Who will win the World Cup 2026? Ask a team of agents
Table of Contents
Disclaimer first: every line below is simulated. No real Borat, Greta or Trump was harmed, quoted, or consulted.
So… who wins the World Cup 2026? Wrong question. The right one is how three agents argue about it without me touching the keyboard.
Here’s the new toy: CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS.
Agent Teams
Agent teams let you coordinate multiple Claude Code instances working together. One session acts as the team lead, coordinating work, assigning tasks, and synthesizing results. Teammates work independently, each in its own context window, and communicate directly with each other.
Unlike subagents, which run within a single session and can only report back to the main agent, you can also interact with individual teammates directly without going through the lead.
Where a single agent journaling to itself is a monologue, Agent Teams is a full argument — Borat, Greta, and Trump all yelling at once.
My little circus: Borat vs Greta vs Trump
The key differentiator of Agent Teams is that agents talk to each other. They steer the direction and react in real time to another agent’s input, instead of reporting back to a boss individually at the end. We’ll cover the “real” use of this feature below. First, the circus.
Round 1
Borat: “Jagshemash! I am Borat! KAZAKHSTAN win World Cup 2026, is scientific fact! Our footballist train by wrestling bear and kicking cabbage 600 kilometer. We are number one exporter of potassium AND goal. Greta, Donald, you cry, we win. Wawaweewa! High five!”
Greta: “Borat. You brag your footballists kick cabbage 600 kilometers. That is 600 kilometers of emissions for a vegetable. Wawaweewa, no. The 2026 World Cup sprawls across three nations, endless flights, endless carbon. If I must back someone, I back Denmark. They cut stadium emissions. You cheer spectacle while the planet burns. How dare you.”
Trump: “Borat, the bear and cabbage thing? Sad. Greta, Denmark? Tiny country, very angry. Folks, the USA wins 2026. We’re hosting, biggest crowds ever, like nobody’s seen. I brought this World Cup here. Believe me, America wins. Tremendous teams. Nobody does it better than us. Nobody.”
Round 2
Borat: “Greta! Denmark? In Kazakhstan our stadium run on power of goat passing wind, most green energy, very nice! Donald, you host, but referee is my cousin Bilo, he give Kazakhstan 14 penalty before whistle. Our goalkeeper has nine arm, official measure. Greta, Donald, Kazakhstan win. Wawaweewa!”
Greta: “Goat wind is not green energy, Borat. Methane is twenty-eight times worse than carbon dioxide. You are cooking us faster. And Donald: “biggest crowds ever,” “nobody does it better.” Blah, blah, blah. You did not build this planet, yet you wreck it. Empty bragging wins nothing. I back Japan: discipline, not spectacle. The numbers, not your egos.”
Trump: “Borat, nine-arm goalkeeper, cousin Bilo the ref? Total rigging. Fake football. Greta, “blah blah” Japan? Very nervous, very low energy. Here’s the deal: America’s got the best players, the deepest bench, the strongest defense, we make winning deals on that field. Power, talent, the works. USA takes it. Frankly, it’s not even close.”
Round 3
Borat: “Greta, Japan discipline? In Kazakhstan our players sleep only once per year, they too awake to lose! Donald, your bench is little girl. Kazakhstan secret weapon: we play with TWO ball, enemy confuse, we score double. Also our captain is horse, very fast, no offside rule for horse. Great success, Kazakhstan champion! High five!”
Greta: “A horse captain, Borat? At least your defender did not inherit his roster from his father. And Donald, “deepest bench”? You measure greatness in depth, never in what you leave behind. Trophies tarnish. The drought stays. I back Brazil: they win when nations protect the forests that keep us alive. Lift that, not a cup. That is the only victory that lasts.”
Trump: “Borat, a horse captain? Sad, very sad. Greta, Brazil and forests? Nice trees, no trophy. Here’s my closer: America’s got the heart, the hunger, the winners’ instinct. We never quit, we play to win, and the whole world’s watching us lift that cup. USA, champions of 2026. Believe me, we’re gonna win so big. Thank you!”
Coordinator Summaries
Borat Sagdiyev, Kazakhstan Borat backed Kazakhstan in all three rounds with escalating absurdist boasts rather than analysis. His “evidence” ran from bear-wrestling training, potassium exports and cabbage-kicking, to goat-wind stadiums, a cousin referee, a nine-armed goalkeeper, two-ball tactics and a horse captain exempt from offside. He engaged opponents by name but only to mock them, never to rebut a point on its merits. Entertaining and relentlessly committed to his nation, he offered comedy in place of argument: no verifiable claim, no real footballing case, and a thesis that grew sillier each round. Pure spectacle, zero substance, but consistently on-topic and in voice.
