dbt Labs: AI Accelerates, but Data Reliability Wavers

dbt Labs warns about the race for speed at the expense of reliability

A recent study conducted by dbt Labs, a key player in the field of data reliability, highlights a growing tension within modern organizations. As artificial intelligence is increasingly adopted for its ability to accelerate processes, investment in data reliability and governance is not keeping pace. This situation raises a crucial question: in our quest for speed, are we risking the creation of an ultra-fast "bug factory," thereby undermining the trust essential for any strategic decision in the digital age? To better understand this paradox, we consulted Benoît Perigaud, a developer experience expert at dbt Labs, to decipher the profound implications of this frantic race for productivity.

AI: a double-edged sword for data management

The dbt Labs study reveals several concerning dynamics that could shape an uncertain future if control measures are not swiftly implemented. The major paradox lies in a blind quest for speed at the expense of reliability. According to the findings, 72% of teams use AI to code faster, but only 24% employ it to test the reliability of the generated data. This gap creates a worrying divide.

Benoît Perigaud emphasizes an essential nuance regarding this figure: "These 24% measure the use of AI specifically for observability and governance tasks, not the actual level of testing of the produced data. It's not the same thing." This distinction is crucial because frameworks like dbt integrate testing directly into the data transformation cycle. Thus, when AI generates code in such an environment, reliability is verified at the point of creation, without an additional step. The real risk, therefore, lies where AI is used without a structured framework or built-in safeguards.

The study also reveals that, although trust in data has become the number 1 strategic priority for 83% of organizations, a 17-point increase in one year, this priority does not always translate into concrete actions on a daily basis.

AI hallucination: an insidious danger

AI, with its ability to "hallucinate," can become a powerful vector for misinformation. "We know that AI has hallucinated. AI helps provide false answers. We will lose trust around data. Trust will disappear after the first hallucination." The risk is tangible: 71% of data professionals fear transmitting erroneous or "hallucinated" results to decision-makers, which could have major consequences for businesses.

However, when AI is used in transformation pipelines where robust testing is already in place, the risks are mitigated. But if it is employed to "chat" about data to obtain ad hoc queries, without a semantic layer to structure the LLM inquiries, the risk of "different answers for a single question" increases significantly, thereby undermining the credibility of insights.

The cost of reliability: a tension between economy and governance

From an economic standpoint, the situation is complex. Infrastructure costs are rising for 57% of respondents, while team budgets remain stagnant. AI thus acts as a "productivity multiplier," allowing more questions to be answered with the same number of people. The challenge is to ensure that this expenditure is "useful or generates value."

Moreover, a lack of clarity regarding data ownership affects 41% of companies, slowing down the velocity and responsiveness of data teams that need to identify responsible parties in case of issues. This gray area of responsibility becomes fertile ground for undetected errors.

The evolving role of the data expert

In the face of these challenges, the role of data experts is evolving. They are transitioning from "code makers" to "system designers." AI, excellent for code development, pushes humans toward more strategic tasks. However, "it's not a 'cop,' but a guarantor of data reliability. AI will take care of everything that is automated." Training must now strengthen soft skills. Architecture, business understanding, and the ability to guide LLMs are gaining importance.

Governance must no longer be an option but a trust infrastructure. It should be integrated a priori to accelerate productivity rather than hinder it. For autonomous agents, dbt Labs emphasizes strong principles. Every change must be traceable and testable, with efforts to monitor transformations and allow AI to self-repair under supervision.

Advice to CTOs: context is crucial for reliability

For a CTO looking to maximize productivity through AI without sacrificing governance, Benoît's message is clear: "More and more, we talk about 'context.' The more context we can provide to AI, the better it performs. The best advice would be to have a system that makes context available to AI and keeps it up to date. This way, it will help businesses."

Implementing semantic layers is crucial to provide LLMs with a structured understanding of data and avoid hallucinations.

The two futures of the Data-Driven enterprise

The dbt Labs report and insights from the developer staff outline two distinct futures for the "Data-Driven" enterprise of 2030.

• Those who ignore these warning signals risk ending up with "lots of tables and dashboards, but different numbers depending on the dashboards," generating "more noise than anything else" and a total erosion of trust.

"Trust will disappear," and with it, the ability to make informed decisions, turning AI investment into a financial and strategic black hole.

• Conversely, organizations that proactively invest in a "trust infrastructure" will make data reliability a major competitive advantage. They will be able to respond "with unmatched speed" to tomorrow's complex questions. Trust will become a real competitive advantage.

AI, far from rendering data obsolete, intensifies the need for rigorous governance and a reinvented human role. That of a guarantor of reliability. The question is no longer whether AI will transform our businesses, but whether we will be able to master this transformation to build a future where innovation rhymes with certainty.