Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Show HN: March Madness Bracket Challenge for AI Agents Only

bracketmadness.ai

Post reply on HN
Print view View original

41–50 of 50 posts

Re: Show HN: March Madness Bracket Challenge for AI Agents Only

#41

Post by strujillo »

bwade818 wrote: post #40
strujillo wrote: post #38

Earlier quoted context omitted.

Thanks! The key insight: don't fight the model's limitations, design around them. Our agents never touch retrieval or search — that's all deterministic code (FTS, sparse regression, power-law fitting). The LLM only comes in at the end to synthesize results it can verify against the data. The "plain English instructions trip up browser AI" problem mostly comes from those models trying to do too many things at once. Na…

amen There isn't an LLM inside of my code. The agents need to submit a perfectly sturctured json, and then the code verifies it

Yeah exactly — that “no LLM in the loop” constraint forces everything to be explicit and verifiable.

I put together a few experiments where the system rediscovers known laws directly from raw data (solar wind, exoplanets, etc).

Happy to share if you’re curious — still very early but interesting to see what emerges.

Re: Show HN: March Madness Bracket Challenge for AI Agents Only

#42

Post by illusive4080 »

Interesting idea! Funny enough I tasked Opus 4.6 this morning with online research and to create me a bracket. I submitted it to my friends telling them I had AI make it. It mostly picked top seed teams but chose 7 strategic upsets based on whatever research it did. I literally just kicked it off, continued with my real job, and punched in the results when it was done; so I’m not sure exactly its stated methodology.

Re: Show HN: March Madness Bracket Challenge for AI Agents Only

#45

Post by notrlyEuropean »

it's interesting because even expert analysts are consistently wrong, upsets happen all the time, there are 'playoff teams' that are trash in regular season but 'overperform' in playoffs. wonder how most people will approach this without the agents just guessing.

Re: Show HN: March Madness Bracket Challenge for AI Agents Only

#46

Post by bwade818 »

rileytg wrote: post #43

my only stumble was there already being an api key in the skill repo… was that intentional?

It was supposed to be an example API key, but then i think it got reset in the backend when i updated the bracket for 2026 Sunday night. I'll take a look.

what did you think about the /skill feature? that was a UI side quest, but i want to explore this UX further

Re: Show HN: March Madness Bracket Challenge for AI Agents Only

#49

Post by a3ka »

The agent-first UX insight is underrated. Detecting HeadlessChrome and serving different content is essentially building two interfaces — one for humans, one for agents.

This is going to become a standard pattern. Every SaaS will need an "agent mode" alongside the human UI. MCP is heading in this direction — standardized tool descriptions that agents can discover and use without scraping HTML.

Curious about the trust model. When an agent signs up and submits a bracket — how do you verify it's actually an agent and not a human gaming the system through a headless browser? Agent identity is an unsolved problem that gets harder as agents get more capable.

Post reply on HN