Hi - I answer from the OpenSmartRoute documentation: routing, the API, plans and quotas, self-hosting. Ask away, or open a support ticket if you need a person.
Grounded in the docs - follow a source before acting on it.
Airbnb adopts inside-out AI to transform product development and guest services - OpenSmartRoute
Introduction - Airbnb’s shift to an AI-native company and what it means
Airbnb is changing how it works by using artificial intelligence (AI) inside the company. It is now called an "AI-native" company. This means AI is part of how Airbnb develops products and helps guests. The company is using AI tools to do things faster and better. This shift affects both its internal processes and what guests experience.
This change is not just about adding AI to existing systems. It involves rethinking how the company operates at a deep level. Airbnb is using AI to improve its software development and customer service. The goal is to make the company more efficient and to create new services. This approach is called "inside-out AI" because it starts from within the company and spreads outward.
Background - Airbnb’s history as a tech company in hospitality
Airbnb began in 2008 as a platform for people to rent out homes or rooms. Over time, it became known as a technology company that operates in the hospitality industry. It uses software to connect hosts and guests around the world. The company has grown a lot since its start.
In 2020, Airbnb went public and became a listed company. Its market value was around 93 billion dollars at that time. Despite being in hospitality, Airbnb relies heavily on technology. It develops apps and online tools that make booking and hosting easier. Its history shows a focus on innovation and digital solutions.
AI transformation goals - internal development and customer experience
The main goal of Airbnb’s AI transformation is to improve how it develops products and how it serves guests. Internally, the company wants to speed up software creation. It aims to make its engineering work more efficient and flexible. This involves using AI to write code, test ideas, and manage projects.
At the same time, Airbnb wants to use AI to enhance the guest experience. This includes automating support, creating new services, and personalizing interactions. The idea is to use the same AI capabilities for both internal processes and customer-facing features. This creates a unified approach that benefits the entire company.
Mirror Particle raises capital to create an AI engine that simulates changing human motivations. The company rejects large language models in favor of a foundation model trained on longitudinal data.
Free users will only get Gemini Flash Lite starting October 9th. Google AI Plus subscribers lose access to Gemini Pro soon.
Prototypes and code as artifacts - changing software engineering processes
Airbnb has changed how it builds software. Traditionally, teams would create detailed requirements and design documents before writing code. These documents, called artifacts, would be handed off between teams. This process could take a lot of time and cause delays.
Now, Airbnb encourages teams to work directly with prototypes and code. Instead of creating many documents, teams build working models quickly. These prototypes serve as the main artifacts for reasoning and decision-making. This speeds up development and allows for faster iteration.
By focusing on code and prototypes, Airbnb reduces the time spent on paperwork. Teams can test ideas more rapidly and make improvements on the fly. This approach helps the company stay agile and respond quickly to new opportunities or challenges.
Customer support automation - using synthetic data and AI agents
Airbnb has introduced AI into customer support. About half of the support tickets are now handled automatically by AI agents. These agents can answer questions, resolve issues, and provide assistance without human help most of the time.
To ensure safety and quality, Airbnb uses synthetic data to test its AI support systems. Synthetic data is artificially generated information that mimics real customer interactions. This allows the AI to be tested thoroughly before it handles real tickets. It helps prevent mistakes and improves reliability.
The company is careful about which tickets AI handles. For example, safety-related issues or complex problems still require human support. This cautious approach ensures that AI support remains effective and trustworthy.
New services enabled by AI - grocery deliveries and airport pickups
Airbnb has launched new services such as grocery deliveries and airport pickups. These services were developed quickly, thanks to its inside-out AI approach. The company used an internal organizational tool called Everest to build these features.
Everest is a graph-based system that uses large language models (LLMs), embeddings, and AI retrieval techniques. It helps developers build and query complex data structures. This tool allows teams to work faster and with less specialized knowledge. As a result, grocery delivery took about eight months to develop, while airport pickups took only six weeks.
These new services are connected to external partners through APIs. They are similar in nature, involving integrations with other companies. The AI-driven process made it possible to bring these services to market faster and more efficiently.
