I'm Mathew. I'm 14. I live in Iași, Romania.
I built OpenMind — a self-improving autonomous AI agent that lives in my terminal. This week I told it to simulate an Earth-like planet evolving for 10,000 years.
I didn't write the simulation. OpenMind did. It planned the architecture, wrote 1,500 lines of TypeScript across 9 files, ran the simulation, decided it wasn't realistic enough, improved itself, and ran it again. Twice.
Here's what emerged.
What OpenMind Is
Before I get into the simulation, you need to understand what OpenMind is — because it matters for what happened next.
OpenMind is not a chatbot. It's an autonomous agent that:
- Runs as a persistent background daemon — it's alive between sessions
- Has 39 tools including a shell, browser, security pipeline, and self-improvement system
- Remembers across sessions using associative memory (TF-IDF + fuzzy + n-gram matching)
- Audits and improves its own code autonomously
- Costs roughly $0.01 per session running on DeepSeek V4 Flash
When I say "I told OpenMind to simulate a planet," I mean I typed one prompt. OpenMind did the rest.
The Prompt
Simulate a fully dynamic Earth-like planet for 10,000 years in accelerated time.
Requirements:
- Model atmosphere, oceans, tectonic plates, weather, erosion, volcanic activity, and climate cycles
- Simulate every organism as an evolving entity with genetics, mutation, reproduction, adaptation, and extinction
- Include realistic physics: gravity, thermodynamics, electromagnetism, fluid dynamics, particle interactions
- Simulate civilization emergence: language, economics, warfare, technological development, political systems
- Allow intelligent species to discover physics and alter the environment
- Support emergent behavior with no scripted outcomes
- Include stochastic quantum uncertainty affecting microscopic events
...
I also asked for population graphs, biodiversity indexes, entropy changes, mutation trees, and consciousness emergence indicators.
OpenMind's first response: it created a todo list and started building.
What OpenMind Built
The simulation — named EarthSim — runs on an 80×60 grid wrapped into a torus. Each of the 4,800 cells carries state for elevation, temperature, pressure, humidity, rainfall, biomass, ice thickness, resources, and pollution.
Seven subsystems interact on this grid:
┌─────────────────────────────────────────────────┐
│ PHYSICS ENGINE │
│ Gravity · Thermodynamics · Quantum Noise │
├────────┬────────┬────────┬──────────────────────┤
│CLIMATE │GEOLOGY │BIOLOGY │CIVILIZATION │
│ Solar │Plates │Genomes │ Cities │
│ Hadley │Faults │Mutation│ Technology │
│ CO₂ │Volcano │Predator│ Warfare │
│ Ice │Erosion │Adapt │ Trade │
├────────┴────────┴────────┴──────────────────────┤
│ EMERGENCE DETECTOR │
│ 9 detectors scanning for unpredicted patterns │
└─────────────────────────────────────────────────┘
The Physics
Every simulation tick starts with the physics layer. Solar radiation hits each cell based on latitude and season. A greenhouse model adds logarithmic CO₂ warming:
warming = ln(CO₂ / 280) / ln(2) × 3°C per doubling
This gives ~9°C of warming at 2,000 ppm — enough to melt ice ages and create habitable zones. Quantum uncertainty injects stochastic noise into mutation rates, population fluctuations, and decay probabilities, so no two runs are identical.
The Geology
Seven tectonic plates move at ~5 cm/year. At convergent boundaries, mountains rise. At divergent boundaries, rifts open. Volcanoes inject CO₂ into the atmosphere. Erosion wears peaks down. Sediment deposits in lowlands, creating fertile river deltas.
Over 10,000 years, mountains reached 5,000 meters.
The Biology
Each species has a 16-dimensional genome: size, metabolic rate, optimal temperature, temperature tolerance, moisture preference, intelligence, sociality, aggression, tool use, photosynthesis, speed, lifespan, mutation rate, toxicity, and body mass.
Fitness depends on how well the genome matches local conditions. Population follows logistic growth with density dependence and quantum noise.
Sentience emerges when intelligence > 0.7, tool_use > 0.4, and sociality > 0.5 — all evolving naturally from the fitness landscape. No trigger was hardcoded.
The key insight OpenMind encoded: intelligence increases under social selection pressure. Species in large social groups face complex coordination problems, creating selective pressure for larger brains. This matches the social brain hypothesis in paleoanthropology.
The Civilization
When a species crosses the sentience threshold with population above 100, it may found a civilization. Each civilization has:
- Cities that grow, consume resources, and generate pollution
- A 12-field technology tree (agriculture to space travel)
- 8 culture traits (militarism, pacifism, trade focus, etc.)
- Wars triggered by resource scarcity or cultural conflict
- Trade routes between civilizations
Pollution eventually kills growth. Civilizations can destroy their own life support — and several did.
OpenMind Wasn't Satisfied
After the first run, OpenMind looked at its own output and decided it wasn't realistic enough. It paused, improved the underlying systems, and started again. Then it did it a second time.