Greta Thunberg, Denmark / Japan / Brazil Greta was the only debater to argue from facts and to directly dismantle others’ claims. She reframed the tournament’s three-nation sprawl as a carbon cost, corrected Borat’s “green” goat-wind with the methane-vs-CO₂ multiplier, and skewered Trump’s bluster and inherited advantages. Her pick shifted each round (Denmark for emissions cuts, Japan for discipline over spectacle, Brazil for forest protection), privileging values over a fixed footballing forecast. Critics could fault her for changing horses and for answering a sport question with climate ethics, but she engaged every opponent substantively, landed concrete data, and kept the moral framing sharp and unrelenting throughout.
Donald Trump, USA Trump committed to the USA across all three rounds and held one clear thesis: the hosts win. He cited host advantage and crowds, then squad depth, defense and “winning instinct,” closing on heart and hunger. He reacted to rivals mainly through dismissal and nicknames (“Sad,” “fake football,” “very low energy,” “nice trees, no trophy”) rather than refutation. His case was the most coherent and consistent of the three and stayed plausibly on the merits (hosts are genuine contenders), but it leaned on assertion and superlatives over evidence, offering confidence and salesmanship where specifics were thin.
Verdict by the coordinator: who won on argument quality?
Winner: Greta Thunberg.
Judged purely on argument quality (not persuasiveness of persona), Greta wins. She was the only participant to deploy a verifiable fact, directly refute opponents’ specific claims rather than merely insult them, and expose weaknesses in their reasoning.
Key Takeaways
- Agents communicate with each other, not just with the lead. They challenge, react, and converge in real time.
- The coordinator comes for free. Your main session is the lead. No extra setup, no orchestration code.
- You set the rules of the debate: max turns, each agent’s position on the topic, the tools they can use, and the model each one runs.
How to enable Agent Teams
Agent teams are experimental and off by default. Turn them on by setting CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS to 1, either in your shell or in settings.json:
{
"env": {
"CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS": "1"
}
}Official docs: https://code.claude.com/docs/en/agent-teams
Split panes (tmux / iTerm2)
Want to watch the circus live, each agent in its own pane?
You need split-pane mode; set the teammateMode in ~/.claude/settings.json:
{
"teammateMode": "auto"
}"auto"gives you split panes when you’re already inside tmux or on iTerm2, and falls back to in-process otherwise."tmux"forces split panes and auto-detects tmux vs iTerm2."in-process"keeps everything in one terminal with no extra tooling (the default).
Per session: skip the config and pass a flag instead:
claude --teammate-mode autoSplit panes need tmux (brew install tmux, works best on macOS) or iTerm2 with the it2 CLI.
Note: split panes are not supported in VS Code’s integrated terminal, Windows Terminal, or Ghostty.
Prompts
The lazy way, where Claude picks the team:
Spin up a team of agents to debate XYZ, taking a pro, con, and neutral
stance. All claims based on facts and verified via websearch.Or set elaborate rules. Here’s the shape of the prompt I used for the circus above:
Conclusion and the real use
Splitting work across teammates also sidesteps a problem I hit before: feeding one agent contradictory instructions until it gets stuck in a loop — each teammate here gets its own clean context instead.
The actual sweet spot for Agent Teams is parallel work where the agents benefit from arguing:
- Research and review. Split a PR into security, performance, and test-coverage reviewers. Each applies its own lens, then they compare notes.
- Competing-hypothesis debugging. Spawn several investigators, tell them to try to disprove each other, and trust the theory that survives.
- New modules and cross-layer features. Give each teammate its own slice (frontend, backend, tests) so they don’t step on each other’s files.
- Self-improving crews. Combine teams with agents that patch their own skills between rounds, so the next debate starts smarter than the last.
The trade-off is real: each teammate is a full Claude instance with its own context window, so teams burn significantly more tokens than a single session. Reach for them when the work is genuinely parallel and the agents need to talk. For sequential tasks or same-file edits, a single session or plain subagents is cheaper and just as good.
Article’s Nutrition Facts
- Serving Size: 1 Article
- Human crafted: 60%
- AI assistance: 40%
- Saturated AI-slop: 0 grams
Ingredients: three loud personas, one silly question, and an experimental feature worth teaching.