For a while, much of the conversation about AI capability centred on prompting. It made sense. Better instructions often produced better results, and prompting gave people a practical way to improve quickly.
I no longer think prompting is the scarce organisational capability.
The harder skill is specification. Can we describe the outcome we want, the rules that apply, the constraints that cannot be crossed, the exceptions that matter, and the tests that show whether the result is good? That is a different discipline from asking a model to produce a polished answer.
This distinction becomes more important as AI starts to generate software and orchestrate work. I have been looking at approaches such as G5 Labs, where structured natural language becomes the source from which software can be generated. I do not know how quickly that model will mature or how widely it will apply. What interests me is the implication if it does.
If the cost of producing code continues to fall, the bottleneck moves. It moves towards the organisation’s ability to express intent clearly. The valuable asset is no longer only the application or its source code. It is the shared understanding of what the organisation is trying to achieve and how the system should behave.
That puts responsibility back into the business. A process owner cannot simply describe a feature and wait for IT to interpret it. They need to explain the outcome, the operating context, and what failure would look like. They need to make hidden rules explicit and resolve contradictions that have been carried for years through human workarounds.
Developers remain critical. Generated code still needs architecture, security, testing, performance engineering, model choice, integration, observability, and cost control. The opportunity is not to remove engineering judgement. It is to stop consuming judgement on avoidable translation and repetitive build.
This is also why specification-led development is not the same as uncontrolled citizen development. The aim is not for everyone to build anything. It is for the people who understand the work to describe it precisely, with engineers providing the standards and scaffolding that make the result dependable.
The more powerful the models become, the more valuable clear thinking becomes. Prompting may help us talk to AI. Specification helps the organisation understand what it actually wants.