dbt founders outline an open infrastructure strategy for AI agents

Agents need more than intelligence

The founders describe the current technology shift as larger than previous changes involving PCs, the internet, mobile devices, social platforms and cloud computing. In their view, it will affect knowledge work, the role of firms and the part humans play in production.

They define an agent as intelligence combined with tools and data. Such systems can have goals, ask and answer questions, and perform actions. The letter says agents are already effective in knowledge-work domains where context is readily available and results can be verified, with software engineering offered as a strong example.

Other tasks require far more organizational context. A contract review may need the contract, redline history, email exchanges and contracting guidelines. Investigating falling billings at a healthcare provider could require structured data as well as departmental OKRs, meeting notes and information about current initiatives. The founders say the complexity of this kind of analysis often exceeds the time available to human analysts.

Building an agent stack on company data

According to the letter, organizations that deploy agents have unlocked 50x more question and answer loops because agents can work continuously and be available immediately to more people. The founders say the main requirement is giving agents access to the data they need.

Agents can use existing infrastructure, including compute, networking, programming languages and APIs. Data infrastructure built over decades can provide similar foundations, but the founders say important parts of the required agent stack are still early or do not yet exist. They identify evaluations and background memory formation as examples.

The proposed stack would provide agents with a lasting, shared picture of a business. The founders argue that agents should learn from the systems where business activity occurs, rather than relying mainly on documents, chats and email. Those systems can provide definitions, logic and evaluations that help agents understand what data means and make decisions with greater confidence.

From AI assistance to autonomous processes

The letter presents augmentation as the near-term direction, with agents helping knowledge workers get more done. It also describes a move toward autonomous agents that take ownership of business processes from end to end.

Customer support illustrates the distinction. Instead of using AI only to reduce the number of human support agents, an autonomous process could resolve a ticket, monitor the customer’s account afterwards, make contact if the account is not in a good state, and send software issues to an engineering agent for broader fixes. The founders say this approach can improve both customer outcomes and the underlying system.

To support this transition, the letter identifies “agent engineers” as a future role. Their responsibilities include redesigning processes around agents, connecting agents to tools and data, defining success measures, monitoring behavior, diagnosing failures, improving results, securing organizational support and managing changes to human responsibilities.

Why open systems and people remain central

The founders warn that businesses could become strategically vulnerable if their autonomous agents depend on a single technology vendor. They argue that companies should own their data and governance and be able to change models and compute systems. For that reason, the letter says open source, open standards and cross-platform portability matter increasingly, and states that the Fivetran platform will be built around Open Data Infrastructure principles.

The founders also say analytics agents are already here while autonomous agents are still coming. They connect that direction to Fivetran’s “Stay Human” value, arguing that empathy, goal-setting and innovation become more valuable as AI systems grow more capable. The letter closes by presenting dbt as a place for analytics engineers to build scalable data infrastructure, and promotes dbt Wizard CLI, an agent for analytics engineering that knows which tool to call, what context to pull and how to check its work.