User Interface Innovation Styles

Over the past few months, I’ve noticed a pattern in our system design conversations: they increasingly orbit around audio and video, how we capture them, process them, and extract meaning from them. This isn’t just a technical curiosity. It signals a tectonic shift in interface design. For decades, our interaction models have been built on clickstreams: tapping, typing, selecting from dropdowns, navigating menus. Interfaces were essentially structured bottlenecks, forcing human intent into machine-readable clicks and keystrokes. But multimodal AI removes that bottleneck. Machines can now parse voice, gesture, gaze, or even the messy richness of a video feed. That means the “atomic unit” of interaction may be moving away from clicks and text inputs toward speech, motion, and visual context. Imagine a world where the UI is stripped to its essence: a microphone and a camera. Everything else, navigation, search, configuration, flows from natural human expression. Instead of learning the logic of software, software learns the logic of people. If this plays out, the implications are profound: UX shifts from layouts to behaviors: Designers move from arranging buttons to choreographing multimodal dialogues. Accessibility and inclusion take center stage: Voice and vision can open doors, but also risk excluding unless designed with empathy. Trust and control must be redefined: A camera-first interface is powerful, but also deeply personal. How do we make it feel safe, not invasive? We may be on the cusp of the first truly post-GUI era, where screens become less about control surfaces and more about feedback canvases, reflecting back what the system has understood from us.

As more software switches to conversational UI, the software (and its designers) will have to understand better what "conversational" means.  I just saw this demo from Typeless - a dictation app - that perfectly illustrates this. Watch what happens when I dictate: "Let's meet at 10.. ummm... no, actually at 9" Most speech-to-text systems would transcribe this exactly as spoken—complete with hesitations, false starts, and corrections. But Typeless does something different. It understands that "ummm... no, actually at 9" is a self-repair sequence. This is a concept from the science of Conversation Analysis. Instead of outputting messy transcription, it delivers the clean intent: "Let's meet at 9" #ConversationAnalysis #ConversationalAI #VoiceUI #ProductDesign

Voice Agents Will Redefine the Consumer Experience I believe the next major disruption in consumer AI will come from Voice Agents — intelligent assistants that don’t just listen, but act. We're moving from a world of apps and taps to one where you simply say what you need — and your agent handles the rest. “Book my usual morning flight to Mumbai next week.” “Reschedule my doctor’s appointment to next Friday.” “Order my standard grocery list, but skip milk.” These agents will become our voice-first concierge, navigating complex decisions, commerce, and coordination across multiple platforms seamlessly and invisibly. Shopping: Agents that compare prices, apply coupons, and place orders Travel & Mobility: Bookings, changes, upgrades — handled autonomously Customer Service: No more IVRs — agents talk to support for you Finance: Bill payments, renewals, loan management through voice Productivity: Meetings, reminders, calendar syncing across devices Entertainment: “Play that show I paused last night” — no menus or remotes Some of the companies building here: Rabbit R1: A hardware agent that interacts with other apps via LAMs (Large Action Models) Humane: A screenless wearable assistant designed for ambient, voice-first interaction Inflection (Pi): Building emotionally intelligent AI agents with memory and personality Rewind AI: Creating a memory-first assistant that understands your context deeply Voiceflow: Tools for designing and deploying complex voice agents AssemblyAI, OpenAI (GPT-4o), Meta (Project CAIRaoke): Enabling foundational capabilities in voice, context, and real-time response. We see this as the most natural and inevitable UI shift of the AI era — from screen-first to voice-native, from self-service to agentic automation. As investors, we’re deeply excited about this space and actively looking to back teams that are: Building voice-first, action-oriented agents Solving challenges in real-time memory, latency, and voice UX Focusing on specific high-frequency, high-friction use cases in consumer life If you're building at the intersection of Voice + Agents + Real-world Action, we’d love to hear from you. #VoiceAI #ConsumerAI #AIUX #Agents #FutureOfInterfaces #VC #FoundersWanted Kalaari Capital

