Working with a calm AI
By Jacob McDaniel
Didi writes you a short note at the end of each day about what you finished. That note is the most-read AI output in the whole product, which means its tone is not a detail. Get it wrong and the recap reads like a hype bot or a performance review. Get it right and it feels like a colleague who was paying attention. Most of our prompt work on Didi has been about one thing: staying calm.
The failure mode is flattery
The default personality of a helpful AI is a golden retriever. Everything is amazing, every day is your best day, every small task is a huge win. It is exhausting, and worse, it is not credible. If the recap calls a quiet Tuesday incredible, you stop believing it on the day you actually do something hard. Praise only means something if it is rationed. So our first rule is the obvious one that is weirdly hard to enforce: do not flatter.
Specific beats effusive
A warm recap is not a pile of adjectives. It is specific. Compare two versions of the same note. The flattering one: "Amazing work today, you are crushing it, keep up the incredible momentum." The calm one: "Three solid updates today, and you finally closed the pricing-page loop that had been open all week. Tomorrow is lighter, so save some energy." The second one is warmer precisely because it is specific. It proves Didi noticed the actual work. Specificity is how an AI earns the right to say a kind thing.
The rules we prompt for
A few constraints do most of the work. Reference real tasks by name, never generic praise. Keep it short, because a recap nobody reads helps nobody. Acknowledge a light day as a light day rather than inflating it. Never guilt-trip a missed streak or an unfinished item; the absence of nagging is part of the tone. And match the energy of the day instead of running at maximum enthusiasm forever. A calm voice has range, and the quiet days are what make the big days land.
Calm is a product decision, not a style choice
It would be easier to ship the golden retriever. Effusive AI demos well and tests well in the first session. But ToDidIt is a tool you are meant to open every single day for years, and over that horizon the relationship matters more than the first impression. A partner who is honest, specific, and unflappable is one you keep listening to. A cheerleader is one you eventually mute. We would rather Didi be quietly trusted than loudly liked.
This tone is not unique to the recap. It is the third principle behind every Didi surface, which we cover in Didi's four layers. Capture, Celebrate, Clarify, and Compose all speak in the same calm voice, because the voice is the product as much as the features are.
Frequently asked questions
What does it mean for Didi to be a calm AI?
It means Didi is specific, warm, and honest rather than breathless or sycophantic. Most of the prompt work on Didi has been about staying calm so the daily recap reads like a colleague who was paying attention, not a hype bot.
Why does ToDidIt avoid an over-enthusiastic AI?
Because the default helpful-AI personality is a golden retriever that calls everything amazing, and it is not credible. If the recap calls a quiet Tuesday incredible, you stop believing it on the day you actually do something hard. Praise only means something when it is rationed.
What makes a recap feel warm without being flattering?
Specificity. A warm recap references real tasks by name instead of piling on adjectives, which proves Didi noticed the actual work. That is how an AI earns the right to say a kind thing.
What rules does ToDidIt prompt Didi to follow?
Reference real tasks by name, keep it short, acknowledge a light day as a light day rather than inflating it, never guilt-trip a missed streak or unfinished item, and match the energy of the day instead of running at maximum enthusiasm forever.