The useful bit of AI is the bit before the useful bit

For a one person business, AI can clear away much of the faffing between thinking of something and doing something useful with it. It can dig through manuals, perform calculations, organise untidy information, produce a first draft and provide something concrete to argue with.

That does not mean it replaces the work that matters most. Judgement, curiosity, talking to people, deciding what matters and taking responsibility still belong to the human being using it. The value of AI is that it can get us to those parts faster.

The paradox of using AI

There is an amusing contradiction built into learning how to use AI well. Sometimes we avoid asking AI for help because the task feels too trivial, too lazy or too easy to do ourselves. Then we spend far longer doing it manually than the task deserved.

That is not really a lesson about laziness. It is a lesson about friction. AI becomes useful when we stop treating every small act of searching, calculating, sorting and drafting as something that deserves our full attention.

AI cannot simply invent the answer to everything

There is also an important limit. We cannot type “find me the cure for cancer” into an AI system and expect a magical discovery to appear. AI can search, compare, summarise, calculate, generate hypotheses and help scientists work with large bodies of information, but genuinely new scientific knowledge still has to survive experiments, evidence, causation and real world validation.

The important distinction is between accelerating intellectual work and replacing the process by which knowledge becomes trustworthy.

An AI muggle’s journey

The most useful way to describe this experience is not as the journey of an AI expert, but as that of an “AI muggle”: somebody learning what these systems can and cannot do through practical use.

The journey began with a problem rather than a grand theory. Long periods of poor health had made sustained concentration difficult. Work could still be excellent in short bursts, but those bursts became less reliable and less frequent. When income became uncertain, a practical question emerged: when will the money run out?

The self healing budget

The answer was supposed to be a self healing budget spreadsheet. The idea was simple enough: gather expenses and expected income into one place, update it periodically from the bank and use it to estimate how long the available money would last.

This was not a new type of spreadsheet. Variations of the same model had been built many times before. The difference was that this time ChatGPT was asked to help.

The first result was disappointing. It was clunky, the formula structure was poor and it broke easily. That failure, however, proved more useful than a perfect first attempt might have done.

The first real lesson: AI gives us something to improve

The important shift was recognising that AI did not need to produce the finished answer. It needed to reduce the cost of getting from nothing to something.

A mediocre first draft is often valuable because criticism is easier than creation. Once something exists, we can point to what is wrong, ask for changes, test the result and continue refining it. The AI becomes less like an oracle and more like an extremely fast first pass.

This is one of the central ideas behind the wider AI Overlord or AI Overload project: AI is valuable because it shortens the distance between thinking and doing.

What AI should not replace

The parts worth protecting are precisely the parts that are easiest to overlook when AI appears clever. We still need to decide whether the answer is sensible, whether the assumptions are right, whether the result is useful and whether the consequences are acceptable.

That means AI is at its best when it moves routine cognitive effort out of the way and leaves more time for judgement, curiosity, conversation and responsibility.

A repository before a book

The AI Overlord or AI Overload material is being developed as a repository of working thoughts rather than a polished manuscript. The aim is to preserve examples, contradictions, half formed ideas and practical lessons as they arise.

Polishing can come later. For now, the useful task is to notice recurring themes, separate lived experience from broader claims and collect the examples that may eventually become chapters, arguments or teaching material.

Working thesis

AI is not valuable because it does the thinking for us. It is valuable because it shortens the distance between thinking and doing.

That is the idea worth testing against everything that follows.