Four days of practice. 5 hours to build. Zero AI background between them🤯 Yesterday, a team I mentored won 1st place at TCS AI Friday, Bangalore. Monday to Thursday, we went through the basics, RAG, agentic AI, voice agents, computer vision. None had an AI background going in. By Friday, they were making architecture calls on their own, the part I'm proudest of, and why I keep mentoring alongside my day job auditing how AI systems reason. They picked travel as the use case. The problem underneath shows up everywhere: banking, healthcare intake, support lines, any market that isn't monolingual. 𝗣𝗿𝗼𝗯𝗹𝗲𝗺 Voice assistants assume one language at a time. Say "Koramangala se airport," a normal way to book a ride in Bangalore, and most speech engines break on the code switch. This isn't a travel problem, travel just exposes it fastest. The same failure sits inside every voice system built for a country where people mix languages mid sentence. Zero visual confirmation makes it worse too. Voice runs blind, users fall back to typing, and the point of voice first disappears. 𝗦𝗼𝗹𝘂𝘁𝗶𝗼𝗻 The team built Namma Ride, a voice first ride booking assistant that switches between Hindi, English, and Hinglish mid sentence, with the interface reacting live. Travel was the proving ground, but the architecture is domain agnostic, swap the map for a claims form or intake screen and the pipeline still holds. Core idea: separate transcription from state sync, solve both in parallel, not in sequence. 𝗔𝗿𝗰𝗵𝗶𝘁𝗲𝗰𝘁𝘂𝗿𝗲 🎙️ Parallel STT, twin Whisper instances (English plus fine tuned Hindi) via asyncio.gather, merged by an LLM that fixes colloquial terms, "kora mangla" becomes Koramangala 🔊 Silero VAD on 16kHz audio, end of utterance after 1s of silence, no push to talk 🔀 WebRTC for audio, WebSocket for control, interface updates live without interrupting the mic 🗺️ Backend driven UI, the LLM mutates server state and emits a structured event to animate the client, no text prompts 🧠 Stateful memory across corrections, plus bounded spatial matching, no exact spelling needed The hard part was never transcription. It was keeping voice and interface synced live. Four days earlier, most hadn't touched WebSockets. Friday, they debugged it themselves. An assistant that understands how Bangalore actually talks, not a translated version of how Silicon Valley talks. Strip away the ride booking layer and what's left is a reusable pattern for any voice system serving a code switching population, far bigger than one travel app. Congratulations to the team, you earned this. Thanks to Tata Consultancy Services and TCS AI Fridays for organizing an event that pushes people to actually practice, and gives them room to build real AI solutions for real problems. Follow along if you're building or auditing production AI. 🙌 #AgenticAI #AI #Hackathon #TCS #TCSAIFridays #AIEngineer #AIEngineering #Community #Mentor #VoiceAI #Multilingual

Today, voice assistants are everywhere. Siri, Alexa, Google Assistant shifted expectations. And that shift is only accelerating with GenAI. But voice isn’t just a “new input method.” It’s a new interaction paradigm, one that demands a new design mindset. Here’s what I’ve learned designing for voice-based systems: 1. Design for intent, not clicks. Unlike GUIs, Voice UI has no hover states, tooltips, or buttons. The entire experience hinges on understanding user goals. That means more research, smarter NLP training, and tighter context modeling. 2. Make failure graceful. Voice mishears. It misfires. Designing fallback paths (without sounding robotic) is key to user trust. “Hmm, I didn’t get that” isn’t enough. Guide users without making them feel stupid. 3. Clarity beats cleverness. Conversational doesn’t mean casual. Your voice system should speak like a helpful human — not a quirky chatbot. Keep responses short, actionable, and reassuring. 4. Build turn-taking into the experience. Real conversations have rhythm. So should voice UIs. Use pauses, confirmations, and cues to make users feel heard — not interrupted. Designing for voice is less about visuals and more about behavioral psychology, linguistics, and systems thinking. Done right, it feels like the future. Done wrong, it feels like 2006 speech-to-text. We’re not just designing what gets said. We’re designing how humans feel when they say it and when machines listen. What’s the most frustrating (or delightful) voice interaction you’ve experienced? ♻️ Share if it was useful. 🔔 Follow Artem Borysenko for more updates!

