Sesame's voice AI sounds great. Will anyone keep using it?

Launched 28 May, 39 countries, four agents, built on Sesame's Conversational Speech Model — the one that genuinely sounds human and whose CSM-1B base it open-sourced under Apache 2.0. Credit where it's due: the naturalness is real. Letting conversation flow instead of stalling mid-thought is a smart, hard-won engineering detail that most voice products get wrong. Sesame clearly got it right. That's the easy part of this analysis.

Now for the hard part. The graveyard of voice-first AI apps is not full of bad voices — it is full of great voices that nobody used twice. The pattern is embarrassingly consistent:

The pattern is not a coincidence. Voice is socially awkward in public, slower than typing for many tasks, and a behaviour people abandon when the novelty wears off. A startup needs that habit to form against Apple's own revamped Siri AI, on Apple's phone. That is a hard ask.

Day-eight retention. Not downloads, not the 39-country footprint, not how good it sounds in a launch video. If Sesame's voice app has 30 % day-eight retention it is a different kind of product than every voice assistant before it. If it has 8 %, it is the same kind of product — just with a better model underneath. We do not have that number. This is a preview, not a verdict. The point of saying it loudly at launch is that it's the only number the post-launch coverage will scramble to forget when it's bad and cherry-pick when it's good.

Sesame is an iPhone-first app competing in a market where the device maker is relaunching its own voice assistant — Apple announced a substantially rebuilt Siri AI at WWDC 2026 in June. That's a peculiar position: you're betting that users will choose a third-party voice app over the one that's baked into the OS, answers 'Hey Siri', and integrates natively with everything on their phone. It's happened before (Chrome on iOS > Safari for many users; Spotify > Apple Music), but those products had genuine platform-level advantages. Sesame's advantage is purely voice quality. Whether that's enough depends entirely on whether daily habit forms before Apple ships its own competitive experience.

So where does this land? Not dunking on the tech — it's excellent, the CSM-1B open-source release already proved that. Flagging that 'excellent voice' and 'product people use daily' have been different things for a decade straight. If Sesame breaks that pattern, it's a genuine story. Check the retention curve, not the launch buzz. That's the receipts that matter.

If you want to know whether Sesame's bet is paying off, ignore the download announcements and watch for: (1) a public retention metric from Sesame, or independent app-analytics data (Sensor Tower, data.ai); (2) when the promised Android preview actually ships — it's been announced with no date, and every month it slips leaves the global majority of phones out and Apple's own Siri AI a clear run; (3) a monetisation announcement — 'free for now' is not a business model, and free previews don't fund model development indefinitely. The moment any of those three signals is clear, you'll know far more about this company's prospects than the launch-day coverage tells you.