AI Lowered the Floor, AI Raised the Ceiling

AI lowered the floor and raised the ceiling. The technical moat at the median of the hackathon ecosystem has collapsed; the differentiator at the top has gotten harder, not easier, to clear. Judging criteria that worked when "ship a working prototype in a weekend" was itself the main filter no longer separate strong teams from weak ones, because every team can now ship a working prototype.

The collapse of the floor is well-documented in the contemporary record. CNBC's October 2025 coverage of a Singapore vibe-coding hackathon reported that some participants who had learned the tooling only two weeks before the event placed high in the rankings, beating engineers with years of conventional experience. Cognizant's August 2025 internal hackathon set a Guinness World Record with more than fifty thousand employees producing more than thirty thousand prototypes in ten days, a scale that would have been unimaginable as recently as 2023. Across Claude Code, Cursor, Lovable, v0, Bolt, Replit, Kiro, and the broader generation of AI coding tools that emerged between 2023 and 2026, the technical barrier to "ship a working app in a hackathon" has fallen close to zero. The implication for organizers is uncomfortable but unambiguous: a rubric that rewards "the team shipped working code" rewards the median submission. It does not separate signal from noise.

But the moat at the top has gotten higher, not lower, and several voices have named the precise correction. Sherry Jiang, a Peek co-founder and the organizer of the Singapore vibe-coding event, named the precise correction in CNBC's coverage: "We've lowered the barrier, but raised the bar." Her broader framing in the same interview — that participants with product sense, taste, and the ability to position their work are starting to outperform engineers because the engineering itself is easier now — generalizes the shift the rest of this site builds on. Long Ren, an AWS engineer who both judged and competed in back-to-back AI hackathons in 2025, made the observation more concrete — with most teams using similar foundation models, the actual differentiator became user interface design, time-to-first-token, and how visibly the agent's reasoning was surfaced to the user. The lablab.ai strategy guide for 2025 hackathons states the corollary in plain terms: a generic wrapper around OpenAI is no longer impressive. None of these voices argue that AI tools should be banned or restricted. They observe that the dimensions on which work used to be evaluated — did it ship, did it work — have collapsed into baseline expectations, and that the dimensions worth evaluating now sit higher in the stack.

The major hackathon platforms have begun adapting their rules to where the moat actually lives. ETHGlobal's submission rules declare that work which relies entirely on AI without meaningful contributions from team members is ineligible for partner prizes or finalist consideration. MLH's standard rules recommend crediting tools used and clarify that submissions should not be reskins of an existing AI tool — focusing on what was created, changed, or built during the weekend rather than what was merely used. The rules do not ban AI. They regulate attribution, because attribution is now where evaluation actually has to happen.

The familiar anti-example is the wrapper team — the submission that is essentially a thin layer over a frontier model with a different color scheme, a polished landing page, and a problem statement the model itself could have generated. The work is real, in the sense that something was built. The differentiation is non-existent. A rubric that does not surface this confuses judges, disappoints participants who did the harder work, and produces winners whose victory is structurally indistinguishable from luck.

The implication for hackathon judging is direct, and the rest of this site builds on it. Taste in problem selection, judgment about which sub-problem is worth solving, system design choices that survive contact with users, UX craft that respects attention, and demo storytelling that surfaces the work without overstating it — these are the differentiators now, and the rubrics, format choices, and integrity mechanics covered elsewhere on this site are load-bearing in a way they were not five years ago. The format taxonomy format-taxonomy catalogues the working hackathon archetypes and how each is shifting under AI. The storytelling principle storytelling-is-not-optional argues, against the hacker-culture instinct that the work should speak for itself, that demo craft is part of the work. The disclosure regimes no-ringers-without-disclosure cover how attribution and class labeling enforce fairness when conventional gatekeeping no longer can. The Cognizant case study cognizant-vibe-coding documents what AI-era scale looks like in practice, and the voices sherry-jiang and long-ren profile the practitioners whose framing has done the most to clarify what changed.

The floor has fallen out from under conventional hackathon design. The ceiling is where the work now lives.