Last month, more than 700 engineers and technical leads from companies like Apple, AWS, Google, NVIDIA, Microsoft, OpenAI, Salesforce, and Snowflake registered for a hackathon hosted by popular AI newsletter AlphaSignal. On August 6, 2026 more than 100 of these devs appeared in person in San Francisco to hack on a deceptively simple challenge: in 90 minutes, with no pre-built code allowed, write an AI agent that can order and deliver a pizza to 3 Embarcadero Center. First pizza delivered wins, with $2,500 in total prize money at stake.
It’s a fun premise. But underlying this novelty, the challenge is a good stress test for a much bigger question in AI development: How do you get an autonomous agent to reliably act in the real world, where the information it needs—what’s open, what’s nearby, what’s actually deliverable—isn’t in its training data and changes by the hour?
Of their own accord, two of the winners (Preston Kwei and Rohan Gandotra) leveraged the Brave Search API to solve the same underlying problem: How to ground their agent in accurate, real-time information about the outside world, when that data falls outside their training corpus. Interestingly, each winner took a different approach, which is a useful illustration of how flexible Brave Search API can be as a grounding layer.
Real-time discovery, then action
Preston used the Brave Search API as a live restaurant-discovery layer, letting his agent identify nearby pizza options in real time rather than relying on stale or hardcoded data (a task aided by Brave’s Place Search API). From there, the agent completed the actual order through the Brave browser on DoorDash, pairing real-time search with real-time action in a single pipeline.
Structured, ranked data for agent decision-making
Rohan took a more deterministic approach to the same problem: which restaurants even exist near here, and which one should the agent pick? He used the Brave Search API to pull nearby pizza spots by address, programmatically scored them by distance, and fed that ranked list back to his agent as clean, structured input. This meant no scraping and no manual parsing, just usable data ready for the agent’s decision logic.
Brave Search API is built for just this kind of challenge #
Both Preston’s and Rohan’s approaches map directly onto why developers reach for the Brave Search API when building agents. An agent is only as good as the information it can act on, and static training data can’t tell you which pizza place near the Embarcadero is currently taking orders. That’s the same core problem developers face at larger scales:
- Chatbots that need to ground answers in current events
- Research agents that need fresh market data
- Any AI system that needs a live, verifiable window into the Web rather than a frozen snapshot of it
Whether the goal is answering a question without hallucinating or fetching a ranked list of nearby businesses in real time, the pattern is the same: pull accurate, current, structured data from an independent Web index, and hand it to the agent in a form it can immediately use.
AlphaSignal’s hackathon is a small, playful example that points to a much larger shift already underway: AI agents are moving from answering questions to taking actions, and that shift depends entirely on giving those agents reliable, real-time access to the open Web. A hackathon built around ordering a pizza turned out to be a relevant proxy for the same challenges developers are solving in legal research, market intelligence, fraud detection, and more. It was just compressed into 90 minutes and a countdown clock.
Congratulations to Preston, Rohan, and Elliot S. for winning the night, and thanks to AlphaSignal for bringing the AI community together to find real-world utility for their agents.
Readers’ note: Several members of the Brave team, including Desktop Product Manager Rafael Ebron, Staff Engineer Anton Lazarev, and VP of Privacy & Security Shivan Sahib, attended to observe and answer technical questions, but winners were judged solely on how quickly their agent got a pizza delivered, and whether it was built live during the event. Brave’s attendance had no bearing on the outcome.