Engineering Director - Infrastructure

Voice AI still lacks its ChatGPT moment, PolyAI and Otter executives say. PolyAI and Otter leaders say voice AI needs faster reasoning, better transcription and transparency before it can deliver natural, trustworthy conversations. Oct 11, 2026 - 11:10 Voice AI has advanced rapidly, but executives from PolyAI and Otter say the technology still faces significant obstacles before it can deliver the natural, reliable interactions that made ChatGPT widely popular. Despite improvements in speech generation and real-time conversations, reasoning speed, transcription accuracy and user trust remain major challenges. Speaking at the HumanX conference, PolyAI Chief Technology Officer Shawn Wen said full-duplex models, which can listen and speak simultaneously, represent an important milestone. However, he argued that AI agents need faster reasoning to retrieve information and respond without disrupting the flow of conversation. Voice AI needs more than human-like speech. Wen said customer service agents must do more than sound natural. They need to show they can resolve problems, so callers build confidence after their first few interactions rather than immediately requesting human assistance. Alex Gay, Otter's chief marketing officer, identified similar challenges in workplace applications. He said effective AI meeting assistants require accurate speaker identification, an understanding of conversational intent and access to organisational knowledge to support automation. Otter is also developing digital twins that could represent individuals in meetings. Gay emphasised that these systems must reproduce the emotional qualities of human conversation, including the ability to participate naturally in discussions and debates, rather than functioning as basic question-and-answer chatbots. Transcription accuracy and transparency remain critical. Both executives highlighted limitations in automatic speech recognition technology. Wen said transcription systems can miss important keywords, while Gay warned that inaccurate meeting transcripts can undermine summaries and any automated actions based on them. Gay explained that transcription underpins Otter's broader productivity features. If the original record contains errors, subsequent tasks can produce incorrect results, weakening confidence in the platform. Transparency is another unresolved issue as voice AI becomes more common in customer service and workplace meetings. PolyAI believes callers should know when they are speaking with an AI agent, while Otter is exploring ways to notify meeting participants when conversations are recorded, even without a visible meeting bot. For both companies, improving voice quality alone is insufficient. Faster responses, dependable understanding and clear disclosure of AI participation remain essential to making conversational systems trustworthy for everyday use.

Wispr Flow looks great. It just doesn't exist for Windows. You read the tweets. Wispr Flow's $14 a month and it handles voice writing better than anything else out there. Cold emails, Slack threads, support replies, and Twitter drafts, all dictated and automatically cleaned up. Category leader. So you try it. The web demo works. Actually works. You dictate a cold email to a design partner, hit send, and it reads like you spent ten minutes polishing it. Fourteen bucks a month isn't nothing if this saves you the Thursday morning ritual of drafting, rewriting, drafting again, hating the tone, rewriting once more. Kristian pushes his bootstrapped SaaS to $8k MRR by blocking Thursday mornings for what he calls his "distribution batch." One hour to draft a week of cold outreach, investor updates, support replies, and social posts. Every minute rewriting eats into the next batch. A tool that cuts that time in half is net free money at his salary math. Then you check your OS. Windows. And Wispr Flow stops existing. The Mac monopoly. Here's what happened: Wispr Flow launched on Mac first. Then iOS. Then Android. Windows never materialized. The founder team probably made a rational choice: ship on the platforms where paying users were (designers, writers, content creators on Mac) and solve the biggest market first. Makes sense. But it created a gap. Every serious voice writing tool's done this. Otter.ai launched premium features on Mac and iOS before Windows got parity. Speechify too. Windows users get the free tier, or nothing. It feels like every creator-software premium vendor took an oath that Windows users shouldn't get the same choices for the same price. The cost of that gap. Kristian's not alone. A solo founder on Windows, bootstrapping a B2B SaaS, is doing 30 to 40 percent of his day in writing. Not code. Writing emails, docs, Slack threads, support replies; the asynchronous handoff that keeps a company moving when it's one person. He already uses Cursor for the coding part. The time-value math is iron: a tool that returns 30 minutes a week at any real salary pays for itself forever. Wispr Flow could be that tool. It just doesn't exist on Windows. So Kristian does what every bootstrapped founder does. He types. He drafts four times. He reuses templates. He hits send and moves on. The Thursday distribution batch takes 90 minutes instead of 45. Why this actually matters. The gap isn't an accident. It's a market structure problem. Premium creator tools assume you own a Mac. If you don't, you get the mobile experience, the web demo, or the free tier, and you're watching high-paying users on the platform you wanted to be on. This isn't about resentment (tho there's some of that). It's about math. If a tool solves a real friction point in your day and it costs less than one cold email that converts into a paying customer, you shouldn't have to switch operating systems to use it. Recitey was built for this gap. What Recitey does differently. It runs Whisper locally on your device. No variable cost. No metering. No word cap per month. You dictate a Slack message, an email, a GitHub comment, or a support reply. It runs the model on your device, polishes the output, and sends it via your system clipboard. Works in every Windows app. Works offline. No word limit. OAuth login, no credit card required. The early traction: bootstrapped founders, consultants, engineers who spend their days writing explanations and follow-ups. Anyone who tried Wispr, loved it, and then checked their OS. The math. Kristian's Thursday batch. One hour every week. Recitey cuts the rewriting time to maybe 20 minutes. That's 40 minutes a week, or 160 hours a year. At a $150k salary, that's about $11,500 in reclaimed time. If it costs $14 a month, that's $168 a year. The break-even happened in the first two minutes of first use. The reason Wispr launched on Mac first is still sound. Most creator tools should. But the reason to use Recitey on Windows is simpler: every other choice makes you leave money on the table.

