Airtable is applying its data-driven design philosophy to AI agents with the launch of Superagent on Tuesday. It is an autonomous search agent that deploys teams of specialized AI agents working in parallel to complete search tasks.
The technical innovation lies in how Superagent’s orchestrator maintains context. Previous agent systems used simple model routing where an intermediary filtered information between models. Airtable’s orchestrator maintains complete visibility into the entire execution journey: the initial plan, execution steps, and sub-agent results. This creates what co-founder Howie Liu calls "a coherent journey" where the orchestrator made all the decisions along the way.
"It ultimately depends on how you exploit the model’s capacity for self-reflection," » Liu told VentureBeat. Liu co-founded Airtable more than a dozen years ago with a cloud-based relational database.
Airtable has built its business on a singular bet: software must adapt to the way people work, and not the other way around. This philosophy has enabled the growth of more than 500,000 organizations, including 80% of the Fortune 100, using its platform to create custom applications tailored to their workflows.
Superagent technology is an evolution of capabilities originally developed by DeepSky (formerly known as Slope), which Airtable acquired in October 2025.
From structured data to free-form agents
Liu sees Airtable and Superagent as complementary form factors that together address different business needs. Airtable provides the structured foundation and Superagent handles the unstructured search tasks.
"We obviously started with a data layer. It’s in the name Airtable: It’s a data table," » said Liu.
The platform has evolved as a scaffold around this core database with workflow capabilities, automations and interfaces scaled to thousands of users. "I think Superagent is a very complementary form factor, very unstructured," » said Liu. "These agents are, by nature, very free-form."
The decision to create free-form features reflects industry learnings about using increasingly powerful models. Liu said that as models have become smarter, the best way to use them is to have fewer restrictions on how they operate.
How Superagent’s multi-agent system works
When a user submits a query, the orchestrator creates a visible plan that divides the complex search into parallel workflows. So for example, if you’re researching a company with a view to investing, it will break that task into different parts of that task, like research the team, research the funding history, research the competitive landscape. Each workflow is delegated to a specialized agent that runs independently. These agents work in parallel, their work being coordinated by the system, each making their contribution to the whole.
Although Airtable describes Superagent as a multi-agent system, it relies on a central orchestrator that schedules, dispatches and monitors subtasks – a more controlled model than fully autonomous agents.
Airtable’s orchestrator maintains complete visibility into the entire execution journey: the initial plan, execution steps, and sub-agent results. This creates what Liu calls "a coherent journey" where the orchestrator made all the decisions along the way. The sub-agent approach aggregates the cleaned results without polluting the context of the main orchestrator. Superagent uses multiple boundary models for different subtasks, including OpenAI, Anthropic, and Google.
This solves two problems: it manages context windows by aggregating the cleaned results without pollution, and it allows adaptation at runtime.
"Maybe he tried to do a search task in a certain way that didn’t work, couldn’t find the right information, then decided to try something else," » said Liu. "He knows he tried the first thing and it didn’t work. So he won’t make the same mistake again."
Why data semantics determine agent performance
From a builder’s perspective, Liu argues that agent performance depends more on the quality of the data structure than on model selection or rapid engineering. It drew on Airtable’s experience building an internal data analytics tool to figure out what works.
Testing of the internal tool revealed that data preparation required more effort than agent configuration.
"We found that the hardest part to do wasn’t actually harnessing the agents, but that most of the special sauce had more to do with massaging the semantics of the data," » said Liu. "Agents really benefit from good data semantics."
The data preparation work focused on three areas: restructuring the data so agents could find the appropriate tables and fields, clarifying what those fields represent, and ensuring agents could reliably use them in queries and analytics.
What businesses need to know
For organizations evaluating multi-agent systems or creating custom implementations, Liu’s experience highlights several technical priorities.
Data architecture precedes agent deployment. Internal experience has shown that companies should expect data preparation to consume more resources than agent configuration. Organizations with unstructured data or poor schema documentation will struggle to ensure agent reliability and accuracy, no matter how sophisticated the model.
Context management is essential. It is not enough to simply assemble different LLMs to create an agent workflow. There must be an appropriate context orchestrator that can manage state and information with a view of the entire workflow.
Relational databases are important. Relational database architecture provides clearer semantics for agent navigation than document stores or unstructured repositories. Organizations that standardize on NoSQL for performance reasons should consider maintaining relational views or schemas for agent consumption.
Orchestration requires planning skills. Just as a relational database has a query planner to optimize results, agent workflows need an orchestration layer that plans and manages results.
"So the bottom line and the short version is that a lot of it comes down to having a really good planning and execution orchestration layer for the agent, and being able to fully leverage the models for what they’re good for," » said Liu.
