This article is based on expert predictions originally published on jaxenter.com in December 2017. The perspectives reflect the technology landscape as it stood at that time, but many of the underlying trends they identified—DevSecOps integration, multi-cloud adoption, and AI hype cycles—continue to shape DevOps engineering today.
Why 2017 Expert Predictions Still Matter
Predicting technology trends is difficult. For every accurate forecast, there is an over-hyped promise that fails to materialise on schedule. Yet revisiting expert predictions from the industry’s practitioners—not analysts—offers genuine signal about how foundational shifts were taking shape. In late 2017, four practitioners shared their views on what 2018 would bring for DevOps, AI, data science, and cloud infrastructure. Their assessments were grounded, measured, and in many cases prescient.
Patrick McFadin, VP Developer Relations, DataStax
AI Entering the Trough of Disillusionment
McFadin predicted that AI would slide into what Gartner’s Hype Cycle framework calls the “trough of disillusionment.” His reasoning was straightforward: hyperscalers like Facebook and Google made machine learning look deceptively simple, but most enterprises lacked the engineering depth, data infrastructure, and specialist talent to replicate those results. The gap between expectation and execution would become increasingly visible.
Stream Processing, Graph Databases, and Multi-Cloud
Beyond AI, McFadin highlighted several practical data infrastructure trends:
- Stream processing would become more deeply integrated with standard backend databases, reducing the need for entirely separate streaming infrastructure.
- Graph database use cases would move from experimental to production-grade as the tooling matured and practitioners accumulated more operational knowledge.
- Multi-cloud adoption would accelerate, driven by competitive pressure between cloud vendors and enterprise concerns about vendor lock-in.
- Data autonomy would emerge as a strategic priority. Large enterprises would pursue greater control over their data estates rather than concentrating workloads with a single hyperscaler—partly out of competitive concern, partly to avoid dependency on potential market rivals.
Chris Carlson, VP Product Management, Qualys
Security Must Be Built Into the DevOps Pipeline, Not Bolted On
Carlson’s core message was that DevOps was changing the speed of IT delivery faster than security teams were adapting. The consequence: security was being treated as a post-deployment gate rather than an integral phase of the delivery pipeline. His prescription was direct—security teams needed to engage earlier, partnering with development and operations during planning, not after deployment.
He warned specifically against a common failure mode: procuring standalone security point solutions that do not integrate with existing tooling, reporting, or workflows. Rather than improving security posture, this approach creates additional silos and introduces friction that undermines the speed and automation advantages that DevOps is designed to deliver.
The DevSecOps Skills and Mindset Gap
Carlson identified a structural problem limiting DevSecOps adoption in 2018:
- The number of security practitioners with genuine DevSecOps expertise was still low.
- Established endpoint and perimeter security vendors had little commercial incentive to advocate for a model that could displace their products.
- Early-stage startups offering DevOps security tooling lacked the breadth of solution and market presence to gain meaningful enterprise traction.
His recommendation was cultural as much as technical: organisations needed to establish a DevSecOps culture explicitly, because the business impact of insecure software—on revenue and growth—was too significant to treat security as an afterthought.
Threat Landscape Forecast
Carlson predicted that enterprise threats would intensify across ransomware targeting endpoints, organised crime and industrial espionage targeting servers, and application-layer attacks against web applications. He noted that many enterprises still struggled with basic security hygiene—identifying and patching vulnerable systems, locking down misconfigured resources, and detecting active threats already present inside their environments. He specifically anticipated that successful attacks would begin to target enterprise applications running in cloud environments, a shift from the predominantly on-premises attacks that had dominated prior breach disclosures.
Matei Zaharia, Chief Technologist, Databricks
Data as a Competitive Asset
Zaharia predicted a significant expansion in organisations offering curated datasets as commercial products. The model—paying per dataset or per query—was already present in some verticals, but he foresaw it becoming mainstream. Critically, he expected the complexity of data cleaning, processing, and model training to be abstracted away, lowering the barrier for consuming intelligence derived from real-world data.
