Geoffrey Huntley has a particular claim to notoriety in AI circles: he invented the Ralph Wiggum Loop, a memory management technique that earns its name partly by burning through roughly $10.42 an hour in compute costs. Last week he sat down with Vivek Bharathi at Frontier Towers for an 11-minute conversation that ranged well beyond any single technique, covering what he sees as the structural collapse of software as a profession.
His central argument is a split. 'Software development,' in Huntley's framing, is already gone — commoditized, accessible to anyone willing to point a tool like Cursor at a problem. 'Software engineering' is something else: designing agentic loops, building safety systems, managing credentials, sandboxing untrusted execution, thinking seriously about risk. His analogy to locomotive engineers lands well. The job today is not laying track by hand. It is keeping the train on it.
On open source, Huntley doesn't hedge. He calls it effectively dead, and his reasoning is economic rather than ideological. Shared libraries emerged to solve two problems: reducing duplication across organizations and building hiring pipelines around common tooling. AI-generated first-party code makes both rationales obsolete. Why carry a dependency chain — abandoned repos, maintainer burnout, npm package hijacking, Dependabot noise — when a model can produce controlled, bespoke code on demand? He carves out exceptions for security-critical infrastructure such as PKI and SSL, but the broader implication for an ecosystem that has underpinned commercial software for decades is severe.
The same logic extends to competitive moats. Software products are becoming hyper-commodities, Huntley argues — as interchangeable as electricity or shared hosting. Any product can be screenshotted and rebuilt. Auto-migration tooling is steadily dissolving vendor lock-in. He raises the consequence for venture capital directly: if technical differentiation disappears, the standard VC thesis for software investment becomes hard to defend. The moats that survive, he says, are non-technical — contracts, distribution networks, relationships, taste. For practitioners building agentic systems, that is a genuine opening. For everyone else, it is a warning.