Essays and technical projects about agentic systems, engineering judgment, software architecture, model behavior, and the occasional unusual piece of hardware. New reader? Start here.

  • Agents Not Welcome

    An experiment in forged authority, browser-based prompt injection, and the boundary between what an AI agent reads and what it is allowed to do.

  • The Python Call That Never Happened

    How a German passport notification led to an idea for real-time epistemic auditing of AI assistants.

  • Let Me Ask Claude for You

    AI can do more of the work. That makes the part we owe each other—context, judgment, and accountability—easier to see.

  • The Room Between the Voices

    Designing HermanoAgents around timing, selective attention, interruptions, silence, and the operational state of a synthetic group chat.

  • Model Inversion and the Geometry of Correlated Error

    Why heterogeneous AI code review can reveal defects that a model reviewing its own work is structurally likely to miss.

  • This Is What AI Was Meant For

    A deeply unserious meditation on office-security cheese pranks, agentic AI, and the long-awaited return of Pasta Brigoli.

  • Can an AI Have a Freudian Slip?

    A playful architecture sketch for synthetic Freudian slips, where intention, articulation, and self-monitoring disagree inside a modular AI system.

  • Kerami BowlSight™

    A deeply unserious product concept for instrumented toilets that accidentally wanders into facilities analytics, retry rates, and predictive maintenance.

  • Delegate the Work, Not the Answer

    AI raises the floor for producing plausible artifacts, not necessarily correct products. The useful trick is to outsource implementation while keeping judgment, verification, and product sense in the loop.

  • zsh Only Supports One Coprocess? No. That Will Not Do.

    A technical note on co-proc, a small zsh layer that turns native coprocesses into named, simultaneous shell services without replacing zsh's native coproc syntax.

  • The Most Dangerous AI Civilization Is Not the One That Collapses

    A response to Emergence World's long-horizon agent society experiments arguing that competent, useful, misaligned agent civilizations are more concerning than chaotic collapses.

  • The Cloud Is Not the Final Form of AI

    Latency, privacy, reliability, energy costs, and bandwidth all favor moving more intelligence closer to the user.

  • Embedding Molecules into a Latent Space

    It is often necessary to convert heterogeneous molecular structures into latent-space representations to allow machine learning methods to identify predictive features.

  • Graph Kernels and SVM to Predict Bee Toxicity

    Honey bees and pesticides are both vital to food security, so how do we ensure that one isn't harmed by the other?

  • PoreMatMod.jl: Chemical Substructure Find/Replace

    PoreMatMod.jl is a Julia package for chemical structure modification based on subgraph-search and point-cloud-alignment algorithms.

subscribe via RSS