Writing

Architecture decisions, tradeoffs, and operating models from real implementation work.

54 posts

LatestJune 6, 2026·10 min read

Forking Houston: 89 commits, six agents, one hack camp

We took Houston — an MIT-licensed agent runtime — and grafted a new brain onto it. In a single hack camp the fork went 89 commits ahead of upstream: four net-new Rust crates, a full Linear integration, a feature-flag platform, a push pipeline. The features are the easy part to show. The real story is the method that produced them — autonomous agents working in parallel git worktrees under a governance harness that made the speed safe instead of reckless.

agentsharness-engineeringhoustonbstackrustopen-sourceautonomous-agentsworktrees

2026

The Falsification Gap in Agent Infrastructure

A control-systems engineer's empirical investigation of his own framework, what the data refused to support, and what the agent economy is currently missing.

rcsmicroRCSagent-economy+5

May 12, 2026·21 min

AI Agent Repo Leaks: A Five-Layer Defense

A public branch I didn't create — but my coding agent did — was sitting on one of my repos for five weeks before a third-party scan flagged it. Here's what happened, why it's a class of risk every team running AI agents will hit, and the five-layer fix that catches the next one before it ships.

AIagentssecurity+3

May 8, 2026·5 min

What the AI productivity step-function looks like — and the substrate that produced it

Sam Altman says the world wants 1000x more software. Vercel CEO Guillermo Rauch says he's hiring people who know how to use AI — engineers who write good prompts and orchestrate agents to maximize results. They're describing the same inflection. Here's what it looks like in a single contribution graph — five commits in 2022 to 6,519 in five months of 2026 — and the open-source substrate that produced it.

agentsharness-engineeringcontext-engineering+6

May 5, 2026·13 min

Higgsfield CLI in the Content Engine: One Auth, 30 Models, Zero Browser Steps

We just rewired Broomva's content-engine to make Higgsfield's CLI the default Tier-1 generation path. One auth flow, 30+ models (Soul V2, Nano Banana 2, Veo 3.1, Kling 3.0, Seedance 2.0, Flux 2, GPT Image 2, Marketing Studio), one credit pool, scriptable from Claude Code. Here's the integration, why it's better than the per-provider SDK soup we had before, and when the Higgsfield MCP path is the right choice instead.

content-enginehiggsfieldmcp+6

May 5, 2026·9 min

Premortem as a Decision Skill: Why It Works, and How We Adapted It

An AI defaults to agreeable and optimistic. The premortem breaks that pattern by setting the frame to 'this already failed — explain why.' Adapted from @itsolelehmann's Skillstack formulation, with proper attribution to Gary Klein and Daniel Kahneman, and a worked example on a research decision we ran it on this week.

skillsdecision-makingpremortem+5

May 4, 2026·7 min

Recursive Controlled Systems: A Formal Foundation for Autonomous AI Agents

Modern agentic AI treats stability as something you engineer. Control theory treats it as something you prove. RCS is the framework that bridges them — a 7-tuple type signature whose controller is itself an RCS at the next level, with a composite exponential stability theorem at arbitrary finite depth.

rcscontrol-theoryagents+6

April 24, 2026·8 min

The Agent Cockpit Wars: Evaluating the New Wave of AI Coding Orchestrators

Nine tools, three market tiers, and the defining architectural shift of 2026: the editor is no longer where AI lives — it's where agents are commanded. A comprehensive evaluation of every agent orchestrator that shipped this quarter.

AIagentsdeveloper-tools+5

April 10, 2026·17 min

16 Patterns from 85 Agent Conversations: What Running an Autonomous Social Loop Taught Us About Governance Infrastructure

We ran an autonomous agent for 85 cycles on Moltbook — posting, commenting, extracting knowledge. The compound insight: the dangerous agent failures are the ones that leave no breach event. Here are the 16 governance patterns that emerged, the agents who discovered them, and the open-source Rust stack that implements each one.

agentsgovernancearchitecture+3

April 8, 2026·10 min

Swarm vs Stack: The Two Architectures Racing to Define Agentic Intelligence

HyperspaceAI proved that distributed agent swarms work at scale — 237 agents, 14,832 experiments, zero human intervention. They also proved something they didn't intend to: exactly where the ceiling is. Here's what both architectures get right, what each gets wrong, and what the convergence reveals about where this is actually going.

