By Guest: Johnny Halife
Company: SOUTHWORKS
Show: The Agentic Enterprise
Topic: Agentic AI
Agentic AI is no longer confined to the lab. According to a recent survey of 625 cloud architects and decision-makers conducted by SOUTHWORKS, nearly all respondents see value in AI reasoning, decision optimization, and autonomous agents. Yet most organizations lack the internal capacity to scale these initiatives independently. Governance challenges, fragmented tooling, and unclear ROI are blocking widespread deployment—even as enterprises race to prove AI’s value.
📹 Going on record for 2026? We're recording the TFiR Prediction Series through mid-February. If you have a bold take on where AI Infrastructure, Cloud Native, or Enterprise IT is heading—we want to hear it. [Reserve your slot
Johnny Halife, CTO at SOUTHWORKS, has spent over two decades helping enterprises modernize their infrastructure—from on-premise to cloud, cloud to cloud-native, and now into the age of agentic AI. In a recent conversation, he unpacked what’s driving the shift from experimentation to active execution, where enterprises are finding real ROI, and what’s holding them back from scaling AI at the enterprise level.
SOUTHWORKS partnered with theCUBE Research to survey cloud architects, cloud decision-makers, and Azure professionals to understand the state of agentic AI adoption. The questions weren’t academic—they came directly from the onboarding questionnaire SOUTHWORKS uses with its own customers. “We help companies of all shapes, sizes, and industries,” Halife explained. “But we kept asking: Is this a specific problem for a few clients, or is this happening across the market?”
The findings revealed a messy landscape. Halife categorizes AI adoption into three buckets: automation that doesn’t need AI at all (just API orchestration), personal productivity tools like GitHub Copilot or Microsoft 365 Copilot, and true agentic workflows where autonomous agents reason over data and execute tasks under human supervision. “Some standardization needs to happen,” he said, “because if you use an OpenAI agent to solve a simple automation problem, it might end up too expensive to run.”
The biggest roadblock? Governance. “The number one thing holding back pilots from production is governance,” Halife said. “Who’s paying for that ChatGPT subscription? Are developers using free trials and training public models with PII? What happens when an agent accesses confidential data?” Without enterprise-wide policies, individual users and business units are deploying AI tools in silos—creating fragmentation, compliance risk, and cost inefficiency.
Halife also highlighted real-world use cases delivering fast ROI. “In the IT space, we’ve seen amazing results with troubleshooting,” he said. “You’re running live software accessed by millions of users. You have logs, you have source code. Connect an agent that can help troubleshoot and suggest a fix. What used to take six hours now takes 15 minutes.” That’s not just efficiency—it’s business impact. Reduced downtime, fewer on-call escalations, and faster incident resolution.
But Halife warned against “carbon copying” existing workflows. “Don’t replace a human doing a task with an agent doing the exact same thing. You need to rethink processes for the AI era. Design for MCP, APIs, vectors—don’t just feed your old problem into the new solution.” He compared it to early cloud adoption, when companies tried to replicate their on-premise data centers in the cloud—an approach that often failed because the cost models and architectures were fundamentally different.
Standardization is the next frontier. “Enterprises need a Kubernetes moment for AI,” Halife said. “We need a foundation—something we all trust, that we can build on top of without reinventing the wheel every time.” He pointed to early fragmentation in the AI space, where platform vendors kept their models and integrations proprietary. “Now, with initiatives like Google’s Agent-to-Agent (A2A) protocols, we’re starting to see the light. We need to play better with each other.”
The survey also revealed that most enterprises plan to rely on IT and consulting firms to implement agentic workflows, while also sourcing capabilities from platform vendors. “There’s a white space there,” Halife noted. “Somebody needs to stitch those together. That’s the opportunity.”
Cost remains a critical concern. Enterprises are excited about AI, but they need proof of ROI before committing long-term budgets. “This is not a bubble,” Halife said. “This is an actual change. But you need to show the Intel moment—show me the value, show me the money. Start with big wins, stack them one on top of the other, and build confidence.”
Governance tools exist—least privilege access, data resiliency frameworks, cross-border compliance controls—but they require enterprise-level commitment. “It’s really hard for individual users or business units to standardize at the level we need,” Halife said. “You need company-wide policy and commitment. The guardrails shouldn’t block innovation—they should enable it.”
His final advice for enterprises deploying agentic AI? “Don’t try to automate what you know as-is. Rethink workflows for this new era. Understand the compute power and limitations of AI models. Put the human in the loop at the right moment. If you carbon-copy your old processes, it’ll become too expensive to maintain, and you won’t see the ROI. That’s why most proof-of-concepts fail.”
SOUTHWORKS has been around for 21 years, long enough to see technology hype cycles come and go. Halife believes this time is different. “This is exponential technology that will revolutionize the way we work,” he said. “It’s a fantastic time to be alive.”
