Product Fieldnotes

Apple’s 512GB Mac Studio costs $22,299, and one of its big selling points is running someone else’s AI models. I can’t decide if that’s a brilliant hardware strategy or the most expensive possible admission that Apple Intelligence isn’t the main attraction.

A seller asked an AI agent to help sell a keyboard. It accepted an offer and shared his home address. You can correct a bad price. You can’t unshare an address. For product teams, this is the real agent UX problem: which actions must require explicit approval?

Anthropic files for an IPO. Meta plans to raise up to $95B for AI. Huge bets. Fair enough. But if your product team can’t name the specific customer headache AI removes, you’re not building a better product. You’re adding an expensive adjective to the roadmap.

Ever tried cancelling a subscription and spent 10 minutes just looking for the button? Some product teams call that retention. I’d call it a product failure. You might save one month’s revenue. But that customer is never trusting you again.

Target just launched an AI shopping assistant. I get the idea. But sometimes the best shopping experience is still three filters and a checkout button. Not every customer problem needs a conversation. Are we solving a problem or just adding AI because we can?

Zoom wants AI to turn meetings into completed work. Good. Because nobody needed another meeting summary. The real problem? Decisions that disappear the moment the call ends. If nobody owns the next step, AI is just a very expensive note-taker.

OpenAI is putting its models on Google Cloud. Two companies competing in AI, sharing a checkout counter. That’s a better product lesson than most strategy decks: customers don’t care about your rivalry. They care how much friction you remove.

Spotify is adding paid podcast memberships. Good for creators. Fair enough. But I'm curious how many subscriptions people are willing to juggle before listening to a podcast starts feeling like managing their electricity bills. Every monetization feature looks great on a roadmap. Not always on the customer's bank statement.

One team updates the offer. Another updates the email. Another updates the account page. All three tickets are marked Done. The customer still sees three different prices. Funny how every team can deliver exactly what was asked, and the product can still be wrong.

In a product meeting this week, someone asked: “What happens if the customer does nothing?” We had a flow for Yes. We had a flow for No. But doing absolutely nothing? That turned out to be the most complicated scenario. Funny how often we forget that silence is a customer decision too.

“We’ll handle those customers manually.” I've heard this in more product meetings than I can count. Sounds reasonable at first. Six months later, there's a spreadsheet nobody trusts, three teams involved, and a workaround that's somehow become the official process. And we're still calling it temporary.

Tesla just showed a Roadster that can fly. Insane engineering. Seriously. But this car was announced in 2017. Now production is targeted for 2028. As a PM, I can’t decide what’s more impressive: the technology or how long people will wait for it.

Figma is making it easier for AI agents to do the work. Cool. But here’s what I’m curious about. If your team already struggles to agree on what to build, what happens when everyone can generate 10 versions of it before lunch? Faster execution doesn’t fix indecision. Sometimes it just makes the mess bigger.

Apple just announced a $299 AI home hub. And I have the same question I have in half the product meetings I attend: Who actually asked for this? Maybe it'll be great. Apple has surprised us before. But I'm getting a little tired of companies treating "we can add AI" as proof that people need another device.

Amazon blocking Meta's AI shopping agent is pretty funny. For years, platforms wanted to remove friction from shopping. Now an AI wants to remove the platform from shopping. Suddenly, friction doesn't look so bad. I get the security concerns. But come on, there's a business model to protect too.

Microsoft’s president says milk is more regulated than AI. And somewhere, a product team is still being told to ship first and figure out the risks later. I’ve seen teams spend weeks debating button colors and 15 minutes on what happens when something goes wrong. Maybe that’s the part we should fix first.

Google, OpenAI, Meta. Everyone wants to build an AI agent that does everything for you. But here's the product question nobody wants to answer: If your AI books the wrong flight, spends $500, or emails your boss by mistake... Who owns the mess? The user, the agent, or the company?

One of the easiest mistakes in product is treating customer requests as requirements. Customers are usually very good at describing pain. They’re not always as good at prescribing the solution. “We need an export button” might really mean: “I don’t trust this dashboard.” The useful question isn’t just: What did they ask for? It’s: What problem are they trying to escape?

I think a lot of product teams confuse activity with progress. More tickets closed. More experiments launched. More dashboards. More meetings. But if the important decision is still unresolved, none of that really moved the product forward. Busy teams can still be stuck.

One thing I’ve learned in product: A slow decision can be more expensive than a wrong one. Wrong decisions create work. Slow decisions create waiting, rework, side conversations, parallel assumptions, and five versions of “what we think is happening.” Sometimes the real bottleneck isn’t delivery. It’s nobody wanting to decide.