post>

The next phase of AI adoption will be decided less by the quality of individual tools and more by how organizations redesign the work around them.

Most companies are still treating AI as a layer that sits on top of existing processes. That can produce useful local efficiency, but it rarely changes how decisions are made, how knowledge moves, or how human expertise develops over time.

The more important question is not simply where AI can automate a task. It is how humans and AI should divide cognitive work across an entire workflow.

➡️ Start with the decision, not the tool

Define which decisions need to be made, what information is required, where uncertainty sits, and who has authority to act. Only then decide whether AI should support analysis, generate alternatives, monitor conditions or execute part of the process.

➡️ Make human contribution explicit

Humans add context, values, experience, judgment and the ability to recognize when the situation has changed. Those contributions should be designed into the workflow rather than assumed to appear automatically when an AI system reaches its limits.

➡️ Use AI to expand the field of possibilities

AI is often used to produce a quick answer, which can make the first plausible option feel like the best one. Better systems ask AI to generate competing interpretations, identify missing information, surface less obvious alternatives and show where the evidence is weak.

➡️ Build challenge into the interaction

People need a clear and accepted way to question AI outputs. This means making evidence visible, showing confidence and uncertainty, creating escalation paths and treating correction as part of normal work rather than as a failure of the user or the system.

➡️ Protect the development of expertise

If AI performs every difficult part of a process, people may become less capable of judging whether the output is appropriate. Junior staff therefore need opportunities to work through problems themselves, while experienced staff need enough visibility into the process to maintain and deepen their expertise.

➡️ Connect workflows rather than optimizing silos

An AI system can improve one team’s performance while making the wider organization slower or less coherent. The value comes from understanding how information, decisions, handoffs and accountability move across the end-to-end system.

➡️ Treat organizational memory as a designed capability

AI can retrieve information and identify patterns across large knowledge bases, but people still need to validate, contextualize and curate what is retained. Institutional memory is not just a search problem. It is also a question of what the organization chooses to remember and how that knowledge is used.

The central point is that successful AI adoption is becoming a problem of organizational architecture.

The organizations that create the most value will not necessarily be those with the most AI tools. They will be those that have worked out how to combine human judgment, artificial cognition, workflows, decision rights and feedback into a system that improves over time.

That requires more than prompt training. It requires redesigning work so that both humans and AI become more capable through the way they work together.

This is a central focus of my current work. More on that soon.

<end of post>