312-499-6335 415.774.2973 When AI takes notes: court allows privacy claims against Otter.ai to proceed. Monday, August 31, 2026 Otter.ai's AI notetaker tool has been accused of violating Illinois's biometric privacy law, and the California court hearing the case recently let the proceedings move forward. As we have previously written in our sister blog, AI notetaker tools carry a growing range of legal risks. These include exposure under wiretapping and biometric privacy laws. According to the plaintiffs, Otter's AI-powered notetaker joins virtual meetings and records and transcribes participants' speech in real time. They also allege that Otter keeps the meeting data and uses it to train its AI models without getting consent from non- subscribers. This, they allege, violates Illinois' biometric law (BIPA) as well as eavesdropping laws. Namely, the federal Electronic Communications Privacy Act (ECPA) and California's Invasion of Privacy Act (CIPA). The court allowed all of these claims to proceed. While the ruling addresses only the pleading stage and does not establish that Otter actually violated any of these laws there are several things worth keeping in mind. Regarding California's wiretapping law (CIPA), Otter argued that its bot was an authorized meeting participant and therefore a party to the conversation. The court disagreed because Otter uses the notetaker for its own commercial purpose: training AI. This was unlike a prior case, Graham v. Noom, Inc., in which the company's recording tool acted as an extension of the customer who deployed it. The court did set limits. It dismissed privacy claims where plaintiffs simply called their conversations "private" without alleging specific facts showing they had a reasonable expectation of privacy. But it allowed claims to proceed when the conversations involved sensitive topics, such as medical discussions. On the BIPA claims, the court found that Otter's speaker-tagging features plausibly involve capturing voiceprints. I.e., biometric identifiers under Illinois law. This allows plaintiffs' allegation that Otter violated the law by building speaker profiles without giving notice or getting consent. It also allows the allegation that the company should have had -and published- a written retention policy to proceed. The court rejected Otter's extraterritoriality defense, noting the plaintiffs were Illinois residents physically in Illinois when their data was collected. With BIPA carrying liquidated damages of $1,000 per negligent violation and $5,000 per intentional one, these claims add meaningful exposure. Putting It Into Practice: This ruling is a reminder of the importance of understanding the extent to which notetaker tools might use information for their own purposes. As always, practical steps include conversations with vendors, not just building terms into contracts (although those matter too). And understanding with the business the true value of using a notetaker, and balancing it against the risks. Current public notices. Published: 27 August, 2026 Published: 26 August, 2026 Published: 24 August, 2026 Published: 20 August, 2026 Published: 17 August, 2026 Published: 12 August, 2026 Published: 10 August, 2026 Published: 4 August, 2026 Published: 31 July, 2026 Published: 27 July, 2026 Published: 1 July, 2026