Reporting in March said Apple had finished HomePod and Apple TV hardware sitting unreleased. The Siri those products were built around wasn’t ready. So the hardware waited.

Most of us have lived a smaller version of that. One unfinished piece holds up everything attached to it, and the cost climbs every month you don’t ship.

I’m sympathetic to Apple here. An assistant that fails is worse than an assistant that arrives late, and three products were riding on this one.

But sympathy isn’t a plan. What I’d want to know, in Apple’s position or my own, is how you tell a considered wait from a stall.

The meeting where this actually happens

Picture the roadmap review. The team wants another quarter. They’re not being evasive. They genuinely believe the thing isn’t good enough yet, and they’re probably right. Someone from sales points out that two customers have been told it’s coming. Nobody in the room has a way to settle it, so the decision defaults to whoever sounds most confident.

What makes it circular is that “not good enough yet” has no shape. You can’t argue with it and you can’t verify it.

A good wait has a shape. The team names the specific failure they’re fixing, shows you the test that proves it, and comes back with the result. A stall says “we’re protecting the experience” and books another review.

What Apple was protecting

Apple’s version had a real answer, at least. In June 2024 it described a Siri that could use your personal information and act across your apps. By March 2025 it acknowledged that delivery would take longer than expected.

The distance between those two things is bigger than it looks. Consider drafting an email against sending one. A draft is a suggestion, and you read it, fix what’s wrong and decide whether it goes. Send the wrong attachment to the wrong client and no amount of improvement afterwards undoes it. That gap between suggesting and acting is the ladder I set out in the five levels of agentic software, and I’d say most businesses sit lower down it than their vendors imply.

Being slow about that seems reasonable to me. I’d want a great deal of confidence before handing software that job in my own business.

The bill lands next door

The cost turns up in whatever was only waiting.

Bloomberg’s March reporting, covered by MacRumors, put the smart home display back from spring 2025 to spring 2026 and then September. Those were reported internal targets rather than dates Apple gave customers, which matters for how harshly you judge the delay. Either way, one difficult dependency was holding up separate finished products.

There’s usually a middle option. Google put its Gemini task automation into beta in February, for a handful of food and ride-hailing apps, in two countries, with the user able to watch it work and interrupt it. That’s a much smaller promise than Apple made. It also buys real customer evidence while the full version is still being built.

Apple took a different route out. Its January agreement with Google provides the model technology underneath Apple’s next generation of assistants, with Apple still running things on its own devices and private cloud. That’s meant to solve the delivery problem, and it creates a different one, because the part customers talk to most now comes from a competitor.

Whether Apple’s integration advantage outweighs that dependency is genuinely open, and I don’t think anyone knows yet. I’d want an answer before making any supplier hard to replace.

Four things to bring to the meeting

None of us are shipping Siri. The argument in the room is the same one, just cheaper to get wrong.

  1. Define what “working” means. Replace “the AI needs to be better” with a task you can test. Be precise about how much the software decides on its own, because plenty of products sell the agent and ship the assistant. For a support assistant, that might be finding the right customer record and drafting an accurate reply, with a person sending it. Include the awkward cases, like two customers with the same name.

  2. Ship the part that already helps. A draft waiting for approval saves time even when you can’t trust the software to send it. You can take the useful half now and earn the rest later, which is what Google’s limited beta amounts to.

  3. Put a date on the delay. Give the team a fixed period and a specific problem. At the end they bring evidence, and you decide together: carry on, cut the scope, change supplier, or stop. A review whose only output is another review date needs a better explanation.

  4. Check what’s queued behind it. Look for launches, hires and customer commitments that assume the feature lands. Some of them could go ahead with a simpler version, or with someone doing the work by hand for a while. Apple’s idle hardware is the expensive version of that problem.

Back in the meeting

The point was never speed. It’s that the roadmap review stops being a contest of confidence.

You end up discussing a named failure, a narrower release or a different supplier, all of which are easier conversations than defending a date. Your sales colleague gets something to tell those two customers: here’s what works today, here’s what we’re still testing.

Siri is due to arrive on 14 September as a beta, in English, on supported devices. Apple will find out soon enough whether the wait bought what it was supposed to buy. The rest of us can find that out a good deal sooner, and more cheaply, by shipping the small useful part and letting customers tell us what’s missing.