The Three Strategic Dimensions of AI – When the Race to AGI Ended
Adam Olofsson Hammare
Summary: Three AI giants – XAI, OpenAI, and Perplexity – all released major updates within 48 hours in April 2026. But the race to AGI is over. The three companies are now running in entirely different strategic dimensions. XAI is trying to become the enterprise "one-stop shop" for voice AI, OpenAI is building foundational execution infrastructure, and Perplexity is aiming to dominate users' daily habits through a superapp. Listen to the podcast above and read details and explanations below!
The Paradox: $10 Billion on the Engine, Money Goes to the Pickaxe Seller
Imagine spending $10 billion to build the world's most powerful AI model. Then you look at the numbers and realize the one selling the safety gear is the one actually making the money.
That's exactly what's happening in the AI industry right now.
"The entire fixation on who has the highest benchmark score or the smartest raw model completely misses what's actually happening in the field."
The narrative of a shared race toward AGI – where everyone runs the same track toward the same finish line – simply isn't accurate. Instead, the three major players are diverging into entirely different strategic dimensions. And that's exactly what we're going to explore.
XAI: From Twitter Bot to Enterprise Platform
XAI's April 17–18 release wasn't about a new foundation model. Instead, they productized the entire voice pipeline.
They took Grok for speech-to-text (STT), text-to-speech (TTS), and packaged them as their own APIs. The pricing is aggressive – bordering on destructive for the industry:
- Speech to text: $0.10/hour (batch), $0.20/hour (real-time streaming)
- Text to speech: $4 per 1 million characters
- Voice agent API: $0.05 per minute
But there's a technical detail worth highlighting: Inverse Text Normalization (ITN). This is an algorithmic translation that makes raw transcription structured and machine-readable.
Imagine a doctor dictating: "Patient received 50 mg of medication at 2:15 PM on October 5th."
A regular transcription gives you the words "fifty milligrams." But with ITN, the system understands the semantic context and automatically formats: 50 mg | 14:15 | 10-05. It opens a bridge between chaotic human acoustics and structured database.
Why does this matter for enterprises?
XAI packaged all of this with SOC 2 Type 2, HIPAA (healthcare compliance), and GDPR. The strategy is crystal clear: imagine a healthcare contact center that can now, legally and practically, pump thousands of hours of patient calls directly to XAI – for $0.10 per hour.
OpenAI: Back to the Building Blocks
If XAI is out selling the finished product, OpenAI is busy building the foundation itself.
Their April 18 updates were technically deep: Python Agents SDK version 0.14.2 and a pre-release of Codex 0.122.0. The focus was on two things:
MongoDB sessions and a new "seatbelt" function.
Why care about a specific database architecture? Here's the critical insight: we're shifting from stateless chatbots to stateful, autonomous agents.
A regular chatbot today: you ask, it answers, the transaction is done. It forgets you immediately. But an agent working on a complex problem for hours or days needs persistent, structured memory – and MongoDB is perfect for storing such constantly evolving, complex nested data.
"You can't have autonomous agents if you can't show exactly how they made their decisions."
The new seatbelt function – sandboxed execution – solves another problem. If an agent that's been running for three days makes a catastrophic error, you need to be able to rewind and see exactly where it went wrong. With absolute isolation, you can inspect exactly which regulation caused the problem. It gives you a complete audit trail.
Perplexity: Capturing Habits Instead of Models
Perplexity completely ignores the model arms race. Instead, they're focusing on something else: getting you to never leave their app.
Their latest move? Building out their product with discover feed, travel booking, and team workspaces. The goal is to dominate users' everyday behaviors – not by having the smartest model, but by offering the best workplace.
Three Strategies, Three Dimensions
Three companies. Three entirely different paths. Here's how they're positioning themselves:
XAI
- Strategy: "One-stop shop"
- Focus: Selling the complete sensorium funnel system directly to enterprises
OpenAI
- Strategy: Foundational architecture
- Focus: Building execution infrastructure so others can build
Perplexity
- Strategy: Habit capture
- Focus: Building super-app to keep the user in the app
"If Perplexity can succeed and retain power users by taking someone else's model – like GPT-4.1 – and putting it in a better workplace and faster mobile app. Will billions of dollars that OpenAI and XAI spend on training these massive foundation models ultimately just end up manufacturing for those who build the user interfaces?"
Thoughts on how this affects the future
The question hanging in the air after this overview is uncomfortable: will foundation models eventually become just electricity?
If you spend $10 billion building a powerful model, but the final profit ends up with whoever builds the sleekest user interface – was the investment worth it?
For companies thinking about their AI strategy today: the choice of platform is no longer about who has the prettiest benchmark number. It's about:
- Do you want a ready-made, packaged solution with everything under one roof? XAI.
- Do you want to build your own agent systems with full control? OpenAI.
- Do you want the smoothest experience for your team? Perplexity.
And for developers: the big value proposition right now isn't in building a new model. It's in understanding packaging, developer ergonomics, and user behavior.
It's a much more complex reality than "who wins AGI." And that reality requires a completely different type of strategic thinking.