Growth of Data Science as a Discipline
With data science and computer science among the most popular undergraduate majors at the time, Zaharia expected the practitioner base to keep growing. He was careful to note that the incoming cohort of graduates would not fully satisfy enterprise demand for these roles, but that more organisations would nonetheless begin building formal data science teams and shipping data-driven products.
AI Moving Into Vertical-Specific Business Applications
Rather than expecting generic machine learning platforms to be broadly adopted, Zaharia predicted that AI adoption would be led by vertical-specific solutions—tools purpose-built for common problems in healthcare, finance, retail, or logistics—that incorporated modern ML techniques without requiring organisations to become ML infrastructure experts themselves.
Deep Learning Framework Convergence
At the time, the deep learning ecosystem was fragmented across a large number of frameworks with overlapping capabilities. Zaharia predicted convergence, with interoperability efforts (he referenced ONNX as an example) handling the foundational model definition and serving tasks. Developer attention would then shift upward to higher-level APIs and tooling that made these capabilities accessible for business applications, not just research.
Cloud-Native Data Architectures
Zaharia argued that cloud-native data systems had already moved beyond being simple hosted versions of on-premises databases. He expected new offerings that genuinely exploited cloud properties—elasticity, scalability, managed operations, and availability—to enable data application architectures that were simply not practical to build on-premises.
Christian Beedgen, CTO, Sumo Logic
AI Is Not Ready to Transform the Enterprise
Beedgen offered the most pointed assessment of AI hype. He argued that claims about AI’s near-term enterprise impact were significantly overstated, and that the most cited potential—AI as an objective, unbiased truth machine—was fundamentally flawed. Human bias, whether deliberately or inadvertently encoded into training data and model design, would be reproduced and reinforced by AI systems rather than eliminated.
His position was nuanced rather than dismissive. He acknowledged that specific AI applications—autonomous vehicles being a clear example—were demonstrating real capability. He expected concrete, visible AI use cases to appear in daily life during 2018. His concern was specifically with the broader enterprise transformation narrative. He noted that at the time, only approximately half of the Global 2000 offered fully digital products, suggesting that many large organisations had significant foundational digital work to complete before advanced AI deployment was a realistic priority.
Multi-Cloud and the Demand for Vendor Choice
Beedgen argued that the cloud debate had effectively been resolved: cloud was the present and future of enterprise IT, and organisations continuing to run predominantly on-premises infrastructure would find themselves at a growing competitive disadvantage as cloud-native competitors scaled faster and operated more efficiently.
The more active question, in his view, was which cloud model and which vendors. Enterprise users were demanding flexibility and choice, and this would drive multi-cloud and multi-platform adoption significantly in 2018. The operational implication was important: organisations would need analytics and observability platforms capable of running consistently across any cloud vendor—AWS, Microsoft Azure, Google Cloud—as well as remaining on-premises workloads. This vendor-agnostic model, in his framing, would underpin what he described as an “analytics economy.”
Key Themes Across All Four Predictions
Reading these four perspectives together, several consistent themes emerge that DevOps engineers and platform teams can draw on:
- Security integration is non-negotiable. DevSecOps was identified as an urgent need in 2017, and the challenge of integrating security into fast-moving pipelines without creating friction remains just as relevant today.
- Multi-cloud is driven by strategy, not just cost. Vendor lock-in risk and data autonomy concerns—not only price—were already shaping multi-cloud decisions, and those motivations have only become more prominent.
- AI hype outpaces enterprise readiness. Multiple experts warned independently that AI expectations exceeded what most organisations could operationalise. This caution applies equally to subsequent waves of AI hype.
- Data infrastructure maturity enables everything else. Stream processing integration, graph databases, cloud-native data architectures, and the ability to sell or consume curated datasets all depend on organisations building robust, well-managed data foundations first.
- Convergence reduces complexity over time. Whether in deep learning frameworks or cloud analytics platforms, the direction of travel is toward fewer, more capable abstractions that let practitioners focus on business problems rather than infrastructure primitives.
Original expert predictions sourced from jaxenter.com, December 2017.