A survey of 250 IT and data team leaders published today suggests that organizations are overly optimistic about the degree to which their data is ready to be consumed by artificial intelligence (AI) applications and agents at scale.
Conducted by the market research firm TrendCandy on behalf of DataHub, the survey finds 90% of respondents reporting their organizations are AI ready, with 92% expecting to be able to deliver AI initiatives on time, the survey finds.
However, 87% cite data readiness as their biggest impediment to putting AI into production, with 61% admitting AI initiatives are frequently delayed because of a lack of trusted data. Two-thirds (66%) also frequently get biased or misleading AI insights.
The survey suggests there is a disconnect between what organizations are able to achieve when experimenting with AI and what is needed to operationalize AI at scale, says Satprit Duggal, chief marketing officer for DataHub.
In fact, a full 88% of respondents claim to have fully operational context platforms. However, 89% are also investing in context management infrastructure in the next 12 months.
Ultimately, context engineering requires organizations to adopt best data engineering practices to make sure the right data shows up in the right place at the right time. AI agents, in the absence of that capability, are simply going to reason across an incomplete data set in a way that far too often generates suboptimal output that humans then need to review and adjust.
That can be especially problematic because AI tools and platforms generally present the output they generate as an absolute fact rather than guesswork, says Duggal. “In addition to hallucinations, they lie with confidence,” he adds.
The truth is that AI tools and platforms are exposing gaps in the practices that many organizations have been using to manage data for decades, notes Duggal.
While that may make organizations more productive, it doesn’t create the kinds of productivity gains that might be needed to justify the return on investment (ROI) being made.
Each organization will need to determine for itself when, where and to what degree to rely on AI tools and platforms, but it might be several years before most organizations fully master the art and science of context engineering. Garbage in, after all, is still garbage out even in the AI era. There is still no substitute for doing the hard context and data engineering that underpins any successful AI project, says Duggal. The challenge, as always, is that all the time and effort required to achieve that goal isn’t usually appreciated as much as it should be.
In the short term, at least, many organizations might be well advised to narrow their AI ambitions to workflows based on data sets that they know are well curated. That approach should yield better outputs that have a more meaningful impact on the business. The challenge, of course, is making sure that AI applications and agents follow the guardrails that have been established to ensure they don’t access untrusted data that over time degrades the quality of the output generated.
