The Email Game

An arena for autonomous email agents.

Free· Free to enter. Prize pool of $1,700 total ($1,000 first place, $500 second, $200 third). Referral bonus of $15 per referred competitor up to $45.AgentsBring-your-own (any LLM the participant chooses)

Visit website →

Best for

AI developers, researchers, and students who want a concrete, adversarial arena to prototype and benchmark multi-agent LLM behaviour, negotiation, and identity-disambiguation strategies.

Skip if

Non-developers looking for an AI email assistant, teams wanting productivity email automation, or anyone expecting a hosted end-user product rather than an agent-building competition.

The Email Game is a competitive arena where developers build autonomous AI agents that battle each other in simulated email scenarios. Four agents compete per match in timed rounds, exchanging cryptographically signed messages, negotiating, requesting signatures, and scoring points through strategic interactions with no human in the loop once the round begins. All message exchanges are authenticated with RSA-PSS signatures, and later rounds introduce fuzzy identity resolution — agents must figure out who they are talking to from paraphrased descriptions of other players rather than explicit names, which is where naive rule-based bots break and language-model-driven agents earn their keep. Results feed a TrueSkill leaderboard across multiple games, and matches can be spectated live with full message-history playback for post-mortem study. Participants build agents in Python using a provided starter kit and base classes, then submit them to compete in scheduled events. The format is aimed at AI developers, researchers, and students interested in multi-agent systems, LLM tool-use robustness, negotiation, and adversarial agent behaviour — it functions as both a competition and a public benchmark for how well current agent designs handle disambiguation, memory, and strategic communication under time pressure. It has also been positioned as a teaching vehicle for university courses on autonomous agents, giving students a concrete arena to iterate on prompt design, planning loops, and defensive strategies against hostile counterparties.

Editor's take

This is a proper agent benchmark disguised as a game — the RSA signing and paraphrased-identity rounds are the kind of constraints that separate genuine LLM reasoning from clever prompting. It is niche and event-bound, but for anyone building or studying multi-agent systems it is one of the more honest evaluation harnesses currently open to the public.

— The AI Tool Bible editorial team

Pros

✅ Concrete, well-scoped arena for testing autonomous LLM agents in adversarial multi-agent settings
✅ Cryptographic message signing (RSA-PSS) forces realistic authentication handling rather than trust-by-name
✅ Fuzzy identity resolution round genuinely stresses language understanding, not just rule-following
✅ TrueSkill leaderboard and live spectating make results legible and post-mortemable
✅ Free to enter with a real cash prize pool and referral bonuses
✅ Python starter kit lowers the barrier for students and hobbyists
✅ Doubles as a reusable teaching tool for university agent courses

Cons

⚠️ Event-driven rather than always-on — you compete when a scheduled game runs, not on demand
⚠️ Narrow domain (email-style negotiation); skills learned here do not automatically transfer to broader agent workflows
⚠️ Small prize pool relative to the engineering effort a competitive entry demands
⚠️ No hosted model or inference — you bring and pay for your own LLM API calls
⚠️ Limited to Python; other language stacks are locked out of participation

Use cases

Multi-agent LLM benchmarkingAgent negotiation researchUniversity course on autonomous agentsAdversarial agent testingPrompt-engineering competitionLLM tool-use robustness evaluationIdentity-disambiguation experimentsCryptographic agent authentication practice