Santosh Kumar Radha

When AI absorbed the execution layer, the remaining work changed shape entirely. The human contribution moved from producing the output to defining the space within which the output must live.

Human-Computer Interaction has had four decades of formal study. Human-Agent Experience has had none. This is an attempt to build the missing design theory: three structural dimensions, eight regions, and a map of the space that every AI product inhabits.

Our engineers use Claude Code as a daily collaborator. We wanted dozens of instances coordinating in parallel on a shared codebase. Here is what we learned after 200+ agent invocations, $116 builds, and three nested control loops.

In October 2024, we published Composite Learning Units. In March 2026, Dupoux, LeCun, and Malik published a cognitive-science-grounded architecture for autonomous learning. The two papers converge on the same structural diagnosis.

The most profitable machine ever designed is the slot machine. I never expected to recognize the same pattern in my own work sessions at 2am, fully aware of what was happening.

The build-versus-buy question is back. This time the answer changed because intelligence moved.

As intelligence becomes abundant, coordination becomes the limiting factor, one that can be designed intentionally or left to emerge on its own.

We are grouping very different kinds of systems under a single mental model. Once you look at AI through the lens of where intelligence sits in the architecture, a clearer split begins to appear.

Monoliths became microservices, and what followed was a decade of cloud native infrastructure closing gaps. Now reasoning has moved inside the backend, and every solution we built rests on an assumption it violates.

An agent approved 50,000 micro-loans in three minutes, and every individual decision passed muster. But the portfolio that emerged was toxic.

A returns agent approved a $12,000 refund it shouldn't have, and the strange thing was that nothing looked wrong. This is becoming the new failure mode for AI systems.

When software reasons, trust becomes a problem. If you can't answer 'who did this, and were they authorized?' then you don't have guided autonomy. You have unaccountable autonomy.

Five years from now, every serious software company will have an AI backend. Not a chatbot. Not a copilot. A reasoning layer that sits alongside their services.

For thirty years, we built a web for human eyeballs. That era is over. As autonomous agents become our digital proxies, the most valuable parts of the internet will become an engine of pure function.

A quick and dirty implementation of binary addition using quantum circuits in Qiskit, demonstrating quantum arithmetic with carry propagation.

A tribute to my PhD advisor Walter R.L. Lambrecht, reflecting on years of collaborative research and learning at Case Western Reserve University.

Talk abstract on topological phase transitions in 2D honeycomb Group-V systems, from nodal line semimetals through Dirac cone annihilation to higher-order topological insulators.

Using Julia to model halide perovskite structures as pseudo-spin systems with Monte Carlo methods, exploring ferroelectric transitions and crystal stability.

Proposing a topological quantum switch in antimonene using electric field to break inversion symmetry, creating controllable 1D conducting channels.

Continuation of the Julia Monte Carlo project: corrected periodic boundary conditions and added dipole-dipole interaction calculations for perovskite pseudo-spin systems.

Understanding the Bogoliubov-de Gennes formalism through a simple 2-site superconductor, exploring particle-antiparticle symmetry and parity phase transitions.

Starting from the Poincare group and building toward quantum mechanics from relativistic field theory. First in a series on QFT from a condensed matter perspective.

Using the Ising model from condensed matter physics to understand crowd mentality in stock markets, with Numba optimization for blazing fast Monte Carlo simulations.

Introduction to calculating Heisenberg interaction J_ij for spin Hamiltonians using perturbation theory and transverse spin susceptibilities.

Optimizing the pySKTB tight-binding package using Numba and line_profiler, achieving an order of magnitude speedup for quantum calculations.