Business
How Businesses Should Prepare for Speech-Native AI Like Dessa
Dessa is not a business product. Block announced it on October 23, 2026 as a research preview, with no pricing, no API and no release date, and early access going first to a small group of AI safety researchers. So why should a business owner spend ten minutes on it today? Because what Dessa demonstrates — a voice model trained only on human speech that listens while it talks, hears frustration and handles interruption the way a person does — is the direction voice AI is moving, whether the winning system turns out to be Dessa or a rival. Preparation is cheap now and expensive later. This guide is about the preparation, labelled clearly as analysis: nothing below is a Block announcement.
First, understand what actually changes
Today's business voice tools mostly follow the old relay: speech is transcribed to text, a text model replies, and the reply is read aloud. Tone, hesitation and emotion are stripped out at the first step. A speech-native model deletes that relay. As we explain in what Dessa AI is, Block trained Dessa from scratch on conversation audio only, so timing and feeling are not features bolted on afterwards — they are the medium the model learned in. For a business, the practical difference is not a better robot voice. It is a system that can notice a caller getting upset mid-sentence, yield when interrupted, and keep a natural rhythm instead of forcing customers through walkie-talkie turns.
That is also why preparation starts with listening, not buying. Spend time with the live demo recordings on Block's official page and note which moments feel trustworthy and which feel uncanny. Your customers will make exactly those judgements, faster and less charitably than you will.
Audit where voice already touches your business
Most businesses already have more voice surface than they realise: the phone line, the drive-through or counter, appointment reminders, delivery updates, support calls, and the voicemail nobody checks. List them. For each one, ask three questions. How much of the conversation is routine information — hours, status, booking, directions? How much depends on hearing emotion correctly — a complaint, a confused elderly customer, a parent in a hurry? And how much involves trust, money or identity, where a convincing machine voice creates risk as well as convenience?
That audit maps neatly onto what speech-native systems are likely to handle first and what they should handle last. Routine, low-stakes exchanges are the natural starting point. Emotionally charged conversations are where a frustration-aware system could genuinely help — the scenario we explored in Dessa AI for customer support — but they are also where a mistake costs the most goodwill, so they need human oversight longest. Anything involving payments, codes or account changes belongs behind verification steps no voice should be able to talk its way around, whoever builds the system.
Fix the knowledge the voice will need
Here is the unglamorous truth every voice project learns: the model is rarely the bottleneck — the business's own information is. If your opening hours differ between your website, your listing and the sign in the window, a voice assistant will confidently repeat one of them and be wrong a third of the time. If your returns policy lives in the owner's head, no AI can state it consistently. Speech-native preparation therefore looks a lot like housekeeping: one accurate source for hours, prices, policies, menus and common answers, kept current by a named person.
There is a Dessa-specific wrinkle worth understanding. A model trained only on speech has, by design, never read your documents — a limit we set out honestly in what Dessa can't do yet. Whatever eventually connects a system like this to a business's facts — if anything does — Block has not announced it: no integrations, no tool use, no product form. Businesses that tidy their knowledge now will be ready for any voice system; businesses waiting for a magic import button will wait a long time.
Decide your disclosure and escalation rules in advance
A voice that sounds human raises a question your business should answer before any vendor asks it: will callers be told they are speaking with an AI, and at what point does a human take over? Decide now, while it is abstract. A reasonable starting position for most small businesses is simple: disclose at the start of the interaction, escalate immediately on request, escalate automatically when the caller is distressed, confused twice, or discussing money, health or a complaint — and never let a voice system handle identity verification by voice alone. Write it down. The businesses that struggle with voice AI publicly are almost never the ones whose technology failed; they are the ones whose rules were improvised after a bad call went viral.
Block's own rollout is a useful model here. A payments company putting safety researchers ahead of customers is telling you where it thinks the real work is. Our guide to whether Dessa AI is safe covers the impersonation and scam angles your policy needs to anticipate — including the mirror risk that criminals use convincing voices against your business, pretending to be you.
What to do this month — and what to ignore
Do: join the free waitlist at block.xyz/dessa if you want Block's own updates; run the voice audit above; clean up your core business facts; draft disclosure and escalation rules; and, if you need voice capability today, evaluate systems that actually ship — with the honest caveat, covered in our Dessa AI API guide, that Dessa offers developers nothing to build on yet. The scenario work in Dessa and Square restaurants shows the kind of labelled, speculative planning that is useful: it explores a possibility without pretending it was announced.
Ignore: anyone selling Dessa access, Dessa integrations or Dessa certification — none exist. Ignore fixed deadlines for readiness; there is no release date to be ready by. And ignore the pressure to rebuild your phone system around a research preview. Preparation means your information, policies and judgement are ready. The technology, whenever it arrives, is the last piece — not the first.
Sources: Block's official Dessa page (block.xyz/dessa) for research-preview status, access order and the absence of announced pricing, API, integrations and release date; The Verge and TechCrunch, October 2026. Business recommendations are DessaAI Daily analysis, not Block guidance.
Frequently asked questions
Can my business use Dessa AI today?
No. Dessa is a research preview with no business product, pricing, API or integrations announced. Early access goes first to AI safety researchers, and the public waitlist at block.xyz/dessa is free.
Has Block announced Dessa for Square or Cash App?
No. Block has announced no product form or integrations for Dessa at all. Articles exploring Square restaurants or support uses — including ours — are labelled analysis of possibilities, not announcements.
What is the most useful preparation a business can do now?
Audit where voice touches your customers, clean up the core facts a voice system would need (hours, prices, policies), and write disclosure and human-escalation rules in advance. Those steps help with any voice system, not just Dessa.
🚀 Get Dessa launch alerts
Dessa is still a research preview. Join the free DessaAI Daily list and we will email you the day access opens — plus the week's biggest Dessa stories, once a week. No spam, no hype.
Form not loading? Subscribe on the signup page →