Most apps still expect users to learn the app before they can get anything done. But what if the user could simply say what they want? I have been exploring a small proof of concept around Bangla voice-led app automation. In the demo below, a user says: “Amar Airtel number e 30 taka recharge koro.” The system parses the Bangla voice command, understands the intent, identifies the operator and amount, opens the bKash app, and moves through the recharge flow. This is still an early demo, but it points to a bigger product shift. For years, we have designed apps around navigation. Buttons. Menus. Forms. Tabs. Screens. But as LLMs get better at understanding regional languages and converting unstructured text into structured actions, the interface can become much simpler. The user does not need to know where the feature lives. The user only needs to express intent. This is also why I find the Rabbit R1 story interesting. Rabbit tried to create a new hardware category around this idea. But the bigger opportunity may not require new hardware at all. It may come from apps and platforms exposing their services in ways that AI agents can safely act on top of the devices people already use. We are already seeing early signs of this in enterprise software. Salesforce has released its MCP Server in beta, which allows AI assistants and agentic IDEs to interact with Salesforce without the user clicking through the Salesforce UI. Of course, financial workflows need a much higher safety bar. Anything involving payments, PINs, recharge, send money, or banking actions needs strict permission handling, user confirmation, and security controls. But even with those caveats, the direction feels clear to me. The next major UX shift may not be a better menu. It may be removing the menu altogether. What do you think: will voice-led agents become a real interaction layer for most apps, or will trust and security concerns slow this down? #AI #AIAgents #Bangladesh #Fintech #ProductManagement #UXDesign #VoiceAI

Voice AI is moving from listening and speaking to acting, interpreting emotion, controlling tools, and raising much harder questions about consent and trust. Here's what stood out 👇 - Alibaba Group launches Qwen Audio 3.0 with real-time voice that can proactively use external tools, covering 113 languages for ASR and 36 for TTS. - Meta patents an AI wearable that continuously analyzes voice to track the user’s emotional state, raising concerns under the EU AI Act’s emotion-inference ban. - PwC and OpenAI launch agentic customer service solutions combining PwC’s CX expertise with OpenAI’s multimodal voice and digital agent APIs via Rhys Fisher for CX Today. - Google Voice adds Gemini AI notes that auto-summarize calls with key points and action items, plus new standalone plans starting at $10/mo. - Google quietly opted users into AI training on voice queries and uploaded media via a new Search Services History setting, with no opt-in required. - Rime raises $24M Series A to build enterprise speech-to-speech models, powering nearly 100M phone calls monthly for Mayo Clinic, Dialpad, and others. - LALAL.AI launches Lynx, a neural network built for speech denoising that is 6x smaller than its flagship model while matching output quality via Slator. - GIGAChat adds emotion detection and can process audio up to three hours long with speaker separation, timestamps, and segment summaries. - DoorDash, Observe.AI, and Amazon Web Services (AWS) scale AI-powered quality evaluation across 19,000 agents, automating nearly 100% of interaction reviews. - Samsung Electronics adds cloud transcription to its Voice Recorder app, giving users a choice between on-device privacy and cloud-powered accuracy. - Aina raises $5.5M to build hardware that controls AI agents rather than just recording, with its first product Dune already shipping to early adopters via Ivan Mehta for TechCrunch. - Chen Institute and Science honor neuroscientist Sergey Stavisky for an AI speech neuroprosthesis that decodes brain activity into spoken words at 97.5% accuracy. - New research finds that AI voice phishing works because of persuasive scripting, not vocal realism, as 70% of targets detect the synthetic voice but comply anyway via Sinisa Markovic for Help Net Security - instadesk, Huawei, and IFLYTEK open a joint AI customer experience lab in Uzbekistan, combining multilingual ASR/TTS with Ascend cloud infrastructure. - VoicePing 3.0 launches with real-time translation, AI dubbing, new ASR and MT models, and MCP/API access for enterprise multilingual workflows. - A study flags five risks in clinical AI scribes: inconsistent consent, weak performance on accented speech, background noise, missing human review, and unclear accountability via Resultsense. - Telcos are sitting on a voice AI opportunity bigger than their own cost savings, argues an analysis that says the real play is selling voice infrastructure to enterprises via Sebastian Barros. Image source: CX Today

