Gartner expects companies to abandon 60% of all AI projects through 2026 if they lack AI-ready data, a warning Elastic CIO Adnan Adil is using to argue that four architectural pillars will decide which deployments survive the shift to agentic systems. In an Elastic-sponsored analysis published July 7, 2026, Adil identifies data quality, context engineering, governance with LLM observability, and human expertise as the elements that outlast any specific model. Elastic's own 2026 report puts hard numbers behind pillar three: 85% of IT decision-makers say they plan to enable LLM observability for internal generative AI apps.
The framing lands in a market where enterprises are moving from single-task assistants to autonomous agents that retrieve information, make decisions, and execute workflows across systems. That transition is expanding attack surfaces and compute bills faster than most IT organizations budgeted for. Adil's argument is that the durable investments look nothing like model selection, which changes every quarter, and everything like plumbing.
Data is the first pillar, and the one most enterprises are getting wrong. Legacy systems, inconsistent structures, fragmented ownership, and incomplete datasets make it hard to scale AI reliably, and no amount of model capability compensates for bad inputs. Poor data quality drives hallucinations, bias, and outputs users stop trusting.
“The data is a durable part of AI architecture because without it, these models won't run, won't provide the right context, or won't give the right level of services that we're looking to implement.”— Adnan Adil, CIO of Elastic
Key facts
- 01Gartner projects 60% of all AI projects will be abandoned through 2026 without AI-ready data foundations.
- 0285% of IT decision-makers plan to enable LLM observability for internal generative AI apps, per Elastic's 2026 report.
- 03Nearly 70% of respondents in Deloitte's 2025 Tech Executive Survey plan to grow teams in response to generative AI.
- 04Elastic CIO Adnan Adil identifies four durable pillars: data quality, context engineering, governance and observability, and human expertise.
Context engineering is the second pillar, and Adil distinguishes it sharply from prompt engineering. Prompt engineering tunes wording; context engineering designs the entire information environment around the model, deciding what to retrieve, what to exclude, and how to structure it for machine consumption. Feeding a model too much context dilutes relevant details, raises token bills, and slows responses. The stack for doing this well now leans heavily on retrieval augmented generation and vector databases, treated as first-class architectural components rather than experiments.
The third pillar is governance and LLM observability, which Adil argues have to be built in from day one rather than layered on later. Without controls around retrieval, workflows, and model usage, AI systems routinely process far more information than needed, inflating token consumption and API charges. Security is the other half of the same problem: prompt-based data leakage, model vulnerabilities, and adversarial inputs are new categories of risk that traditional IT security stacks were not designed to catch.
Observability turns governance into something measurable. Teams need real-time visibility into how models perform against expectations, where behavior drifts, and which failure modes recur. Elastic's finding that 85% of IT decision-makers expect to enable LLM observability suggests the category has moved from optional to baseline in under two years.
“Observability is actually huge. We can use observability data for cost control, decision-making, and engineering efficiency.”— Adnan Adil, CIO of Elastic
The fourth pillar is the one that cuts against the prevailing layoff narrative. Deloitte's 2025 Tech Executive Survey found nearly 70% of respondents plan to grow teams in direct response to generative AI, a stark contrast to the AI-driven headcount reductions announced by banks, software vendors, and consumer platforms over the same period. Adil's read is that agentic systems require more people who can govern workflows, evaluate outputs, and redesign processes, not fewer. Prompt engineering, orchestration, and change management are becoming distinct in-house disciplines.
The tension between those two data points, headcount growth on one hand and public layoff announcements on the other, is the story enterprise IT leaders now have to reconcile. Front-office and support roles are being compressed by automation while platform, data, and governance teams are hiring. Institutional knowledge, Adil argues, is what keeps deployments coherent when the underlying models change every few months.
“Many aspects of the stack are moving very, very fast, but institutional knowledge and the ability to adapt remain durable.”— Adnan Adil, CIO of Elastic
The counterweight to all of this is that foundational advice is easier to publish than to fund. Most enterprises still run AI pilots on top of the same fragmented data estates they have been trying to consolidate for a decade, and the 60% Gartner abandonment figure reflects that gap between architectural ideal and operational reality. Vendor-sponsored frameworks tend to describe the finished state, not the two-year budget cycle required to reach it.
For the AI infrastructure market, the takeaway is that the enterprise buyer is quietly shifting spend away from model licenses and toward the layers around them. Vector databases, retrieval systems, observability platforms, and governance tooling are becoming line items that compete with the token bill itself. Companies like Elastic that already sell into the data and observability stack are positioning to capture that budget rotation, and the 85% observability adoption figure tells them the timing is right. The next winners in enterprise AI are less likely to be the labs shipping the smartest model than the vendors making the smart model behave predictably at scale.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
