“The model is the brain. But a brain without a nervous system is just a very expensive paperweight.”
For three years, the AI industry has been obsessed with one question:
Which model is best?
This week, that question became irrelevant.
Not because models stopped mattering. But because something bigger happened.
Three frontier models launched in the same week — Claude Opus 4.6, GPT-5.3-Codex, and Gemini 3. Each one extraordinary. Each one pushing the frontier.
And yet, none of them is the story.
The story is MCP.
Model Context Protocol. The universal standard that connects any AI model to any real-world system. Built by Anthropic. Adopted by OpenAI. Adopted by Microsoft. Embraced by Google. Now governed by the Linux Foundation’s Agentic AI Foundation.
One protocol. Every lab. Every agent.
If you’ve been following my thesis on the post-search paradigm, this is the moment it goes from theory to infrastructure.
The model race is over. The system race just started.
Here’s what happened in one week:
Anthropic released Claude Opus 4.6 — their most advanced reasoning model. The Claude 4.5 family now spans Opus, Sonnet, and Haiku.
OpenAI launched GPT-5.3-Codex — 25% faster agentic coding, state-of-the-art on SWE-Bench Pro. Over a million developers on Codex. OpenAI used it to train itself.
Google shipped Gemini 3 with a CLI tool for building “AI employees” from your terminal. 750 million users on Gemini products. 48% Cloud growth.
Three frontier models. One week. Capabilities converging.
But the real signal?
MCP became universal. Anthropic built it as an open protocol. This week, every major lab either adopted it or committed to building for it. The Linux Foundation took governance. It’s no longer Anthropic’s protocol. It’s the protocol.
Think of MCP as USB-C for AI.
Before USB-C, every phone had a different charger. Wasteful. Frustrating. Fragmented. USB-C unified everything.
MCP does the same for AI agents. One standard that lets any model connect to any tool, any database, any API. The brain-and-nervous-system framework I’ve been writing about — this is it becoming real.
Why this matters for leaders: Models are commoditizing. The gap between Claude, GPT, and Gemini is narrowing every quarter. But an AI agent that can act in the real world — check your inventory, process a return, adjust pricing in real-time — that needs connective tissue.
MCP is that connective tissue.
The model you choose matters less every quarter. The system you build around it matters more.
Perplexity quietly launched something profound this week: Model Council.
It runs Claude, GPT-5.2, and Gemini — simultaneously. In parallel. On every query.
Then it cross-validates the answers. Shows where models agree. Flags where they disagree. Delivers a unified, verified response.
This is the death of single-model trust.
Think about what this means for commerce:
A customer asks which laptop is best for video editing under ₹80,000. Instead of one model guessing — three models reason, compare notes, and deliver a stress-tested answer.
Then an agent acts on it.
The search becomes the sale. The question becomes the transaction.
For decision-makers: If you’re still evaluating “which model to use,” you’re asking yesterday’s question. The future is multi-model systems where AI committees deliberate, not individual models that guess.
Here’s a number that should reshape your AI strategy:
$650 billion.
That’s the combined AI infrastructure commitment from the top hyperscalers for 2026.
Alphabet alone: $185 billion. More than their past three years combined. Amazon: up to $200 billion. Microsoft, Meta — each pushing records.
But here’s what most leaders miss:
This spending isn’t on models. It’s on everything else.
Memory bandwidth. Power capacity. Chip supply chains. Data center cooling. Samsung is shipping HBM4 samples — next-generation memory that determines which companies can actually run frontier models at scale.
The mental model shift:
2023-2024: The company with the best model wins. 2025-2026: The company with the best infrastructure wins. 2027+: The company with the best system — model + infrastructure + connective tissue — wins.
We’re in the middle transition. Most companies are still optimizing for the first era while the world moves to the third.
Burn the blueprint. The old playbook was: pick the best model, build a chatbot, call it AI strategy. The new playbook: build a system — agents, protocols, infrastructure, memory, coordination — and let the model be interchangeable.
Two stories that didn’t make headlines but should have:
1. Claude Sonnet 5 “Fennec” is imminent. It leaked in Google’s Vertex AI logs. Rumored to be 50% cheaper than Opus 4.5 with comparable performance. The “better AND cheaper” trend in AI is the most underrated story of the decade. Every quarter, the frontier gets cheaper. Every quarter, the barrier to entry drops.
For every startup founder in India reading this: the cost curve is your friend. What cost $10,000 to run last year will cost $1,000 next year. Build now.
2. OpenAI is putting ads in ChatGPT. Sam Altman once called AI combined with advertising “uniquely unsettling.” Now it’s happening — ads for free and lower-tier users.
When the product is free, you’re not the customer. You’re the product.
This is worth watching carefully. The companies that resist the ad model — that charge directly for value — will build deeper trust with users. And in the age of agents, trust is everything.
MCP going universal is especially significant for India’s AI ecosystem.
Here’s why:
Indian companies don’t need to build frontier models. Nobody does — that race has three runners and they’re all spending $100B+ a year. What Indian companies can build: the systems, agents, and applications that sit on top of those models.
MCP levels the playing field. A startup in Bengaluru can build an agent using Claude, GPT, or Gemini — and connect it to Indian commerce infrastructure — using the same universal protocol as a team in San Francisco.
The emerging market mirror is real: the most innovative AI commerce applications in 2027 won’t come from Silicon Valley. They’ll come from markets where necessity and low legacy create the perfect conditions for AI-native innovation.
India’s moment in AI isn’t about building the next GPT. It’s about building what sits on top of it.
And this week, the foundation for that was laid.
Step back from the individual stories and a pattern emerges:
Models are converging. Claude, GPT, Gemini — each extraordinary, increasingly similar. The differentiation window is shrinking.
Infrastructure is diverging. Who can actually deploy at scale? Who has the memory, the power, the chips? That’s where the real competition is.
Protocols are unifying. MCP as universal standard means the connective tissue is commoditizing too. Good. That’s how platforms get built.
Systems are the moat. Not the model. Not the infrastructure alone. The system — how your models, tools, data, and agents work together. That’s what compounds. That’s what no one can copy.
This is what I mean by the shift from channels to context. The right model, connected to the right systems, delivering the right outcome at the right moment.
Not just smart. Connected.
Welcome to the week AI grew a nervous system.
Choose to be wise.
Suman Guha Founder, recodeai
recodeai is free for 100,000 leaders. If you found this useful, forward it to one person who needs sharper signals in the AI noise.
Donations support scholarships for girl children — education, health, dignity.
No posts
Comments
Nothing yet. Say the first thing.
Sign in to join the conversation.