Does prompting still matter? The new rules for getting what you want from AI (Forward Future)

Note to readers: what was downloaded is only the first part of a Forward Future article by Nick Wentz. The text cuts off at a paywall («Keep Reading», available to paying Frontier Pass subscribers). This summary covers only the readable portion. We don't know what the rest says, including the new rules promised in the headline.

The article argues that the old prompt-engineering tricks are becoming obsolete. At the same time, the skills that set advanced users apart from everyone else matter more as models improve.

The piece recalls the days when getting a good answer from ChatGPT required «magic words»: asking the model to act as an expert, to think step by step and to follow highly elaborate instructions. Forward Future acknowledges that it published its own role-based prompting guide in March 2025.

To illustrate the shift, the outlet surveyed its own team internally and found fairly aligned views:

• Alex McCraw, producer, used to write prompts such as «you are an expert graphic designer», on the idea that assigning an expert role would improve the AI's performance, but he no longer bothers with them.

• Matthew Berman, founder, considers prompt engineering, understood as squeezing out maximum performance with techniques like «think this through step by step», to be «basically useless» today.

• Brian Knepprath, editor, believes that overspecifying or spending too much time on a prompt is now «actively a waste of time». Until a few months ago he frequently used the /goal command, designed to keep agents from losing sight of the objective, but he now rarely uses it because ordinary prompts achieve the same thing.

• Nick Wentz, co-founder, agrees that the old techniques are losing relevance, but believes the bar for good prompting is rising at the same time. In his view, the main shift is moving from telling the AI how to do something to communicating what you want, why it matters and what a good result looks like. For him, the best users know how to frame the problem, provide relevant context and recognize a good result.

The article acknowledges a caveat. Those who say prompting is dying still get results that less experienced users struggle to replicate. It therefore concedes that the team is not representative of people who use AI for work. Its explanation is that the better the AI gets, the less you need to tell it how to think. Even so, you still need to know what you want, which instructions are critical and when to step aside.

On why the old tricks are losing their point, the piece notes that reasoning models are trained to work through complex tasks without the user having to ask. It adds that OpenAI has advised developers that simple, direct prompts tend to work better, and that asking for step-by-step reasoning may be unnecessary or even counterproductive. Nick also dropped elaborate personas, such as asking for writing in the style of well-known authors or for problems to be tackled like a world-famous researcher, because they rarely improve the result.

Berman offers an example from his software experiments. When he asks the AI for something similar to an existing product, such as an Excel-like app, he no longer needs to describe every feature in detail. Models know what a spreadsheet does and what its interface looks like, so he can focus on the outcome rather than the implementation. The article's conclusion is that if a short request is enough for a convincing prototype, it isn't worth spending much time perfecting the prompt.

The last readable section argues that the most important decision is how much freedom to give the model, and it captures a disagreement between two team members. Alex maintains that detailed instructions remain essential if you have a specific result in mind. If you are open to any interpretation, a short prompt can deliver a great result. But if you expect a particular style, structure and set of details, the result is likely to disappoint, because «AI can't read your mind». In his video work, for example, he has to convey the visual style, camera movement and sequence of events for the animation to match his vision.

Brian sees the opposite risk: by specifying every detail, users can prevent the model from discovering a better approach, and many end up «prompting themselves out» of surprising solutions. Nick adds that the value of detailed prompting depends on the complexity of the work. The text breaks off there, so we cannot report his full reasoning or the article's conclusions.