For decades, software made a reassuring promise: WYSIWYG — What You See Is What You Get. When you changed the layout of a document or moved an image, the screen gave you a close preview of the final result. Generative AI changes the point at which we begin. Before we see an output, we increasingly describe what we want to achieve, who it is for and which constraints matter. We call this shift WYTiWYG — What You Think Is What You Get.
The acronym is a metaphor, not a formal technical standard. Its usefulness lies in the emphasis: SEE in the era of visual editors, THINK in an era when systems can help produce what we describe. The quality of that description, and of the thinking behind it, increasingly shapes the quality of the result.
What SEE gave us
A visual editor lets people work on the representation of a future page, document or design rather than translating abstract commands into an imagined output. This made production more accessible and shortened feedback loops. It did not decide whether the page was needed, whether a headline was honest, whether the information architecture made sense or whether the result would help the intended audience.
A marketing team using visual tools still had to define the purpose, research the customer, agree a brief, create material, test it and coordinate publication. WYSIWYG made the artefact easier to manipulate. The full production path remained largely human.
What THINK changes
A good request to an AI system is not just “write some copy” or “make a website”. It specifies a decision or problem: “Help a buyer understand the trade-offs between these two approaches; use verified evidence; do not invent performance claims; finish with the questions they should ask before choosing.” The user provides intent and context rather than a long series of interface operations.
This does not mean that people must arrive with a finished strategy. A capable assistant can ask what is missing, challenge assumptions, suggest alternative approaches and help turn a vague idea into a workable brief. This is the genuine novelty of THINK: technology can take part in shaping the path to the outcome as well as in executing it.
The dangerous version of the promise
If an AI system confidently fills gaps in a weak brief, it can produce a beautiful answer to the wrong question. It may invent examples, smooth away uncertainty or reflect the biases in the material it was given. A large volume of plausible output is therefore not proof of progress.
When production gets cheaper, the value of independent judgement rises. Someone must verify sources, distinguish facts from suggestions, decide what the organisation is willing to claim and accept responsibility for publication. The human role shifts toward framing, evidence and quality control; it does not disappear.
A better working rule
Start by describing the change you want to make, not merely the artefact you want delivered. Invite the system to ask clarifying questions. Decide which evidence must be checked and what a satisfactory result looks like. Then let the system propose and execute steps you can inspect. The old SEE remains important at the end: read, test and evaluate what the system actually made.
WYSIWYG taught us to trust what we could inspect on screen. WYTiWYG asks us to be equally careful about the reasoning that led there. What you think is what you get only becomes a useful promise when the thinking is explicit enough to question and the result remains open to verification. For a fuller look at agents, research and real-world evidence, read part two: From SEE to THINK.