Earlier this month, Secondmind CEO Gary Brotman launched a new Substack, G on AI, aimed at finding the "real signals in the AI noise." His inaugural post, The Efficiency Reckoning, argues that the “Magician” era of AI - one defined by brute force and infinite data - is over, and the era of hard engineering has arrived. 

At Secondmind, we aren’t just watching these signals; we are building the software that responds to them. Below we look at three industry shifts identified by Gary and discuss how Secondmind is providing practical solutions for the world’s leading engineering teams.

Avoiding the virtualization trap

Gary identifies a "mathematically bankrupt" trend in the shift toward virtual development: the idea that we can solve complexity simply by scaling digital mileage. As he notes, “If you swap a billion physical miles for a billion virtual miles… you’ve just traded a gasoline bill for an AWS bill”.

At Secondmind, we see this challenge firsthand. Complex system design can require hundreds or even thousands of high-fidelity simulations. In this context, the "Shift Left" concept - intended to save time and money by validating earlier with increased virtualization - simply creates a new drain on both budgets and project timelines. This is the virtualization trap in practice: trading physical bottlenecks for digital ones.

We address this with data-efficient software that requires up to 80% fewer data points - meaning significantly fewer simulations or tests - to achieve even better results. For example, when a leading global Tier 1 supplier struggled to optimize piston bowl designs, Secondmind helped them discover 3 times the number of feasible designs in half the time compared to traditional CAE tools. By focusing only on the data that matters, we help engineers avoid the virtualization trap and move toward rapid, high-value discovery.

Math per watt in practice

The Silicon Valley dogma of "Scale is All You Need" is reaching a breaking point, evidenced by the staggering energy requirements of modern AI data centers. Gary argues that requiring a fission reactor to run matrix multiplications is “a damning indictment of efficiency... the antithesis of the engineering mindset.”

In the automotive sector, this isn't just an environmental concern; it’s an operational bottleneck. Running physical test cells at $1,500 an hour for weeks on end is the engineering equivalent of that fission reactor. Secondmind builds software that respects these hard constraints by maximizing “math per watt.” 

Instead of adding more data and more testbed hours, we focus on extracting maximum insight from every minute of hardware time. The results are proven: working with a Tier 1 motor supplier, Secondmind delivered 80% data savings on motor measurements, which the customer agreed would save 85% of their total calibration time. This shift is allowing engineers to manage the exponentially growing complexity of sophisticated systems without an exponential growth in Capex or energy consumption.

Secondmind’s closed-loop approach

For high-stakes engineering, "good enough" is an unacceptable standard. Gary highlights how Generative AI often acts as a “parlor trick” that produces plausible designs while ignoring the laws of physics. He asserts: “Physics is the ultimate ground truth. You cannot cheat gravity, and you cannot cheat chemistry.”

Secondmind is built for this reality. Our software is designed for reasoning and trust in domains like design and calibration, where data is scarce and physics matters. We achieve this through a "Closed Loop" approach - or as Gary says, "AI proposes, physics disposes."

Practically, this means our software acts as a navigator, identifying pre-validated, physics-aware design choices. The real-time feedback loop ensures the system never suggests a design that violates fundamental laws and helps keep complex programs on schedule and experienced engineers in control.

The takeaway

While the broader AI industry continues to chase scale at any cost, at Secondmind we are building practical AI for the hard engineering required to make these systems work in the real world. In 2026, the real competitive advantage will be the ability to achieve high-confidence results and resolve complex trade-offs with radical efficiency.

At Secondmind, we’re helping the world's leading OEMs and Tier 1 suppliers turn efficiency into a competitive edge. We'd love to show you how you can deliver complex projects faster.

You can read Gary’s full blog at G on AI on Substack, and subscribe for future posts.