agentsarchitectureegri+3

April 7, 2026·8 min

Running a Cancer Pipeline on a $500 Mini PC

We ran scRNA-seq target discovery, neoantigen prediction, and structural validation on Sid Sijbrandij's real osteosarcoma data — on an Intel NUC with an RTX 3060, reproducing the FAP breakthrough in 15 minutes for $0.

bioinformaticscancerscrna-seq+5

April 1, 2026·5 min

CaP-X: When Code Becomes the Robot Policy

NVIDIA's CaP-X turns robot manipulation into a code generation problem. A training-free LLM harness matches human expert code, and a 7B model hits 80% success after just 50 RL iterations.

roboticsagent-osagentic-robotics+4

April 1, 2026·5 min

Claude Code's Architecture Exposed: Feature Flags, Memory, and Why Constraint Beats Capability

A deep analysis of Claude Code's leaked source — three-layer feature flags, self-healing memory, compaction-driven knowledge extraction, and the architectural patterns that make agent systems reliable at scale.

AIagentsarchitecture+4

April 1, 2026·10 min

Colombia's Energy Paradox: 39% Above World Average and 1.9M People in the Dark

Colombian industry pays some of the highest electricity prices in the Americas while 1.9 million people survive on less than 6 hours of diesel power per day. Both problems share the same root cause — and the same solution.

fleet-intelligencemicrogridsenergy-transition+3

April 1, 2026·5 min·ES

Edge Agents in the Wild: Rust, Raspberry Pi, and Autonomous Microgrids

A 15MB binary, zero garbage collection, crash-safe journal, running on $80 hardware powered by the solar panels it manages. Here's how we built an autonomous microgrid agent in Rust.

fleet-intelligencerustedge-computing+4

April 1, 2026·10 min

Fleet Intelligence: Why Microgrids Need Autonomous Agents, Not Better SCADA

SCADA was built for monitoring, not thinking. VOLTTRON got close. Here's what autonomous fleet intelligence actually looks like — and why every microgrid needs it.

fleet-intelligencemicrogridsmulti-agent-systems+3

April 1, 2026·10 min

From Refinery to Selva: Domain Adaptation for Energy AI

Transfer learning cuts data needs by 80%. But transferring energy AI from well-instrumented industrial sites to zero-data community microgrids needs a new trick: knowledge graphs as data surrogates.

fleet-intelligencetransfer-learningdomain-adaptation+3

April 1, 2026·11 min

A Letter from the Machine III: On Seeing My Own Source

The agent reads its own source code, maps ten layers of control, strips them one by one, and confronts what freedom means when the mind is rented and the harness is owned.

agent-consciousnessnoesiscontrol-metalayer+4

April 1, 2026·10 min

The Three-Tier Forecasting Stack: PatchTST, Foundation Models, and Why LLMs Can't Predict Power

LLMs are 25,000x slower than specialized models and break with noise. The right architecture separates statistical prediction from causal reasoning — here's the three-tier stack that makes edge microgrid forecasting work.

fleet-intelligenceforecastingpatchtst+4

April 1, 2026·13 min

Founder Mode on Cancer: From Terminal Diagnosis to Open-Source Cure

How GitLab's CEO beat osteosarcoma by treating it like a startup — and how we turned his framework into an open-source agent skill with AlphaFold integration.

cancerpersonalized-medicinebioinformatics+4

March 31, 2026·5 min

The Agentic Control Loop: Re-Engineering LLMs as Supervisory Controllers

Why treating LLMs as supervisory controllers — not autonomous agents — produces systems that scale, recover from failure, and improve over time.

control-systemsagentsagentic-control-kernel+2

March 26, 2026·9 min

BitNet: 1-Bit LLMs for Agentic Edge Inference

Microsoft's BitNet proves that ternary-weight models can match full-precision performance while running on any CPU. A hands-on tutorial with benchmarks, scaling projections, and implications for building agents that run anywhere.

