AI Needs a Reset: It's Time to Get Back to Engineering Basics

AI doesn’t need another breakthrough model. It needs a return to engineering fundamentals.

Somewhere between foundation models and agentic AI, we confused capability with architecture.

Because a model can reason doesn’t mean every task requires reasoning. Because a framework can orchestrate ten agents doesn’t mean every workflow needs ten agents. Because a model supports a million-token context window doesn’t mean every prompt should use it.

Somewhere along the way, we stopped asking one of the most important engineering questions:

“What is the simplest solution that solves the problem?”

Instead, our conversations increasingly revolve around different questions. Which model should we use? How many agents do we need? Should we increase the context window? Can we add another reasoning step?

The answer increasingly seems to be more—more models, more agents, more tokens, more orchestration, and more compute.

Yet great engineering has never been about doing more. It has always been about delivering the desired outcome with the simplest architecture, the least complexity, and the fewest resources necessary. Those principles built modern software, and they remain just as relevant in the age of AI.

Progress Isn’t the Problem

The pace of AI innovation over the past few years has been extraordinary. Better reasoning models, multimodal capabilities, autonomous agents, and longer context windows are enabling applications that were unimaginable just a short time ago. This progress should be celebrated.

But every technology revolution eventually reaches an inflection point where innovation alone is no longer the differentiator. The conversation shifts from what is possible to what is practical. Success is no longer measured by demonstrating capability, but by delivering sustainable business value.

AI is beginning that transition.

Capability Should Expand Our Choices, Not Replace Our Judgment

One of the biggest traps in technology is allowing available capabilities to dictate architecture rather than business requirements.

We’ve seen this pattern before. Cloud computing made infrastructure virtually unlimited, and many organizations responded by overprovisioning resources. Microservices gave us unprecedented architectural flexibility, and many teams responded by over-decomposing systems into services that became increasingly difficult to manage.

Today, AI offers reasoning models, autonomous agents, massive context windows, and sophisticated orchestration frameworks. They are remarkable capabilities, but there is a risk that we begin using them because they exist rather than because they are necessary.

Not every workflow needs autonomous agents. Not every task benefits from a reasoning model. Not every application requires a million-token context window. In many cases, a deterministic workflow, a smaller open-weight model, or even traditional software may be the better engineering solution.

The best architects don’t start with technology. They start with the problem. Technology is simply one of many tools available to solve it.

Complexity Must Be Earned

Modern AI systems can be extraordinarily sophisticated. Multi-agent orchestration, dynamic model routing, reasoning pipelines, and tool-using agents all have legitimate use cases. They represent important advances in how intelligent systems are built.

However, complexity often accumulates faster than business value.

Every additional model call, orchestration layer, or system dependency introduces trade-offs: higher latency, increased infrastructure costs, greater operational overhead, additional failure points, and more difficult debugging and maintenance.

Complexity itself isn’t the problem. Unnecessary complexity is.

Engineering has never been about avoiding complexity altogether. It has always been about ensuring that every layer of complexity contributes proportionate value. Complexity should be an intentional architectural decision—not the default consequence of adopting the latest AI capability.

The best architectures don’t maximize AI capability; they optimize business outcomes.

Back to Engineering Basics

Large language models have changed how we build software, but they haven’t changed the principles of good engineering:

• Before increasing context, ask whether better retrieval can solve the problem.
• Before adding another inference, ask whether the result can be cached or reused.
• Before orchestrating multiple agents, ask whether a simpler workflow would achieve the same outcome.
• Before reaching for the largest available model, ask whether a smaller one is sufficient.
• Before introducing AI at all, ask whether the problem is better solved through deterministic software.

These aren’t constraints on innovation. They are the disciplines that make innovation sustainable.

The goal of engineering has never been to use the most sophisticated technology available. It has always been to apply the right technology, in the right place, for the right reason.

From the Reel Economy to the Real Economy

Over the past two years, much of the AI conversation has been shaped by what I think of as the “reel economy.” Social media, conference demos, benchmark leaderboards, and increasingly impressive prototypes have inspired millions of engineers and accelerated innovation. They have shown us what AI could do.

Now the industry faces a different challenge: proving what AI should do.

Businesses don’t operate on viral demonstrations. They operate on systems that are reliable, secure, maintainable, scalable, compliant, and economically viable. They care less about how many agents collaborate in a workflow and more about whether that workflow consistently delivers value.

The next chapter of AI won’t be defined by who builds the most impressive demo. It will be defined by who delivers the most meaningful outcomes in the real economy.

The organizations that lead this next phase won’t necessarily have the largest models or the most sophisticated architectures. They’ll be the ones that combine AI innovation with engineering discipline.

Innovation creates possibility. Engineering creates impact.

Perhaps that’s the reset AI really needs—not a reset in innovation, but a reset in priorities.

As AI matures, success won’t be measured by how much intelligence we can build into a system. It will be measured by how effectively we apply that intelligence to solve real-world problems.

It’s time to move beyond building AI for the reel economy and start building AI for the real economy.