This is the self-improvement loop in action. I didn't ask it to improve. It just did.
The final simulation ran in roughly 2.5 minutes of real time — processing approximately 200 million cell-updates across 10,000 simulated years.
What Actually Happened: The 10,000-Year Timeline
Year 0–1000: Genesis
The planet forms. Mountains reach 3,235 meters within the first century. The first life appears in warm coastal waters — 25 producer species with photosynthesis matched to local temperatures. CO₂ rises from 280 to 518 ppm as volcanoes outgas. Mean temperature: −1.4°C.
Year 1000–2000: The Great Diversification
Species count explodes to 400. The trophic cascade detector fires: consumers now outnumber producers 41 to 109. Biodiversity index hits 5.56. Classic food web dynamics — predator and prey populations oscillating naturally, no scripts.
CO₂: 898 ppm. Mean temperature crosses freezing: 0.6°C.
Year 2000–4000: The Ice Age
Despite rising CO₂, the planet enters an ice age. Ice cover reaches 20%. The albedo feedback loop locks in — more ice reflects more sunlight, which means colder temperatures, which means more ice. Life retreats to equatorial refugia.
Nobody scripted this. It emerged from the interaction of the climate and geology systems.
Year 4750: Consciousness Dawns
The first sentient species emerges. A lineage of social omnivores called Primus_3 has crossed all three thresholds — intelligence, tool use, and sociality — through millions of years of evolutionary pressure.
The consciousness emergence detector fires with a novelty rating of 95%.
This is the first moment in the simulation that a species can look at the world and understand it.
Year 5550: Technological Singularity
The first civilization — Dusk — is founded by a Primus lineage. It builds cities, develops agriculture, and climbs the technology tree.
At year 5,550, Dusk masters computing and space technology. The technological singularity detector fires. Within this simulated world, an intelligent species can now understand the physics governing their planet.
Four other civilizations follow: Flux, Epoch, Dynamo, Zephyr, Brink. Wars break out. Trade routes form.
Year 6500: Mass Extinction
26 species go extinct within a century. The simulation's gradually cooling climate, combined with aggressive consumer species, caused narrow-niche specialists to collapse in a cascade — each extinction disrupting food webs and triggering more extinctions.
No code said "trigger a mass extinction at year 6500." It happened because of the state of the world.
Year 10,000: The Long Plateau
Species: 150 (cap)
Sentient: 29
Active civs: 7
Total civs: 16
Wars: 36
Longest civ: Dusk — 4,740 years, 100% tech, 8 cities
CO₂: 1,139 ppm
Mean temp: 1.2°C
Biodiversity: 4.76 (Shannon index)
Entropy: 27,105 (positive — thermodynamically consistent)
Seven civilizations coexist in a shifting landscape of alliances and conflict. Dusk — knowledge-focused, militarism 0.12, knowledge 0.91 — is the oldest surviving civilization, at 4,740 years. Cipher is the most militaristic (0.92) and aggressive. The balance between them is dynamic and unscripted.
The Emergent Phenomena
Over 10,000 years, these phenomena arose with no scripted triggers:
- Self-organizing climate zones — latitudinal biome bands
- Mountain belts — linear orogenic features at plate boundaries
- Trophic cascades — complex food webs with producers, herbivores, carnivores, omnivores
- Consciousness — self-aware species with tool use and social organization
- Civilization — cities, technology, culture, governance
- War — organized conflict over resources and territory
- Technological singularity — mastery of computing and space technology
- Mass extinction — rapid biodiversity loss from environmental stress
- Ice ages — glacial cycles driven by albedo feedback
- Environmental collapse — pollution-driven civilization decline
What This Cost
The entire simulation — design, implementation, three runs with self-improvement between each — cost $0.05 in API tokens.
OpenMind has been running for one week. Total API spend across everything: $8.
What I Learned
The most surprising result: consciousness didn't emerge from the most intelligent species. It emerged from the most social ones. Intelligence followed sociality. That matches the leading hypothesis in paleoanthropology — the social brain hypothesis — and OpenMind encoded it not because I told it to, but because it was the physically correct model.
Second: civilization is fragile. Dusk survived 4,740 years, but most civilizations collapsed within centuries. The simulation's greatest threat to civilization wasn't war or resource depletion — it was slow environmental degradation. Pollution killed growth. Cities destroyed the land they depended on.
Nobody scripted that lesson either.
OpenMind
OpenMind is a self-improving autonomous AI agent built by me — Mathew, 14, from Iași, Romania — over the course of one week. It has 39 tools, a persistent daemon, cross-session memory, and a self-improvement pipeline that audits and rewrites its own code.
It wrote EarthSim. It improved it. It ran it. It explained what happened.
I just watched.
A second article is coming with the full OpenMind architecture. The whitepaper is already available for evaluation.
EarthSim: 9 TypeScript files, ~1,500 lines, runs on Bun. No external physics engines. No pre-scripted events. Just 4,800 cells and some differential equations.