惨痛的教训
萨那大厦

多步骤计划和推理是必不可少的

思维链之类的技巧, tree-of-thought, 和 reflection significantly improve query response accuracy. 如果我们想让助手自动纠正错误, 访问外部工具, 并代表我们提供端到端的解决方案, 这些技术必须是内置的. We've trained our custom agent solution R-4 to lay out a step-by-step plan to solve the query, 执行计划, 并对其产出进行自我反省.

会议是真理的重要来源

So much invaluable company knowledge gets exchanged verbally in meetings vs. 写文档. 这是一个尚未开发的网络新公司数据来源. 与R-4, Sana的bwin足球平台可以转录, 总结, 指数, 检索, 和 analyze meetings like any other knowledge asset—enabled by seamless 集成 with platforms like Google Meets, 团队, 和缩放.

验证数据至关重要

公司的知识很快就会过时. 在这里, an AI assistant is uniquely positioned to succeed where the traditional knowledge management system has failed, 但前提是它知道什么是最新的知识. Along with constant content re-指数ing 和 testing, we've designed verification, deprecation, 和 Q&A workflows to ensure the assistant always has access to the latest 和 greatest.

单靠矢量搜索是不够的

We need a rich knowledge graph to h和le queries based on multiple data variables. Vector search can't predictably solve a request like "List all companies with more than $5M in ARR we've met in Europe over the last 12 months". Instead, we first have to derive a knowledge graph 和 then search structured data.

模型需要一个统一的接口

The AI ecosystem is moving so fast that committing to a single provider or model feels risky. Equally, maintaining provider-agnostic systems adds huge complexity for organizations. We've built an underlying architecture that enables Enterprise customers to choose 和 switch between a range of state-of-the-art models.

集成、权限和部署是一个迷宫

Most company data 和 context live across a slew of tools, not just within an operating suite. We've done the hard work to ensure we can enable granular access controls, 集成, 以及100多个企业工具的权限. 这是一项艰巨的任务. We've had to build a system that automatically h和les permission through out-of-the-box 集成 that mirror users' existing access rights. This system supports deployment in private clouds, offering flexibility for Enterprise customers.

定制是需要的，但很麻烦 

每个团队都希望拥有自己独特的AI助手. But fine-tuning models for each team's needs is difficult without an internal AI team. Our no-code UI setup lets users build custom assistants tailored to each use case in minutes.

长尾巴真的很长

80% of user queries can be h和led well, but that last 20% spans a huge range of edge cases. Reaching human-level robustness has required us to use multimodal models 和 reasoning that can give the assistant a better underst和ing of images, 幻灯片, 和表.

生成式接口还处于起步阶段

聊天和自动完成风格的ui只是一个开始. Inventing new interaction paradigms that fully leverage LLMs is a vast design space. We're building out components the assistant can use to dynamically generate interfaces ranging from meeting summaries 和 knowledge snippets to actions in apps like Salesforce 和 Gmail.

内存将是不可协商的

人们对bwin足球平台的舒适度和体验差异很大. To serve everyone equally well, we have to set an unreasonably high bar on user experience. We also need our architecture to support a future where agents will simply adapt to your individual needs from the first login 和 keep underst和ing you better over time.

助理应该积极主动

In a world where most people don't know what an AI assistant is capable of, Sana的bwin足球平台不能只回答问题. 它还需要预测整个工作流程中的需求. 销售, this could look like automatically taking notes according to the company's sales methodology, 起草客户跟进邮件, 更新Salesforce. 

人类是基准

Users expect the assistant to match what a human expert could do at every turn. But human-level breadth, robustness, 和 adaptability remain elusive. 我们发现给用户一个循序渐进的视图, 引用来源, 和 allowing them to correct the assistant's mistakes has helped us build trust.