Better information beats better AI

Early movies about computers treated them as oracles. From WarGames to 2001: A Space Odyssey, the computer sits out of sight producing answers from thin air. Reality is less cinematic, and far more dependent on how we prepare the information those systems rely on.

The Hitchhiker’s Guide to the Galaxy captured this well. Deep Thought produces the answer “42” after an extraordinary amount of computation, only to reveal that the question itself was flawed. The problem was not the intelligence of the machine, but the structure of the problem it was given.

Most of what I hear about AI today follows the same pattern. The focus is on the model, on access, on speed. You need the best model, or so the thinking goes. What is discussed far less is the discipline required to make that intelligence useful in practice.

In a previous piece, I described AI as needing opposable thumbs. Intelligence on its own is not enough. Dolphins may be highly intelligent, but without thumbs they cannot translate that intelligence into tools that persist and compound. The same constraint applies to AI. It must operate across the realities of business, where information is fragmented across documents, systems and conversations, requiring access, integration and the ability to act.

An AI system may have access to your documents, emails and data, but if that information lacks structure it remains unusable. This is why using AI is an information discipline problem. In most organisations, information sits in random folders, half-finished presentations, convoluted spreadsheets and inboxes that have become informal archives.

The starting point is a discipline that feels familiar, even old fashioned. Meetings need to be captured in a way that preserves their substance. Not transcripts, but structured summaries that record what was decided and why. Without that context, a meeting is reduced to a moment in time rather than a building block in a larger body of work.

The same applies to analytical work. Spreadsheets that contain numbers without narrative are effectively opaque. They can be manipulated, but they cannot be understood. The explanation is part of the analysis.

Documents also need to be considered differently. Strategy papers, project updates and emails often exist as standalone artefacts, each capturing a slice of activity. For them to be useful, they need to be connected. Decisions should reference the data that informed them. Execution should link back to the strategy it is intended to deliver. Information needs to form a chain that can be followed, rather than a collection of isolated points.

For many professionals, email continues to function as a system of record. Important decisions and context are embedded in threads that are rarely revisited in a structured way. If those decisions are to become part of a usable dataset, they need to be extracted, clarified and connected to the broader body of work in which they sit.

Over time, I have found that desktop AI tools are most powerful when you treat them less as clever assistants and more as an extension of how you already work. I use them to capture the raw material of leadership work and shape it into something useful.

The approach is deliberately iterative. I let the system do a first pass, whether that is summarising a meeting, extracting actions, or turning rough notes into readable prose. I then step in as editor. Tone, sequencing and emphasis matter far more than speed at senior levels, so the system functions as a drafting partner rather than an author. The human pass is where judgement sits, deciding what really matters, what needs careful handling, and what should be recorded for the long term rather than just the next conversation.

A further shift is thinking longitudinally rather than moment by moment. Used well, AI helps connect work across time. Themes, risks and commitments are carried forward across meetings instead of being rediscovered repeatedly. That continuity is particularly valuable in complex or politically sensitive environments, where consistency of framing and memory matters. The system reduces cognitive load so that energy can be directed towards decisions rather than recall.

The important point is that this only works when it is embedded into a broader way of working. AI can support preparation before meetings, discipline immediately afterwards, and follow through over weeks and months. Used in this way, it does not make leadership mechanical or impersonal. It creates space for better conversations, clearer accountability and more considered judgement, which is where the real value still sits.

None of this is new. It reflects practices that have long been associated with effective operators. What has changed is the consequence of doing it well. When information is structured, connected and explained, AI systems can operate across it in a genuinely useful way. They can identify patterns that span multiple pieces of work, challenge assumptions based on prior reasoning, and connect decisions that were made at different points in time.

In the end, using AI is not a question of prompts or access, but of whether your work can be understood and used in a way that allows it to be extended. That is a higher standard than most current practice. It requires decisions, assumptions and reasoning to exist in a form that can be revisited and connected over time.