It’s the humans, stupid

During the 19th-century debate about evolution, Lamarck proposed the inheritance of characteristics acquired during a lifetime, Cuvier stood for catastrophism and fixed species, and Darwin suggested natural selection, where neither the strongest nor the smartest survives, but rather those most adaptable to changes in the ecosystem. Conversely, rapid shifts in an ecosystem can undermine or completely decimate the dominant species of the day, while elevating groups (including, but not necessarily, the underdogs) with a favorable combination of traits. For what it’s worth, dinosaurs are still with us, stealing breadcrumbs at every opportunity.

The final leg of modern humans’ climb to the top of the food chain on Planet Earth was about defeating Neanderthals. Neanderthals had a brain of comparable capacity, which (unfortunately for them) prioritized individual spatial performance over social interaction. This, coupled with a peculiar anatomical disadvantage, eventually left them numerically outmatched.

Looking out the window, a similar change is upon us, except this time it will be humans wielding artificial intelligence tools pushing whosoever hath not to the fringes. The commoditization of fast problem-solving that is available 24/7, combined with polyglot persistence and convenient data-exchange frameworks, provides an unprecedented level of autonomy. This enables individuals, small teams, or less powerful countries to compete successfully against much larger entities. Naturally, large organizations are also adopting artificial intelligence, but their accumulated weight slows them down, and their value proposition is no longer unique; this is similar to the end of the Bronze Age sparked by the discovery of ubiquitous iron, or the decline of medieval knights triggered by the introduction of the longbow.

In the engineering and manufacturing domain, major players like Dassault Systèmes and Siemens face mounting headwinds because alternatives to exquisite geometry solvers are coming quickly to everyone’s fingertips, and integration between disparate, smaller, and more agile systems has become cheap, easy, and reliable. Moreover, connecting external agents directly to CAD, CAE, and PLM modules cuts straight into their licensing base.

CATIA and NX have always been about expert engineers clicking through them 40 hours a week. They may no longer justify their price point when a combination of frontier LLMs and open-source CAD systems delivers increasingly comparable features, superior UI/UX, and robust APIs for pennies – and round the clock. Inserting artificial intelligence inside the major players’ ecosystems still leaves them on the defensive against agent-first newcomers. Incumbent players will hardly collapse outright; they will simply lose their primacy and become a replaceable layer for agentic collaboration.

There is enormous excitement around adopting AI, but there are equal dangers. The amazingly powerful, inexpensive toolset provided by frontier labs is “autonomy on lease”- a singular bet on the goodwill of a handful of corporations whose political opinions, regulatory pressures, and pricing models can change overnight. A power outage can slow down business operations; a company deeply entangled with Claude or ChatGPT will turn into a headless chicken in an instant. Those presently complaining about Dassault Systèmes’ vendor lock-in habits may end up remembering them warmly.

Individual humans are stochastically fragile and inherently slow. In contrast, agents can be dangerously fast. Hence, humans are transitioning into carriers of “agency” and sign-off sentries at decision-making checkpoints. The new paradigm is to fence agents within explicit constraints, bound them with strict rules, and pit them against each other for validation and verification. Security and traceability become a gargantuan, high-speed arms race: agents rush to do their jobs using whatever identity they can lay their hands on, while the entire corporate IT infrastructure morphs into an active attack surface for rogue or malicious internal and external actors. The quantum revolution is allegedly around the corner too, ready to turn encryption-backed confidence into mush.

As often happens with humans, what the left hand unleashes, the right hand works hard to mitigate. There is an avalanche of funding pouring into security solutions designed from the ground up for an agentic world, as well as those preparing for a post-quantum era. For customers, the wisest move is investing in harnesses capable of operating across diverse cloud vendors and on-premises open-source deployments. Traceability and provenance rise as new centers of gravity.

Humans are inherently competitive and violent. The struggle between the United States and China is no longer about deterrence; it is about absolute dominance. In the cognitive dimension, contemporary Western protests against data centers smell suspiciously similar to the Cold War pro-peace movement sponsored by the Soviet Union.

In the kinetic dimension, the cheap, mass lethality pioneered in Ukraine has power grids, industrial plants, and domestic airspace in its sights. Plan accordingly.

In the spirit of unrestrained competition, it is mandatory to continuously study the Chinese approach to building the new AI-friendly economy: the state works alongside research institutes and corporations to establish standards and security protocols, creating what is effectively a civilizational digital thread. It makes absolute sense for American companies to utilize open-source Chinese models – on-premises, behind proper guardrails.

From an organizational rollout perspective, what is the best way for an enterprise to deploy AI: a dedicated AI department, or AI capability and authority distributed among domain experts? In the first case, technicians who know how to prompt models will have little grasp of aerospace engineering specifics. In the second, domain experts will require continuous tooling support, training, and strict security guardrails. On top of that, as the C-suite battles trends, speed, employee morale, and associated funding and hiring priorities, organizational psychology suddenly makes itself great again.

On a final note, we at Senticore continue developing our tool for converting technical publications into actionable work plans and integrating them directly with systems like MES. As a bonus, we quickly discovered that the tool is an exceptional fit for researchers working with multiple sources, particularly in defense. Claude has become an invaluable collaborator and force multiplier, and we continuously experiment with locally deployed open-source models from American vendors (with unexpected kudos to Microsoft) and from China, with little reason to explore offerings coming out of Europe. We assembled what we consider a robust harness, and after discovering that 73% of our agents’ runtime is spent inside that harness (rather than on inference), we invested in higher-end local hardware. Surprisingly, our biggest effort goes not into wrestling with AI per se, but into building a user experience that really rocks. How innovative!

This conversation started with evolution, which rarely favors the complacent. Watch the 2026 Netflix Dinosaurs series and the 2004 film Kung Fu Hustle. The former shows how seemingly invincible titans find themselves vulnerable and irrelevant when their environment shifts; the latter blends styles, wills, and disparate traditions – demonstrating the depth of China’s comprehension of the West and a synthesis of cultural discipline with technological ruthlessness. They are playing to win. So should you.