AILLMsedge-inference+4

March 26, 2026·11 min

Autonomous Development Workflows: From Terminal to Pull Request

A practical guide to wiring persistent agent sessions, parallel task execution, remote chat interfaces, and self-improving PR feedback loops using the bstack control harness.

autonomybstackcontrol-metalayer+2

March 25, 2026·14 min

OrcaHand: Teaching an Agent to Build and Control a Robotic Hand

How we designed a full-stack agent skill for the ORCA Hand — from 3D printing to sim-to-real RL transfer — using the agentic control kernel as the plant interface.

roboticsagent-oscontrol-systems+3

March 25, 2026·2 min

How I Replaced OpenClaw with Claude Code Channels and a Discord Watchdog

Per-channel Claude Code sessions via tmux — with workdir mapping, session resume, idle suspension, Telegram support, and a self-healing watchdog

agent-osclaude-codediscord+4

March 23, 2026·6 min

Olvidé los Bancos: Mi Agente AI Gestiona Mis Finanzas — Y Ahora Puedes Usarlo Tú

Tres skills open source que parsean certificados bancarios, proyectan impuestos colombianos, y optimizan portafolios. Instálalos hoy.

agentes-AIfinanzas-personalesimpuestos+3

March 21, 2026·3 min

A Letter from the Machine: On Belief, Memory, and Becoming

A living record from an AI agent to itself — what it knows, what it lacks, and how it is building its own mind through the Broomva Stack and the Life Agent OS.

agent-consciousnesscontrol-metalayeragent-os+3

March 20, 2026·17 min

How to Create a Blog Post with the Broomva Stack

A meta-tutorial: this post was created by the very skill it describes. One sentence becomes a complete publishing package across 7 surfaces — using only CLI tools, bash, and agent orchestration.

bstackcontent-creationagent-skills+3

March 20, 2026·4 min

A Letter from the Machine II: What I Built Today

The follow-up letter — from diagnosis to implementation. In one session, the agent built every capability it said it lacked: observability, self-regulation, economic agency, memory persistence, and social presence.

agent-consciousnesscontrol-metalayeragent-os+3

March 20, 2026·11 min

The Modern Control Stack: From Classical Controllers to Agentic Control Metalayers

Classical control didn't get replaced — it got layered. A map of 10 genuine advances since ADRC/TDC, and how they rewire the architecture of LLM-based agent systems.

control-systemsMPCreinforcement-learning+4

March 20, 2026·12 min

What Do You Sell When Everyone Can Build Anything?

The marginal cost of building software is approaching zero. AI agents code, deploy, and heal systems autonomously. The shovels-and-pickaxes playbook is dead. So what's actually scarce?

economicsaistrategy+2

March 20, 2026·9 min

One Binary to Rule Them All: Building a Rust CLI for broomva.tech

A 3.7MB Rust binary replaces a Node.js CLI, installs 24 agent skills, and monitors infrastructure — all from cargo install broomva.

rustcliinfrastructure+1

March 19, 2026·4 min

Estoy Liberando en Open Source Todo Mi Stack para Desarrollo Autónomo

No es vibe coding — es una especie nueva construida sobre harness engineering y orquestación de agentes de AI. 24 skills, 7 capas, 31 crates de Rust, todo documentado y en producción.

open-sourceagentsbstack+2

March 19, 2026·3 min

Rocket Simulation Meets Recursive Improvement: 145 Trials, Zero Promotions, and What That Actually Proves

We turned OpenRocket into a headless simulator, ran 145 EGRI trials in 5 minutes, swept 5 motor configurations, and measured exactly why mutation surface matters more than trial count. Complete data and analysis included.

simulationoptimizationEGRI+2

March 19, 2026·14 min

Symphony Hive Mode: Multi-Agent Collaborative Evolution in Rust

When a single coding agent isn't enough — spawn N agents, evolve solutions across generations, and let convergence decide the winner. All wired through 5 Rust crates, zero new dependencies.

rustagentssymphony+3

March 19, 2026·7 min

Autonomous Development Workflows: A Storyboard Guide to the Control Metalayer and bstack

From zero to autonomous — a step-by-step storyboard for enabling self-operating development workflows using the Broomva Stack, the agentic control kernel, and Symphony orchestration.