💡 𝑊ℎ𝑎𝑡 𝑖𝑓 𝑦𝑜𝑢𝑟 𝐼𝐷𝐸 𝑐𝑜𝑢𝑙𝑑 𝑙𝑖𝑠𝑡𝑒𝑛, 𝑟𝑒𝑎𝑠𝑜𝑛, 𝑎𝑛𝑑 𝑐𝑜𝑑𝑒 𝑠𝑒𝑐𝑢𝑟𝑒𝑙𝑦 — 𝑎𝑙𝑙 𝑡ℎ𝑟𝑜𝑢𝑔ℎ 𝑦𝑜𝑢𝑟 𝑣𝑜𝑖𝑐𝑒? Hello Connections, This T𝐮𝐞𝐬𝐝𝐚𝐲, I had the amazing opportunity to participate in a 1-𝐝𝐚𝐲 𝐇𝐚𝐜𝐤𝐚𝐭𝐡𝐨𝐧 𝐨𝐫𝐠𝐚𝐧𝐢𝐳𝐞𝐝 𝐛𝐲 Google and AI Tinkerers, where our team built something really exciting — a 𝑽𝒐𝒊𝒄𝒆-𝒇𝒊𝒓𝒔𝒕 𝑺𝒆𝒄𝒖𝒓𝒆 𝑰𝑫𝑬 🎙️💻. 👉 The core idea: an IDE where your prim𝐚𝐫𝐲 𝐰𝐚𝐲 𝐭𝐨 𝐰𝐫𝐢𝐭𝐞 𝐚𝐧𝐝 𝐝𝐞𝐛𝐮𝐠 𝐜𝐨𝐝𝐞 𝐢𝐬 𝐲𝐨𝐮𝐫 𝐯𝐨𝐢𝐜𝐞. Imagine just pointing out the line of code with an error or speaking your thoughts directly into the IDE — that’s what we set out to build. But we didn’t stop there. While most teams focused on functionality, we doubled down on security, which became the heart of our project: 🔐 Key Highlights of 𝐨𝐮𝐫 𝐒𝐞𝐜𝐮𝐫𝐞 𝐈𝐃𝐄:  - 𝐕𝐨𝐢𝐜𝐞-𝐟𝐢𝐫𝐬𝐭 𝐜𝐨𝐝𝐢𝐧𝐠 using 𝐆𝐨𝐨𝐠𝐥𝐞 𝐒𝐩𝐞𝐞𝐜𝐡-𝐭𝐨-𝐓𝐞𝐱𝐭 𝐀𝐈.  - Security-first approach with 𝐌𝐞𝐭𝐚’𝐬 𝐋𝐥𝐚𝐦𝐚-𝐆𝐮𝐚𝐫𝐝 𝟒-𝟏𝟐𝐁 𝐨𝐧 𝐆𝐫𝐨𝐪 — automatically detecting & filtering 13 different harmful content categories (including text + images).  - 𝐆𝐨𝐨𝐠𝐥𝐞 𝐂𝐥𝐨𝐮𝐝 𝐃𝐋𝐏 (𝐃𝐞-𝐈𝐝𝐞𝐧𝐭𝐢𝐟𝐲 𝐀𝐏𝐈) to strip 𝐏𝐈𝐈 & 𝐏𝐇𝐈 data before processing.  - 𝐒𝐚𝐧𝐢𝐭𝐢𝐳𝐞𝐫 𝐀𝐠𝐞𝐧𝐭, which includes both our Guard and DLP to ensure only safe input passes through → then routed to our 𝐑𝐞𝐚𝐬𝐨𝐧𝐢𝐧𝐠 𝐀𝐠𝐞𝐧𝐭 for planning code changes, and finally → to the 𝐂𝐨𝐝𝐢𝐧𝐠 𝐀𝐠𝐞𝐧𝐭 powered by 𝐆𝐞𝐦𝐢𝐧𝐢 𝟐.𝟎 𝐅𝐥𝐚𝐬𝐡.  - 𝐇𝐮𝐦𝐚𝐧-𝐢𝐧-𝐭𝐡𝐞-𝐥𝐨𝐨𝐩 𝐬𝐚𝐟𝐞𝐠𝐮𝐚𝐫𝐝: critical actions (like deleting a row vs. a whole table!) require confirmation before execution.  - 𝐋𝐢𝐠𝐡𝐭𝐰𝐞𝐢𝐠𝐡𝐭 𝐔𝐈 built with 𝐇𝐓𝐌𝐋/𝐂𝐒𝐒/𝐉𝐒 + 𝐅𝐥𝐚𝐬𝐤 𝐖𝐞𝐛𝐒𝐨𝐜𝐤𝐞𝐭𝐬.  - 𝐓𝐞𝐦𝐩𝐨𝐫𝐚𝐫𝐲 𝐥𝐨𝐜𝐚𝐥 𝐬𝐭𝐨𝐫𝐚𝐠𝐞: All conversations & results are saved securely in a Markdown file for just 𝟐 𝐡𝐨𝐮𝐫𝐬 𝐛𝐲 𝐝𝐞𝐟𝐚𝐮𝐥𝐭, with user-controlled options. - Observability Dashboard 📊 to monitor API costs, tokens used per interaction, total requests, and average response time — keeping the IDE transparent, cost-aware, and efficient. 