Debate at AI Engineer World’s Fair highlights tensions between agentic loops and engineering discipline

A panel at the AI Engineer World’s Fair examined whether autonomous software factories and agentic loops are viable now or if the engineering discipline is lagging behind the ambition. The debate featured advocates arguing that loops are inevitable and already in use, while skeptics warned that hype is outpacing discipline and economic viability.

Advocates like Geoffrey Huntley, creator of the Ralph Loop, and Ian Livingstone, CEO of Keycard, argued that loops are already here and represent a natural evolution of software development. Huntley stated that loops are "inevitable" and that he does not "see myself going back to writing code by hand," while Livingstone emphasized verifiability and the iterative nature of loops in software development.

Skeptics, including Dex Horthy from HumanLayer and Greg Pstrucha from Subroutine, countered that the hype around loops exceeds current capabilities. Horthy argued that "the hype is outrunning the discipline" and that deterministic control loops, like those in Kubernetes, are not yet matched by agentic systems. Pstrucha raised concerns about the economic sustainability of agentic loops, stating that "you can’t orchestrate your problems away by buying more tokens."

The discussion also touched on software factories, a metaphor for highly automated agent environments. Horthy cautioned against automating end-to-end from the start, advocating instead for starting small and building intuition with agent loops. Huntley acknowledged that software factories represent the future but noted that the approach is not yet solved in the market.

Anthropic’s internal model, Claude Tag, was presented as an early example of a software factory model in practice. Mike Krieger, Head of Labs at Anthropic, described Tag as more delegated, asynchronous, and proactive than Claude. He shared how his team uses Tag to delegate responsibilities, such as monitoring feedback channels and proactively taking on tasks, which has changed their operational approach to a multiplayer, async workflow.

However, Krieger also highlighted challenges, noting that his team is "bottlenecked on reviews" and constrained by the "human ability to fully conceptualize what we’re doing."

A 2026 AI engineering survey by Amplify, presented by Barr Yaron, found that 95% of respondents now use agents, nearly double the share from the previous year. Among teams using agents, 89% said those agents could write data, up from 52% the previous year. The survey also revealed that human approvals and permissions remain the leading safeguards, with no settled control layer for agents. Token usage is now the second-most monitored production metric, behind quality, and 40% of respondents said AI costs regularly limit ambitious AI usage.

Closing keynotes at the conference emphasized building AI-native companies rather than treating AI as mere autocomplete. Theo Browne showcased projects that demonstrate the expanded scale of what individual developers can attempt with AI, while Garry Tan, CEO of Y Combinator, argued that the fastest-growing founders are "not treating AI as autocomplete, they’re treating it as a workforce."