Are you willing to prompt in emojis? Should your AI respond in Chinese?

Article — Feb 11, 2026

By Piers Fawkes · 4 min read

That's a thought I had last night while walking the dog and thinking about the conversations I've been having with enterprise data leads as I evangelize Fodda.

What's surprised me lately isn't how far behind corporates are on AI — it's how pragmatic they've become. There's a lazy narrative that brands are slow, agencies are smarter, consultancies are whizzier still. But the data leads and digital VPs I've been speaking with are quietly getting their houses in order.

At CES in Vegas, one head of data at a major alcoholic beverages company told me they had licensed 29 AI dashboards. What they really wanted was five.

He really meant five APIs.

"No one remembers their passwords anyway," he added.

The Token Economy Phase

The issue we're facing isn't access to AI. It's sprawl. And increasingly, cost discipline. When you're on a $20 subscription, tokens feel abstract. When you start wiring LLMs into internal tools via API calls, tokens become very real. Every long prompt, every uploaded deck, every agent trained on documents "just in case" burns budget.

Multiply that across teams and the CFO starts to grumble on the all-hands. The more serious enterprise conversations now aren't about which model to use - they're about how to reduce unnecessary token burn.

Instead of giving AI everything, the smarter move seems to be giving it exactly what it needs, exactly when it needs it. Thin queries. Structured retrieval.

Your marketing agent doesn't need every retail trend ever written to produce a shopper marketing plan. It needs the handful relevant to the task. That's a big part of why I've been building Fodda the way I have (yes, you knew I had to get there). The knowledge graphs in our marketplace can be queried precisely rather than ingested wholesale.

Language Compression?

Which brings me back to Chinese.

Chinese characters are semantically dense. In certain contexts, you can encode more meaning in fewer tokens than in English. So yes, I briefly wondered whether the next efficiency frontier is language compression. Could prompts become more symbolic? Could an AI system burn materially fewer tokens using Chinese characters than English strings?

It sounds absurd - until you start looking at the bills faced by corporations and startups alike. They need a precise set of insights used in real time, instead of mega-data dumped wholesale into prompts and context windows.

Less speculative compute, more surgical context: We're entering a token economy phase of AI. Exploration is still happening, but application has arrived - as signaled by how the finance guy is now popping up in weekly status calls.

(Sadly, after some research, I learned that Western LLMs actually burn more tokens writing Chinese than English - so until DeepSeek USA arrives, perhaps emojis are the real efficiency play??)

Piers Fawkes, Founder PSFK