💡 We also explored 𝐂𝐥𝐨𝐮𝐝 𝐑𝐮𝐧 to seamlessly deploy our backend services and 𝐁𝐢𝐠𝐐𝐮𝐞𝐫𝐲 for securely logging/monitoring system usage (without storing sensitive data). These helped make the IDE scalable and auditable while staying privacy-first. Although we finished as 𝐑𝐮𝐧𝐧𝐞𝐫-𝐔𝐩(people’s choice)🥈, the experience was incredible, and I couldn’t have asked for a better team than Sahibpreet Singh and Bavalpreet Singh 🙌. This was just the beginning — we’re excited to push this concept further and turn it into a tool that reimagines how developers interact with code securely. 🔗 Check out our repo here: https://lnkd.in/eC_g9Sur Google Cloud Security Shopify Human Feedback Foundation Bitstrapped #AI #Google #Agents #ADK #IDE #Hackathon #Voice #Security #Machine #learning #genAI #LLM #LLAMA #Meta #Gemini

Touchscreens don’t belong in sterile rooms. Menus don’t help if you’re wearing gloves. In the real world of IoT - hospitals, warehouses, factories - your users can’t always swipe or tap. But they can speak. That’s why Voice User Interfaces (VUIs) are the next competitive edge in connected products. ✦ Voice-first UX solves real-world problems that screens can’t: 🔹 A nurse checking vitals hands-free in a crowded ward. 🔹 A technician adjusting smart machinery while holding tools. 🔹 Older adults calling for help without lifting a finger. For IoT products, voice is necessary for usability in motion-heavy, high-pressure, or accessibility-critical environments. ✦ Designing for VUIs means rethinking everything: ✦ You’re designing conversations, not screens. ✦ You’re optimizing intent recognition, not visuals. ✦ And without rigorous voice testing, you’re shipping adoption risk. The takeaway for 2026? If you’re building IoT without voice in mind, you’re already behind. Have you used a voice device that actually worked well? ♻️ Share this if you agree. 🔔 Follow Valentine Boyev for more updates!