0. Looikos: the apex (the simulation compiler)
Looikos (L-O-O-I-K-O-S) is the ecosystem itself, and what it's becoming is best named a simulation compiler. Andy says the concept is early enough technologically that he struggles to define exactly what it is, and "simulation compiler" is the closest handle he can put on it. A category this new gets named by what it does, because no existing label fits it yet.
The thesis runs along a lineage Andy lays out in order.
Linux is the proof. He has been heavily inspired by the Linux ecosystem and specifically by how damn robust it is. Many have tried to overthrow it; none have managed it. That durability is what convinced him that open-source and decentralization are certainly where the future is moving. The only open question was in what format an open, decentralized future arrives. He believes he now has a decent thesis on the format, which is why he describes himself as past the thinking and into the execution phase.
Compilers are universal. It's the load-bearing intuition he couldn't shake. He genuinely thinks compilers are a universal technology, on the order of a physical law of any computing civilization. Even aliens will have a compiler of their own source, because to turn inert silicon into instruction, at some point electrons have to become data. There are different ways to format that translation, but a different format is just a complete compiler running different instructions. A compiler is the thing that necessarily exists wherever computation exists.
The move is to point that universal wizardry at a new substrate. Andy couldn't let go of the idea: what if the same apparently-universal compiler wizardry could be aimed at something other than source code? Instead of source into machine code, it would compile the lived structure of a market, an audience, or a domain into the agent-native systems that operate inside it. The simulation compiler compiles a modeled world into the systems that act on that world. (Extrapolation, consistent with his framing: this is why the data platform, the world-modeling primitives, and the agent harnesses are all one project rather than several. They're the front end, intermediate representation, and back end of the same compiler.)
Looikos becomes a simulation compiler emergently, not by declaration. He's explicit that this is a consequence of scale. Once the ecosystem runs dozens of projects collectively managing tens of thousands of customers globally, across what's likely billions of impressions that are closely managed, attributed, and analyzed, all of that data, content, and traffic becomes exactly the substrate a simulation compiler needs to learn from. The ecosystem generates the corpus, the corpus is the best possible place to learn how to build agent-native systems, those systems run the next projects, and the loop closes and compounds. He frames the whole portfolio as the apparatus that produces that corpus.
The strategic frame underneath it: the third door and alpha engineering. Picture the competitor Andy describes: a team of hundreds in perfect coordination, far smarter than you, far better funded, doing this far longer, willing to wait far longer. How do you beat them? The front door and the back door are both locked. You find a third door, and a third door takes creativity. He defines alpha like this: alpha is the thing your competitors are aware of, have probably already tried, have likely even seen the results of, and still won't do, because for their particular structure it doesn't make sense to do it. Alpha engineering is finding what no one else in the market is willing to do, then doing it better than anyone else can, in a way that's scalable, which is the third door. Looikos is a third-door bet: own the means of agent-native content and system generation the way the great consolidators owned their era's leverage.
He names the consolidators on purpose, and each is a different shade of the same lesson. Hearst bought a glut of printing presses and so controlled content generation and distribution, and through it the thinking of populations, and through that politics and business, to the point of literal castles. Ted Turner is the closer comp: he accumulated assets and then improved the efficiency of the advertising across them to produce greater profitability. Conrad Hilton looked at hotels and asked how to wring the most profit per square inch, and was meticulous about it. Rockefeller is the one Andy keeps returning to: he could pick up a flawless 300-page accounting report, the best a senior team could produce, skim it in under five minutes, and calmly hand back the one critical error that every person and every system had missed, sending them back to start over. Andy's answer to "how did he do that" is that Rockefeller wasn't redoing the arithmetic by the book. He was reading multidimensionally, from perspectives others weren't even considering, on a future timescale others weren't comfortable operating in, and he already knew what the unfavorable conditions looked like and the specific measures that led to them, the way a private placement memorandum forces you to enumerate everything that can go wrong in gruesome detail. (The Rockefeller read is the explicit model for the quality-control and strategist posture across the ecosystem: judge from perspectives and timescales others won't occupy, and know the edge cases before they arrive.)