Thoughts from a No-Code CAIO: Is the Sun Setting on Superintelligence?

Jason Padgett
Nov 14
5 min read

The Aurora Borealis - One of those sights that makes you look up, breathe, and say… yeah, things are changing.

I watched the aurora last week. Those rippling bands of light, each doing its part, no single beam trying to illuminate the whole sky. It reminded me of a prediction I've been sitting with—one that Microsoft's Mustafa Suleyman and the practitioners at the Midwest AI Summit seem to be confirming.

The AI world has been chasing superintelligence. But what if that chase is ending?

The Superintelligence Era (2017–2024)

The "Attention Is All You Need" paper dropped in 2017 and introduced the Transformer architecture that revolutionized natural language processing. From November 30, 2022, when

OpenAI introduced ChatGPT to the world, until very recently, the playbook was simple: bigger models, more GPUs, more compute. That vision alone was enough to raise billions in VC dollars.

As we approach ChatGPT's third birthday, a different question is surfacing: What if superintelligence looks less like a super-brain and more like swarms of specialized agents?

What People Actually Want

When you talk with people who actually run companies: the manufacturers, life sciences teams, accountants, hospital systems, and SMB leaders across industries—you hear pragmatic requests. They don't want a box that can solve "Humanity's Last Exam" style puzzles. They want tools that work on their data, according to their goals and values.

Real value shows up when AI understands the domain.

It's CPAs shaping their own reconciliation agent. It's clinicians designing decision trees the model simply executes. It's a logistics director running a secure planning agent on a local server because the data can't leave the building. If anything, superintelligence may get replaced by super-relevance.

The Rise of Small, Useful Intelligence

We're watching the era of "one model to rule them all" quietly fade while something far more practical takes its place. Smaller language models. Tight clusters. On-prem agents. Systems that fit inside a rethought workflow rather trying to replace it.

Mustafa Suleyman talked openly about Microsoft staying "three to six months behind the frontier" so they can publish lightweight, affordable models that run on your laptop instead of on a warehouse of GPUs. That's a radical strategic signal from one of the biggest AI players on earth → prioritizing deployability over raw capability.

People don't need omniscience. They need specificity and operability.

Ask any leader around here what they want from AI and you'll hear some version of the same request: "Make my back office faster." "Fix the overtime problem." "Give me natural language analytics." "Just help my people stop jumping between twelve different apps."

All of which involves context engineering, process mapping, and rethinking how data flows through the organization.

This is why smaller models may start making a big splash. They sit behind the firewall, grab only what they need, and execute in real time. They also enable something that large frontier models struggle with: private agency. The AI doesn't think for you, it acts for you. It runs queries, drafts reports, updates tickets, and sends alerts. That's where the ROI actually comes from.

The Swarm Future

And then there's Palmer Luckey and the Anduril team. If you want to glimpse the future of small models and agents, watch what they're doing with drone swarms. It's the same architecture we're moving toward in the enterprise: a network of small, fast, specialized units that coordinate without a single giant brain in the middle.

Imagine dozens of tiny AI agents inside a company, each with one job, all talking to each other. A dispatcher agent. A staffing agent. A documentation agent. A billing accuracy agent. A compliance guardrail agent. Not superintelligence. Super-coordination. Super-speed.

What Preparing for This Looks Like in Practice

At Phoenix Paramedic Solutions, we're not building swarms (yet). We're doing something harder: building the foundation that makes the swarm possible.

Right now, I'm deep in learning about data governance. Learning to think like a schema-free Chief Data Officer. Before you can deploy agents that coordinate across systems, you need to know what data you actually have, where it lives, how clean it is, and who owns it.

We're asking fundamental questions about what our stack looks like going forward:

Do we go Copilot-enabled and Microsoft-centric—betting on integration with predefined tools our teams already use but at the mercy of the market?
Do we reorganize around data tool like Snowflake and select best-of-breed tools carefully, prioritizing flexibility and interoperability?
Do we shift to a platform like Databricks and start focusing on in-house education and talent acquisition that lets us build on open source?

These questions used to be additive and reside primarily in the IT domain but going forward they're strategic bets about what kind of AI organization we want to become.

The swarm future that may be on the near horizon: It requires data infrastructure that doesn't exist yet in most small and mid-sized organizations. It requires governance frameworks. It requires people who understand both the clinical mission and the technical architecture.

So that's what we're focusing on first is not the 'half-baked' agents and questionable MCP servers it is... yes, the boring, essential groundwork: mapping data flows, aligning schemas, establishing access controls, and figuring out which platforms give us the most leverage with the least vendor lock-in.

The agents will come. But they'll only work if the foundation is solid.

The One Model That Might Rule Them All

If there's a "one model to rule them all" in this future, I don't think it's going to be a reasoning engine. It's more likely to be a data orchestration mode → something that can take a lake full of unstructured information and organize it, semantically layer it, define relationships, and make it queryable in real time.

That's the middleware. The coordinator. The thing that lets all the small agents talk to each other without rebuilding integrations every time you add a new tool.

Frontier models will still matter for complex reasoning tasks→ when you need deep multi-hop thinking or breakthrough synthesis. But for execution, coordination, and speed? Small models win every time.

Counterpoints Worth Considering

Hybrid is currently winning. In the GPT-5.1 era, hybrid intelligence means intelligently combining GPT-5.1’s reasoning ability with smaller, specialized, or deterministic models to optimize speed, cost, and compliance for each task. Rather than one model orchestrating a swarm, GPT-5.1 works as part of a dynamic ecosystem, with systems like Teneo routing queries to the right model or workflow component automatically based on complexity. Frontier models like GPT-5.1 now act as powerful “building blocks” within multi-model, future-proof architectures rather than as monolithic dependencies.

Breakthrough risk is real. A single architecture leap→ like we saw with test-time scaling or Alpha Go's famous move 37 could recentralize power overnight. The small-model bet assumes continuous, predictable progress. Disruption doesn't work that way.

SLM limits persist. Multi-hop reasoning and hallucination remain issues. Mitigation means ensemble approaches and keeping humans in the loop for critical decisions.

None of these invalidate the shift. They just remind us that both worlds coexist—and the smartest organizations will use both.

First Step If You're Reading This

Run a quick audit: List your five most painful workflows. Tag every data source they touch. Then ask: "Which of these could a small model update in real time if the schema were clean?"

You'll surface your first agent→ and the governance gap blocking it.

So Is the Sun Setting on Superintelligence?

I would not go as far at to predict the sun is setting. I think the light is shifting. Frontier AI is likely keep pushing boundaries. But the center of gravity in real organizations may shift quickly in 2026 toward smaller, safer, cheaper systems that deliver actual results.

And as much as it pains me to have to address my deficiencies in understanding data analytics, I love a good challenge and the Age of Intelligence appears to hold unlimited potential.