Apple Intelligence has me worried for Apple

Yesterday’s Apple event showcasing their new AI features left me worried for Apple.¹

Apple’s strategy, to huge success, has been one of vertical integration - their software can do things that no one else can do because they also design and manufacture the hardware.

Apple’s brand, then, makes an extremely compelling promise: It just works, and it works like magic. AirPods immediately sync, compared to the usual fighting with Bluetooth pairing. You can seamlessly copy + paste from your iPhone to your laptop. You can use your iPad as a second monitor.

If you want a beautiful, polished, frustration free experience, Apple is for you.²

Yesterday, Apple showed off a bunch of genAI-powered features: Notifications that will summarize incoming email and text messages; a contextual camera app that will magically know your intent behind whatever picture you took and trigger the corresponding action (a ChatGPT search, adding an event to your calendar, looking up a restaurant’s reviews and menu); re-writing your notes into long form text; inventing your own emojis and images; a better Siri that integrates even deeper into the content and apps on your phone.

My concern is this: if it only works perfectly 70-80% of the time, similar most LLMs these days, Apple’s brand promise of “It just works, like magic” will be broken.

Our phone is our most personal and valuable gadget we own. We use it and depend on it hundreds (thousands?) of times a day, so often and so dependably that we forget just how much of a marvel it is.

Apple’s done the smart product strategy thing of integrating their AI into their existing functionality for maximum impact — IF it works. But what happens when, every time you pick up your phone to see the latest email or text notification, you can’t trust that the AI summary you are reading is accurate?

Getting it to 100% perfect — which is what people expect from Apple — is super hard. Like, no-one’s-figured-this-out-yet hard.

What they might do, in the face of this, is shrink the feature down to the most trivial, in hopes they increase the success rate or at least minimize the damage. In other words, make it less valuable. Not exactly a win-win.

As a steadfast and loyal iPhone customer since the very first iPhone in 2007, I’m rooting for them. But I’m also nervous, as they are wading into waters that are deeper and more chaotic than usual.

The Workshop

This is a newsletter-only section where I share a half-baked idea in hopes that y’all who are smarter than me can work it out with me.

When I first read about prompt engineering as a job title, I scoffed. And now, I’m coming around to it.

My initial reaction was, someone actually gets paid to use ChatGPT? Is it really that hard?

And now that I’m actually working with it every day, yes, it’s really that hard. Or at least, time consuming and specialized enough to be its own skill set.

Here’s my newfound appreciation: similar to Product Management, a great prompt engineer needs to be a hybrid expert that is good at hard skills and soft skills.

Hard skills:

• Wrangling the prompt itself: Obviously, knowing enough about how LLMs work to craft the prompt to do what you want it to do, and to do it reliably as opposed to “sometimes”.

• Subject matter expertise: Presumably, the application that is using the LLM is about something specific - healthcare, in my company’s case, but also could be engineering or law, etc

• Analytics: Is the version we just released objectively better than the prior? What data would we look at to determine that?

And also soft skills:

• Communication structure: As a conversational technology, it’s not just what it says, it’s how it says it. Does it respond in paragraphs and bullet points? Or 1-2 short sentences and then ask if you’d like more information?

• Tone: Should the response lean towards authoritative? Empathetic? Curious? Motivating?

• User experience: How do we want our users to feel when engaging with the AI?

To be honest, I’m somewhat begrudgingly coming around to the idea that this job sits in Product Management, for the companies that can’t afford a specialist hire. I’ve long preferred to stay out of the content — at Amazon, we build the UX containers for the products to live within, but we leave what goes inside those containers to the merchandising or marketing teams — but maybe in this new LLM world, we’re the right home for this work.

1  This Washington Post article was originally headlined “iPhone 16’s unfinished Apple Intelligence is useful except when it’s bonkers.” The headline is now the much more demure, “iPhone 16 is all about Apple Intelligence. Previews show it can be kind of dumb.” Presumably, to clarify that the journalist’s experience was based on a pre-release version of the software.

2  At a cost, which is you’re locked in to the choices Apple has made for you. Fewer settings, less customization, higher prices.