The AI Race Is About Distribution, Not Benchmarks
The conventional narrative around AI leadership focuses on benchmark results: faster inference, lower perplexity, better instruction following. But here’s the reality:
You don’t win the front page of the internet by winning benchmark tests. You win by owning user behavior.
OpenAI understood this when they launched ChatGPT. It wasn’t the most complex application—but it was the right interface at the right time. A simple chat window unlocked an entirely new paradigm for interacting with information. For the first time in 20 years, Google’s dominance over the front door to the internet felt... shakier.
Where the AI Race Is Actually Headed
Foundational model capabilities will continue to matter—but what’s going to really matter is how quickly those capabilities get turned into things people want to use. And that means we’re entering the era of interface velocity.
The next breakout experiences will likely come from companies that ship fast, experiment faster, and bridge the gap between new model capabilities and daily life. Here’s where that frontier is unfolding:
1. Real-Time, Multimodal Communication
We're on the edge of AI that can:
Hold voice conversations in real time
Understand and react to what's happening on your screen
Screenshare, co-browse, or narrate as you interact
Combine voice, vision, and touch into one fluid conversation
Early demos (like OpenAI’s GPT-4o) hint at what’s coming—but the race is on to make this usable, delightful, and ubiquitous. Whoever gets this right will redefine how we interact with devices and each other.
2. Agent-to-Agent Social Coordination
Soon, your AI won’t just help you—it’ll talk to other AIs that belong to your friends, coworkers, or family. Imagine:
Coordinating hangouts or work sessions between AI assistants
Planning trips, syncing calendars, suggesting mutual times
Sharing relevant data across trusted AI networks
This creates a layer of persistent social presence—without constant human micromanagement. And whoever builds the best protocol and UX for it wins.
3. Cross-Device, Context-Rich Experiences
True ambient computing means:
Talking to your headphones
Seeing the result on your phone
Getting follow-ups on your TV
Picking up where you left off on your car dashboard
Multimodality isn’t just input/output—it’s context continuity across devices. That requires deep coordination of hardware, software, and AI agents. Right now, no one has nailed this yet.
4. Deep Integration with User Accounts & Services
Today’s AI “agents” often feel like disconnected toys. Connecting them to your real accounts is clunky and slow.
The future? AI that can:
Take actions inside your calendar, email, files, apps
Do so safely, reliably, and with minimal setup friction
Remember what you’ve done across services
The battle here is twofold: building trust and minimizing friction. The first to get this right will become the new “command center” for how people get things done online. Things like ChatGPT’s Connectors and MCP offer a window into the future here, but it’s still early days.
Why Real-Time Matters—And Why It’s So Hard
Here’s the kicker: most of these AI-powered experiences require real-time interaction. That’s where the existing developer toolchain breaks down.
Real-time audio, in particular, is tough. The cloud tools out there today—LiveKit, Agora, Twilio, WebRTC—handle the pipes, but they don’t manage the complexity on your device. You still need to hand-roll things like:
Audio graphs
Speech detection
Noise cancellation
Voice changer integration
Synchronization across media streams
Low-latency DSP and hybrid workflows
That’s where Switchboard comes in.
Switchboard: Turbocharging Real-Time AI Experiences
Switchboard makes it dramatically easier to build real-time, on-device and hybrid audio experiences—the kind of experiences next-gen AI apps need.
Cross-platform audio graph abstraction
Plug-and-play modules for VAD, STT, TTS, filters, effects, and more
Real-time streaming that’s compatible with popular cloud solutions
Built for speed, experimentation, and iteration
Whether you’re building a voice assistant, a social audio app, or a new kind of screenless AI interface, Switchboard reduces time-to-market by 10x.
We’re already partnering with teams pushing boundaries in this space, and through our Venture Studio, we also continue to experiment directly with what comes next.
The Future Belongs to the Fast
The AI race is shifting from compute power to consumer resonance.
From benchmark wins to behavioral change.
From centralized dominance to distributed real-time UX.
Whoever ships the most beloved products—fast—will own the next front page of the internet.
We have the tools to make that happen. Let’s build amazing new experiences together.