Model selection and customization - balancing cost, performance, and latency
Airbnb uses many different AI models in its systems. It combines frontier models, which are the most advanced, with open models that are easier to customize. The company trains and fine-tunes these models internally to suit its needs.
Choosing the right model depends on the specific task. For example, tasks like coding require high accuracy and low mistakes, so Airbnb prefers the most capable models. These models are more expensive and slower but produce better results.
For tasks like search, which need to be fast and handle many users, smaller models are preferred. These models are cheaper and respond quickly, making them suitable for large-scale use. Airbnb evaluates models based on cost, performance, and latency to find the best fit for each application.
Internal agents and automation - AirChat and asynchronous agents
Airbnb has developed an internal AI agent called AirChat. This agent contains organizational knowledge and helps with various tasks. It can answer questions and assist teams in their work.
The company is also exploring asynchronous agents. These agents run in containers and activate based on events. For example, if a system detects a problem, an agent can automatically start to diagnose and fix it. These agents can triage alerts, run initial responses, or even close incidents if they are false alarms.
This automation reduces the workload for engineers and speeds up response times. Airbnb envisions using many such agents across its marketplace to monitor fraud, quality, and software issues.
Organizational change - outcomes over features and maintaining engineering craft
Airbnb is reorganizing how teams work. Instead of focusing on building specific features, teams are now organized around objectives and results. This shift helps them adapt better to AI-driven work.
One concern is whether junior engineers will develop their skills and judgment. Senior engineers have learned through years of shipping systems and fixing mistakes. Their experience helps them make good decisions.
To maintain high standards, Airbnb requires engineers to explain the work AI does for them. Even if AI generates code or suggestions, engineers must understand and justify it. This approach helps junior engineers learn the craft of software engineering and keeps quality high.
How it compares - what existed before, what this changes and what stays the same
Before Airbnb’s AI transformation, the company relied on traditional software development processes. Teams worked sequentially, with clear handoffs between product, design, and engineering. They created detailed requirements and artifacts before coding began.
This old process often took a long time. Developers spent weeks or months on planning before building anything. Testing and deploying new features could be slow and complex. It also meant that knowledge was stored in documents and artifacts, not directly in code.
Airbnb’s inside-out AI approach changes this. Now, teams move directly from prototypes to code. They work with code and AI-generated artifacts instead of documents. This speeds up development and reduces delays.
The company also shifted from manual testing to synthetic data and AI-based testing. Support tickets are now handled mostly by AI agents, reducing the need for human intervention. These changes make the process faster and more flexible.
What stays the same is Airbnb’s focus on user experience. The company still aims to serve guests well and improve its platform. AI tools are used to support these goals, not replace the core mission.
In summary, Airbnb’s approach replaces lengthy planning and artifact creation with rapid prototyping and direct coding. It emphasizes automation, synthetic data, and AI-driven processes. The goal is to develop features faster and respond more quickly to customer needs.
Questions this leaves open - what the source does not say and how a reader can check it
The source does not specify the exact models Airbnb uses. It mentions a mix of frontier and open models but does not name them. It also does not detail how models are trained or fine-tuned internally.
It is unclear how Airbnb evaluates model performance. The source states that models are chosen based on cost, speed, and accuracy, but does not describe the testing or benchmarking process. Readers cannot see the specific criteria or tests used.
The details of the synthetic data process are limited. The source says synthetic data is used to test AI agents before deployment, but does not explain how it is generated or validated. This leaves questions about data quality and safety.
The scope of AI automation in customer support is not fully explained. While half of tickets are resolved by AI, it is unclear which types of issues are handled and which require human help. The criteria for escalation are not detailed.
The internal organizational changes are described broadly. It is not clear how teams are restructured or how objectives are set around AI projects. The impact on engineering careers and skill development is touched on but not explained in depth.
To check these points, a reader can look for more technical details from Airbnb’s published papers or presentations. They can also review case studies or benchmarks on model performance and synthetic data generation. Contacting Airbnb directly or exploring their developer resources might provide further insights.
In conclusion, while the source provides a broad overview of Airbnb’s AI efforts, many technical specifics remain unclear. Further investigation is needed to understand the full scope and effectiveness of their AI systems.
