ICFP 2026 Contest Allows AI Agents and Introduces Spatial Language

Organizer Nolen Royalty wrote that the team wanted to avoid a competition where the strongest strategy was simply running an unmonitored coding agent for three days. Their response was to build the challenge around a new language called littleman.

Littleman programs are two-dimensional. Instead of executing a conventional linear instruction stream, small actors move across instructions placed in a spatial layout and exchange data through pipes drawn into the program. The organizers also published an open textbook explaining the language.

The design does not ban automated coding tools. It changes the problem around them: contestants may use AI agents, but they still need to reason about a novel execution model, spatial structure, and scoring behavior. The official contest rules keep the event open to teams of any size and require prize-eligible teams to submit their source code when the contest ends.

The contest includes a 24-hour lightning round followed by the full 72-hour event. Winners will be announced at ICFP 2026, the ACM SIGPLAN conference on functional programming scheduled for August 24-29 in Indianapolis.

For AI and software-engineering teams, the notable part is the evaluation design. Rather than treating agent use as misconduct, the organizers permit it and try to create a task where tool orchestration alone is not sufficient. That makes the contest a practical example of how assessments may adapt when participants have access to increasingly capable coding agents.

The event is not a controlled benchmark of human versus AI performance, and results were not available when this article was prepared. The useful evidence will come after the contest: submitted approaches, score distributions, and whether successful teams relied on autonomous agents, human-designed solvers, or a combination of both.