If you've spent any time on X this week, you've probably seen clips of someone typing a single sentence in the chat and demo’ing a working video game in their browser, with no code or effort required to build out a game.

The current wave runs on Kimi K3, came out last Thursday, and it's tempting to put the whole thing down to the tool getting smarter.

That's only part of it.

The bigger factor is the person on the other end wrote a good request, and the way they did it is worth borrowing no matter what you use AI for.

Partner with us

Pazi

Takes a goal, such as launching a shop or selling a skill, and hands it to a team of AI agents that build a website, draft outreach, and create content. It comes from the team behind the open-source coding tool GPT Pilot and is backed by Y Combinator.

Jockey

Lets you search an entire video library in plain language, jump to a specific moment or spoken line, and create highlight reels. Built by the video AI company TwelveLabs, it is currently a free research preview and can connect to Claude.

Ditto

Copies any public website and returns it as clean, ready-to-build code in a few minutes. Made by ion.design, it is free and open source, and can be self-hosted or connected to coding assistants.

Take one of the builds making the rounds, a spinning roulette wheel you can play in your browser.

The request behind it ran to about 2 sentences, and instead of asking for something loose like "a casino game," it described a European roulette wheel with all 37 numbers in the correct order, the ability to give it a spin, and a ball that circles the edge before settling on a number.

4 small choices are doing the work there, and none of them involves writing more.

Say exactly what you want.

The example named a European roulette wheel rather than reaching for something generic, and that alone stops the AI guessing at what you had in mind.

Add details someone could check.

A line like "all 37 numbers, in the right order" can be verified just by looking at the result, while a vague wish such as "make it realistic" leaves the AI nothing concrete to work toward.

Describe what it should do.

Anything meant to move or react has to be spelled out, which is why the request included "spin it, and the ball lands on a number"; leave that out and you tend to get a static image that looks fine and does nothing.

Say what form you want it back in.

Naming the format up front, "as a web page I can open," heads off a round of back-and-forth about how to deliver the thing.

That's the entire recipe, and it fit in two sentences.

Once the approach clicks, the ceiling turns out to be surprisingly high.

The same method, handed a longer and far more detailed request, produced an entire city you can drive around in, along with an electric guitar you can pull apart on screen to inspect the parts inside and even "play" by clicking the strings.

You might expect that guitar to have come from some short, clever prompt, but the request behind it was enormous, naming every component it wanted shown and every sound it should make.

That seems to cut against everything above, until you look closer and notice that each of those extra lines was another concrete, verifiable instruction.

Every inch of that length was doing a job, which is exactly why it worked.

None of this is limited to people building games.

Say you ask an AI for a trip plan or a birthday message and it comes back not quite right; the reflex is to pile on more words and run it again, when what usually rescues it is a single concrete detail the AI can be held to.

Asking for "a 3-day trip for 2 with no flights before 9am" will get you much further than "a short trip," and "warm and casual, under 100 words" will land better than "make it nice" nearly every time.

Detail beats length.

The next time one of these tools hands you something you didn't ask for, resist the urge to bury the problem under more words.

The move that tends to work is the opposite instinct: pin down one thing you can check, whether that's a number, a name, a word count, or a particular behaviour you're after.

The people getting extraordinary results out of AI right now got there through that kind of precision, and precision has surprisingly little to do with how much you write.

Until next time,
Vaibhav 🤝🏻

If you read till here, you might find this interesting

#AD 1

Across product, ops, and CX teams, a new kind of role is taking shape: the person responsible for making AI actually work, day to day. In this roundtable, three people living this shift share what it's really like: Simone Santiago Broad (Yoco), Yelva Espinoza (Zumba Fitness), and Fin's Dave Lynch. You'll hear how they carved out these roles, what the job looks like across industries, the skills they'd hire for, and the challenges they're tackling right now.

Watch the full conversation on demand.

Watch now

#AD 2

The DTC beauty category is crowded. To break through, Jennifer Aniston’s brand LolaVie, worked with Roku Ads Manager to easily set up, test, and optimize CTV ad creatives. The campaign helped drive a big lift in sales and customer growth, helping LolaVie break through in the crowded beauty category.

Learn more