Hud CEO Roee Adler says runtime intelligence will define the next era of software operations

Hud CEO Roee Adler argues that the rise of AI-assisted development has exposed the limitations of traditional observability platforms, which were designed for human operators rather than autonomous coding agents. He proposes 'runtime intelligence' as a new approach to provide structured, function-level context for AI systems, enabling them to operate confidently in production environments.

Artificial intelligence has dramatically accelerated software development, with coding agents https://thenextweb.com/news/openai-codex-agents-shift-employees-non-developers now capable of producing large volumes of production-ready code in minutes. Yet while writing software has become faster, ensuring that software behaves correctly in production remains one of engineering’s biggest challenges. For decades, observability https://thenextweb.com/news/tsuga-35-million-series-a-observability platforms have helped teams monitor infrastructure through logs, metrics, and traces. But according to Roee Adler, CEO of Hud http://hud.io/ , the rapid rise of AI-assisted development https://thenextweb.com/news/cursor-anysphere-2-billion-funding-50-billion-valuation-ai-coding is exposing the limitations of systems originally designed for human operators rather than autonomous coding agents. Software Is Being Written for a Different Era Adler believes the shift isn’t simply about adopting AI tools. Instead, it’s about recognizing that software development itself is undergoing a fundamental transformation. As he explains, “ The fundamental change is that software is no longer written primarily by humans using the engineering processes we’ve refined over decades; it’s being written by coding agents who are a completely different species. ” Those agents, he says, are “ fast, impatient, aggressive, ” but they also lack the production context needed to consistently make safe decisions in complex environments. Rather than slowing AI adoption, Adler believes the industry should build the infrastructure that enables coding agents to become trustworthy collaborators. In his view, the goal isn’t to replace engineers, but to equip AI with enough real-world understanding to operate confidently in production. Faster Development Has Created a New Bottleneck Although AI is dramatically increasing the amount of code developers can produce, Adler argues that organizations aren’t seeing the same acceleration in overall software delivery https://thenextweb.com/news/gitlab-19-intelligent-orchestration-agentic-devops . The problem, he says, isn’t that engineering teams suddenly face new operational blind spots. Instead, existing review and validation processes weren’t designed for the pace of AI-generated development. As more code reaches production, confidence becomes harder to maintain. Adler notes that while individual engineers are becoming more productive, “ the bottleneck shifted to the gate, which was trying to prevent bad changes from breaking the system. ” He believes engineering organizations now face two pressing questions: how to review an ever-growing volume of AI-generated code while ensuring business intent is preserved, and how to preserve institutional knowledge as fewer engineers understand every corner of increasingly AI-generated codebases. Why Observability Falls Short for AI Engineering teams already invest heavily in logs, metrics, traces, and application performance monitoring https://thenextweb.com/news/full-stack-baby platforms. Those systems remain valuable for identifying when services become unhealthy, but Adler argues they weren’t designed to provide the kind of evidence AI systems require. As he puts it, “ They’re built to surface that something is wrong, not why. ” While logs, metrics, and traces can help engineers investigate incidents, “ Agents iterating over logs is what we have today, and it’s insufficient. ” Instead, Adler believes AI requires “structured, function-level context describing what is actually executed,” giving models direct visibility into how code behaves under real production conditions rather than asking them to infer behavior from fragmented telemetry. That philosophy forms the foundation of what Hud calls runtime intelligence. Why Context Matters More Than More Data Traditional observability often produces enormous volumes of telemetry, leaving engineers to reconstruct failures after the fact. Adler argues that simply collecting more information doesn’t solve the problem if the critical details remain buried. As he explains, “ More telemetry is a bigger haystack, and the needle is the only thing you came for. ” When production incidents occur, engineers don’t need terabytes of logs. Instead, they need the exact execution flow, affected parameters, code paths, dependency behavior, and surrounding forensic context that explain why a failure occurred. Hud’s approach centers on automatically capturing that forensic context as incidents happen, rather than requiring engineers to predict which logs, dashboards, or instrumentation they’ll need beforehand. Adler argues that this shift is increasingly important because manual configuration inevitably leaves coverage gaps. “ Anything that depends on setup gets applied unevenly, ” he says, adding that organizations often miss the very failures they never anticipated. Runtime Intelligence as a New Architectural Layer Although many observability vendors are adding AI assistants to existing products, Adler doesn’t view runtime intelligence as an incremental feature. Instead, he argues that “ A different layer, not a feature you append. ” Simply placing AI on top of sampled telemetry means the model inherits every limitation of the underlying data. Runtime intelligence, by contrast, changes what information is captured from the beginning, producing production-aware context specifically structured for AI reasoning rather than human investigation. That distinction also explains Hud’s emphasis on capturing execution context without sampling. According to Adler, sampling works well for dashboards and long-term trends but becomes problematic during incident response because the rare event that triggered the outage may be exactly what gets discarded. Since AI reasons only from the information it’s given, missing production data doesn’t merely slow investigations; it can lead models toward incorrect conclusions. The Future Is Reviewing AI, Not Debugging It Looking ahead, Adler expects engineers to spend far less time manually investigating production incidents and significantly more time validating AI-generated fixes. He points to organizations already seeing dramatic reductions in investigation time once root cause information is captured automatically rather than reconstructed manually. But he emphasizes that autonomous software maintenance depends on giving AI direct access to production truth instead of assumptions. That same philosophy extends to autonomous deployments. Adler believes organizations will eventually trust coding agents to make production changes independently, but only after those systems consistently demonstrate transparent reasoning and operate within clear safety boundaries. As he says, “ Trust is built on evidence and reversibility, not faith. ” AI should reason from actual production behavior, explain its conclusions, and allow humans to quickly verify or reverse its decisions. Ultimately, Adler believes the defining challenge of the AI software era isn’t writing code faster but understanding how that code behaves once it’s deployed. As he concludes, “ As AI writes more of our software, the scarce resource is no longer writing code; it’s knowing how that code behaves in the real world. Whoever holds that understanding and can put it in front of both people and agents is the one who ships with confidence and truly harnesses the power of the AI code-gen revolution. “ Get the TNW newsletter Get the most important tech news in your inbox each week. TNW newsroom and editorial staff were not involved in the creation of this content.