Emergent agency & multi-agent systems
How populations of interacting agents discover niches, cooperate, and develop higher levels of causal control over their environment.
Currently building AgentFarm
Chris Mangum
Independent researcher · artificial agents & computational modeling
I study how agency and intelligence emerge in simulated worlds — building open-source platforms for multi-agent simulation, reinforcement learning, and neuroevolution. Blending engineering and art to learn the logic of nature.
Research focus
Studying the emergence of agency
My work sits at the intersection of artificial life, reinforcement learning, and complex systems — asking how simple agents, under selection pressure, develop increasingly capable and adaptive behavior.
Neuroevolution & reinforcement learning
Evolving heritable hyperparameter genomes and decision policies, and probing when learning beats its own initialization.
Complex adaptive systems
Treating simulation as a microscope — measuring fitness, selection, and inheritance with paired seeds, confidence intervals, and robustness gates.
Featured work
The Dooders project
An open-source research effort investigating the emergence of intelligent agents within a computational model of reality. Its flagship platform, AgentFarm, is where the experiments run.
Dooders · open-source platform
AgentFarm
A Python-first platform for agent-based simulation, reinforcement learning, and emergent-behavior research. Multi-agent simulations with configurable genomes and actions, a DQN/neuroevolution decision stack, an experiment runner with SQLite-backed metrics, and an analysis pipeline — all wrapped in a documented, reproducible workflow.
Selected projects
What I'm building
Current open-source work across the Dooders organization and my personal account — simulation platforms, agent memory, and speech- and language-native agents.
Dooders
AgentFarm
Open-source research platform for simulations of complex systems — agent-based modeling, RL experiments, and analysis.
Dooders
AgentMemory
A lightweight, pluggable memory backend for agent-based simulations — temporal data, experience replay, and persistent state logging.
Dooders
Dooders
The original agent/arena experiment — evolving agents that fit their environment and discover unrealized niches. Reality works. Simulate it.
Personal
opusagent
A low-latency realtime bridge integrating telephony for a stateful, speech-native rational agent.
Personal
AgenticPoker
A 5-card draw poker simulation where LLM-driven agents strategize, learn, and trade natural-language table talk.
Personal
GCA
Generative Cellular Automata — exploring how local rules give rise to rich, emergent global structure.
Writing
From the AgentFarm devlog
Build notes, design decisions, and experiment outcomes — written as the research happens, with real results and the occasional dead end.
-
2026-06-20
The transferable-signal gate: do learned policies beat their own init?
Before building richer inherited payloads, confirm there is anything worth inheriting — paired held-out rollouts show a modest but robust early-age decision-quality signal.
-
2026-06-09
When every agent has a different goal
Making the reward function itself a per-agent, heritable trait — un-curated objective diversity measurably lowers collective fitness across 20 paired seeds.
-
2026-06-04
Are we measuring at the wrong level?
Re-scoring a 36-run inheritance A/B at the newborn level — small robust behavioral shifts, but the population-level null wasn't a measurement artifact.
-
2026-05-21
Baldwinian vs Lamarckian: policy warm-start across three resource regimes
A full 36-run matched matrix — warm-start applied most of the time and paired runs diverged, but no profile cleared the robustness gate.
-
2026-05-18
Gene flow and the buffer
A crossover-enabled rerun closes the buffer arc with a profile-dependent result on how gene flow reshapes speciation.
About
Independent research
I'm an independent researcher focused on artificial agents and computational modeling. My through-line is agency — how it arises, how it can be measured, and how it grows when agents are placed under real selection pressure inside simulated worlds.
Most of that work lives in the open under the Dooders project, where AgentFarm serves as the experimental platform. Alongside the simulation research, I build speech- and language-native agents and tools, and I write up results — successes and dead ends alike — in the devlog.
I treat simulation as a microscope and care about doing it rigorously: paired seeds, confidence intervals, and robustness gates before claiming an effect is real.
Attempting to blend engineering and art to learn the logic of nature.