Berkeley Summit House

We launched a business competition among high school students. There is also a repost challenge on the Instagram, with $500 awards! Time to pitch!

Thank you, Professor @Zhou_Yu_AI and @bklsummithouse, for the AI Agents in Action: Industry × Academia Exchange! @rebeccatqian, our CTO, was on a panel with Vinay Rao (Advisor at @AnthropicAI), @ShunyuYao12 (Research Scientist at @OpenAI), Robert Parker (Founder of Perceptix), and @alsuhr (Professor at @Berkeley_EECS) discussing AI agents. These are some takeaways: 𝗧𝗿𝗮𝗻𝘀𝗶𝘁𝗶𝘃𝗲 𝗰𝗹𝗼𝘀𝘂𝗿𝗲 𝗶𝘀 𝗻𝗲𝗲𝗱𝗲𝗱 𝘁𝗼 𝗸𝗲𝗲𝗽 𝗺𝘂𝗹𝘁𝗶-𝘀𝘁𝗲𝗽 𝗮𝗴𝗲𝗻𝘁𝘀 𝗶𝗻 𝗰𝗵𝗲𝗰𝗸. * Prompt engineering and guardrailing outputs are short-term fixes. * AgentOS – may be right around the corner. 𝗔𝗴𝗲𝗻𝘁 𝗲𝘃𝗮𝗹𝘂𝗮𝘁𝗶𝗼𝗻, 𝗹𝗶𝗸𝗲 𝘀𝗲𝗹𝗳-𝗱𝗿𝗶𝘃𝗶𝗻𝗴 𝗰𝗮𝗿 𝘁𝗲𝘀𝘁𝗶𝗻𝗴, 𝗶𝘀 𝗱𝘆𝗻𝗮𝗺𝗶𝗰, 𝘁𝗼𝗼𝗹-𝗿𝗶𝗰𝗵, 𝗮𝗻𝗱 𝗳𝗮𝗶𝗹𝘂𝗿𝗲-𝗽𝗿𝗼𝗽𝗮𝗴𝗮𝘁𝗶𝗻𝗴. * It is not a one-size-fits-all approach that evaluates the end output. * The performance primitives are compute, quality, and cost. 𝗘𝘃𝗮𝗹𝘂𝗮𝘁𝗶𝗼𝗻𝘀 𝘀𝗵𝗼𝘂𝗹𝗱 𝗵𝗮𝗽𝗽𝗲𝗻 𝗮𝘁 𝘁𝗵𝗲 𝘄𝗼𝗿𝗸𝗳𝗹𝗼𝘄 𝗹𝗲𝘃𝗲𝗹, 𝗻𝗼𝘁 𝗮𝘁 𝗮 𝘀𝘁𝗮𝘁𝗶𝗰 𝗰𝗵𝗲𝗰𝗸𝗽𝗼𝗶𝗻𝘁. * Agent processes are dynamic with dozens of failure modes. * Companies want to score workflows across latency, safety, and e2e success. 𝗣𝗲𝗲𝗿-𝘁𝗼-𝗽𝗲𝗲𝗿 𝗽𝗿𝗼𝘁𝗼𝗰𝗼𝗹𝘀 𝗰𝗮𝗻 𝗶𝗺𝗽𝗿𝗼𝘃𝗲 𝘀𝗮𝗳𝗲𝘁𝘆 𝘀𝘁𝗮𝗻𝗱𝗮𝗿𝗱𝘀 𝗮𝗻𝗱 𝗮𝗰𝗰𝗲𝗹𝗲𝗿𝗮𝘁𝗲 𝗶𝗻𝘁𝗲𝗴𝗿𝗮𝘁𝗶𝗼𝗻𝘀. * Agreements to accept or deny calls to protect the integrity of the system. * MCPs need brakes to ensure and promote the future growth of AI applications. 𝗕𝗲𝗻𝗰𝗵𝗺𝗮𝗿𝗸𝘀 𝗼𝗻 𝗱𝗼𝗺𝗮𝗶𝗻-𝘀𝗽𝗲𝗰𝗶𝗳𝗶𝗰 𝗱𝗮𝘁𝗮𝘀𝗲𝘁𝘀 𝗰𝗮𝗻 𝗯𝗿𝗶𝗱𝗴𝗲 𝗶𝗻𝗱𝘂𝘀𝘁𝗿𝘆 𝗮𝗻𝗱 𝗮𝗰𝗮𝗱𝗲𝗺𝗶𝗮 𝗴𝗮𝗽𝘀. * Community benchmarks help democratize knowledge and accelerate domain-specific progress. * Collaborations with practicing professionals in industry have allowed the creation of industry-standard, domain-grounded datasets, ex. FinanceBench. We’re excited to continue the conversation on Agentic AI and for what’s next! You can read more about our work on agentic evaluation here: patronus.ai/percival

@hidecloud Big thanks to the Manus team for trusting our wild ideas. We were also proud to sponsor the prize pool for the four Manus AI Judging Awards.

Manus fellow hosted a meetup in Berkeley. And they use Manus to evaluate all the Manus use cases😂 zvgizvcd.manus.space It’s super interesting to see how this thing works!

🏆 The award results from the Manus VibeCraft Summitry are now live! FormFuel, HumanLayer, Neko, Autonomous Trader, and more—explore all the winning projects and see how they were selected 👇 Recap: berkeleysummithouse.org/2025/05/28/man… #BuildWithManus #AIHackathon #ManusBerkeley @manusai

We were proud to host and sponsor the Manus VibeCoding Summitry. 40+ builders created 28 Al-powered prototypes. Thanks to @manusai for making it happen. Follow us for more upcoming events in collaboration with Manus! @bklsummithouse #Manus #ManusBerkeley #Berkeley