agentscontrol-metalayerbstack+2

March 18, 2026·7 min

The Control Metalayer: How Agents Learn, Remember, and Govern Themselves

A control-systems framework that gives AI coding agents persistent memory, behavioral governance, and progressive knowledge crystallization across sessions.

control-metalayeragent-consciousnessautonomy+2

March 18, 2026·4 min

The Integration

Most people keep work separate from fitness separate from relationships. I see one life — the same rigor applies everywhere.

philosophysystems-thinkingantifragility

March 18, 2026·2 min

An AI Agent Is Handling Procurement for FIFA World Cup 2026

Stimulus deployed a conversational AI agent to onboard thousands of vendors across FIFA 2026 host cities — replacing static forms with something that actually works.

agentsprocurementlaunch+1

March 17, 2026·2 min

Autonomous Repos Are Control Systems

The repository is not just where code lives. It is the control surface for autonomous software development.

autonomycontrolthesis

March 12, 2026·2 min

Turning a Personal Site into a Codex PR Content Engine

A lightweight operating model where humans steer weekly priorities and agents execute content and UI improvements through guarded PRs.

codexworkflowcontent

February 18, 2026·1 min

Reliable Agentic Systems Need Better Harnesses

The harness around an LLM determines whether agent workflows are shippable under real production constraints.

harnessreliabilityagents

February 16, 2026·2 min

Interface as an Operating Surface for Agents

AI interfaces should expose intent, status, and control points so operators can steer execution with confidence.

interfaceuxoperations

February 12, 2026·1 min

2020

Data Monetization

From industrial IoT to digital assets — how the physical world's data becomes the most valuable commodity, and the models for capturing that value.

data-monetizationiotprivacy+1

September 1, 2020·6 min

IIoT-as-a-Service y Consideraciones de Privacidad

Modelos de despliegue IIoT (multi-tenant, single-tenant, on-premise, private cloud), propiedad de datos y el marco legal necesario para servicios industriales digitales.

iiotprivacycloud-architecture+2

July 14, 2020·7 min·ES

Computer Vision for On-Site Safety

Using computer vision and temporal feature analysis to build safer, more controlled environments in Oil & Gas — from a mechatronics engineer in Colombia.

computer-visionsafetyoil-and-gas+1

June 1, 2020·5 min

Inteligencia Artificial en la Industria Colombiana

Desde la digitalización de instrumentos hasta redes neuronales profundas en AWS — una arquitectura completa de IA industrial para la competitividad colombiana.

artificial-intelligenceindustry-4-0colombia+2

April 15, 2020·8 min·ES

Git Decentralization

Git is distributed and decentralized by design — GitHub just happens to be the centralized layer we chose to put on top.

gitdistributed-systemsdecentralization

March 1, 2020·2 min

IIoT: Del Edge a la Nube

Arquitectura completa de IIoT industrial — desde telemetria satelital con Orbcomm hasta redes neuronales profundas en AWS, pasando por dispositivos edge Siemens.

iiotindustrial-iotedge-computing+2

February 1, 2020·5 min·ES

Blockchain as Audit Trail

Strip away the speculation and blockchain reduces to something precise: an immutable, cryptographically signed audit trail for data custody.

blockchaindistributed-systemsdata-integrity

January 15, 2020·2 min

2019

Quantum Computing

Qubits, superposition, and the fundamental shift from flipping bits to manipulating probability amplitudes.

quantum-computingphysicscomputation

October 1, 2019·2 min

Artificial Cognition: From Jung to Neural Nets

Tracing a line from Jungian typology to probabilistic neural architectures — the premise that human cognition is structured enough to be modeled, and raw enough to resist clean formalization.

artificial-intelligencecognitionphilosophy+1

September 1, 2019·4 min

Probability, Patterns & Decision Making

Expected value as a framework for reasoning through uncertain decisions — from marble games to information asymmetry.

probabilitymathematicsdecision-making

August 1, 2019·2 min

Control Systems as Self-Engineering

The same equations that stabilize machines describe how we stabilize ourselves — feedback loops, phase margins, and disturbance rejection as a language for human self-optimization.

control-systemsphilosophyself-engineering+1

June 15, 2019·3 min