GPT-5 just launched, and it’s fascinating to watch the predictable response cycle unfold. The tech community is buzzing with benchmark comparisons, demo videos, and thoughtful debates about capabilities versus competitors.

I personally think that GPT-5 isn’t really about taking a revolutionary step towards AGI, but all about lowering the inference costs for OpenAI.

Anyway, what is more important is that these discussions are about important questions, but those questions are likely the wrong ones to ask!

The Integration Imperative

In a recent post on Substack, MIT’s Andrew McAfee makes a compelling argument that deserves more attention amid the GPT-5 hype: The real competitive advantage in AI doesn’t come from having access to the best model — after all, most leading models are available to anyone willing to pay. The true differentiator is systematic integration. And as Andrew points out this isn’t a new phenomenon in technology adoption.

Lessons from the Electric Revolution

Consider the shift from steam to electric power in manufacturing — a transformation that offers striking parallels to our current AI moment.

When electricity first became viable for industrial use in the late 1800s, factory owners did what seemed logical: they replaced their massive steam engines with electric motors. But they changed nothing else. The factories still relied on the same complex systems of belts, pulleys, and drive shafts that had characterized the steam era. The same rigid, multi-story buildings. The same workflows designed around a single, central power source.

The results were predictably underwhelming. Yes, electric motors were somewhat more reliable than steam engines, but the productivity gains were marginal. Many industrialists wondered if the switch had been worth the investment.

Then came the real innovators. They realized that electricity’s true power wasn’t in replicating steam — it was in enabling entirely new possibilities. Why have one massive motor when you could have dozens of smaller ones? Why constrain your factory to the layout demanded by drive shafts when each machine could have its own power source?

This insight unlocked a manufacturing revolution. Assembly lines became possible. Single-story factories with optimized workflows replaced cramped multi-story buildings. Machines could be arranged for efficiency rather than proximity to power. The companies that grasped this didn’t just improve their existing operations — they invented entirely new ways of making things that their steam-bound competitors couldn’t match.

The AI Parallel

Today’s AI adoption follows an eerily similar pattern. Walk into most organizations experimenting with AI, and you’ll find them doing the digital equivalent of swapping steam for electric motors.

They’re using ChatGPT to write emails faster. Claude to summarize meetings more efficiently or to generate reports that used to take hours. These are useful applications, but they’re fundamentally about doing the same things slightly better.

It’s the equivalent of those early factory owners who saw electricity as just a more reliable steam engine.

Beyond Incremental Improvement

The organizations that will define the next decade aren’t asking “How can AI make our current processes 20% more efficient?” They’re asking fundamentally different questions:

What workflows become possible when every employee has an AI collaborator?
How do organizational structures change when information synthesis happens at machine speed?
What new products and services can we create that were impossible before?
How do customer relationships transform when personalization can happen at scale?

These companies are redesigning their operations around AI’s unique capabilities rather than retrofitting AI into existing processes. They’re creating competitive moats not through access to technology — everyone has that — but through novel applications of it.

DeepSeek vs. GPT-x — the Implementation Gap

The contrast becomes even starker when you look at how different regions respond to new AI capabilities. When DeepSeek launched earlier this year, the Western tech community’s response was the same as today with GPT-5: benchmark comparisons, performance metrics, debates about whether it could match GPT-4’s reasoning capabilities.

Meanwhile, in China, the conversation was fundamentally different. While Western observers were arguing about MMLU scores and coding benchmarks, Chinese entrepreneurs and companies were already mapping out implementation strategies. Manufacturing firms were exploring quality control applications. Financial services companies were prototyping risk assessment tools. Healthcare organizations were designing diagnostic assistance systems.

The difference is telling. One culture treats AI models as technological artifacts to be analyzed and compared. The other treats them as tools to be deployed and integrated. One asks “How good is it?” The other asks “What can we build with it?”

This isn’t about model superiority — it’s about mindset. China’s rapid AI adoption across industries isn’t happening because they have better models. It’s happening because they’re focused on implementation over evaluation and on building over benchmarking

The Real GPT-5 Opportunity

This brings us back to GPT-5. Yes, it’s important to understand its capabilities. Yes, the benchmarks matter. Yes, we should experiment with what it can do.

But the companies that will extract the most value from GPT-5 — or any AI model — won’t be the ones with the earliest access or the biggest compute budgets. They’ll be the ones who most thoroughly rethink their operations around what AI makes possible.

They’ll be the ones who recognize that the question isn’t “Is GPT-5 better than GPT-4?” but rather “What can we build with GPT-5 that we couldn’t even imagine before?”

A Playbook for Transformation

The challenge, of course, is moving from theory to practice. How do organizations actually make this shift from incremental improvement to transformational integration?

Andrew McAfee’s post offers a practical 8-step playbook for organizations ready to move beyond the hype cycle. His framework provides a systematic approach to AI integration that focuses on fundamental transformation rather than surface-level improvements.

So, as we watch the bumpy GPT-5 rollout unfold, it’s worth remembering that the most important developments won’t be in the model’s capabilities — they’ll be in what creative organizations choose to build with those capabilities.

The real question isn’t whether GPT-5 is revolutionary. It’s whether we are…

For a detailed framework on systematic AI integration, I highly recommend Andrew McAfee’s post on his Substack channel Geekway.

And if you want to explore how to re-think digital beyond AI, join our workshop series Re-Thinking Digital with Insights from Disruptive China (German version).

From Steam Engines to AI: The 150-Year-Old Lesson We Need to Remember for GPT-5.