Faisal Fareed has over 24 years of experience in the Financial Services Industry. Principal Solutions Architect - Amazon Web Services.
getty
AI is no longer just a feature added to software. It is becoming part of the software stack. Teams now work with agents, prompts, tools, memory, permissions, retrieval systems and model-powered workflows. For industry leaders, this is a talent issue. Organizations need people who can turn AI capability into secure, measurable, governed production systems.
The universities built the foundations that made this moment possible: computer science, computer engineering, software engineering, data science, cybersecurity, information systems and applied computing. The market now needs people who can work across models, software systems, data pipelines, tools, evaluation, security, governance and domain workflows.
The AI engineer is not a data scientist with a new title, a machine learning researcher or a software engineer who uses ChatGPT. The role turns foundation models, data, tools, workflows, controls and evaluation systems into reliable products.
In a 2026 AI Ascent interview, Andrej Karpathy described three software eras. Software 1.0 is explicit code: humans write rules and computers execute them. Software 2.0 is learned behavior: humans prepare datasets, objectives, architectures and training loops. Software 3.0 is model-mediated computation: humans provide goals, context, constraints, tools and feedback to large models.
That framing changes what technical workers need to know. Students still need to write software. They also need to know when a function should exist, when a model call is better, when retrieval or validation is required, when human approval is necessary and how to measure behavior.
Karpathy also separates “vibe coding” from “agentic engineering.” Vibe coding expands access. Agentic engineering keeps the production bar in place. Professionals can use agents, but they still own security, reliability, maintainability and user outcomes.
The World Economic Forum’s "Future of Jobs Report 2025" projects that AI and information processing will transform employers by 2030. It identifies AI and big data as the fastest-growing skills, and lists AI specialists, big data specialists, fintech engineers and software developers among fast-growing technology roles.
LinkedIn’s 2026 Jobs on the Rise list identifies AI engineer as the fastest-growing role in the U.S. Microsoft’s "2025 Work Trend Index" describes “human-agent teams,” where agents take on more execution and people shift toward judgment, workflow design, coordination and accountability.
The signal is consistent: Employers need people who can build, evaluate, secure and govern these agentic AI systems.
AI engineering draws from computer science, computer engineering and data science, but it has a different center of gravity: How do we build reliable, secure and accountable systems around intelligent models?
The answer includes models, but it does not stop there. It includes prompts, tool APIs, retrieval, orchestration, evaluation datasets, guardrails, observability, cost, privacy, governance and deployment.
For executives, this distinction matters because “AI talent” is not one job. A research scientist, data analyst, software engineer, power user and AI engineer solve different problems.
Existing programs often contain pieces of this skill set, but the full AI engineering profile is still uncommon. A computer science program may teach algorithms, systems and AI fundamentals, but not LLM application architecture or AI product operations. A data science program may teach statistics and modeling, but not secure deployment or real-time AI product design. A computer engineering program may teach hardware and embedded systems, but not foundation-model orchestration.
The gap is not AI awareness. Many universities teach machine learning and deep learning. The gap is the production lifecycle: evaluation, deployment, monitoring, security, governance and workflow design.
Carnegie Mellon is one model to study, but not the only signal. Universities are launching programs under names such as artificial intelligence engineering, applied AI, AI for product innovation, AI and engineering systems, AI infrastructure and professional master’s programs in AI. Their syllabi vary across machine learning, engineering domains, product work, agentic systems, trustworthy AI, security and deployment.
That variation is expected. It also shows the gap. A handful of programs cannot supply the talent pipeline employers are requesting.
AI engineering should be built as a competency-based curriculum, not a loose bundle of AI electives.
The curriculum should include computing foundations: programming, systems, databases, APIs, testing, CI/CD and secure software. It also needs math, statistics and machine learning: probability, uncertainty, model training, validation, transformers, fine-tuning, benchmarking and model limits.
It needs the new applied layer: prompting as specification, structured outputs, retrieval-augmented generation, tool use, latency, cost and orchestration. Agentic systems should be taught directly: task decomposition, tool design, permissions, human approval, observability, recovery and agent-native documentation.
Evaluation deserves its own track. Students should learn golden datasets, regression tests, human evaluation, LLM-as-judge evaluations, red-team tests, trace-based evaluations and release gates.
Security and governance must be core: prompt injection, data leakage, model poisoning, supply-chain risk, excessive agency, secure tool execution, NIST AI RMF, ISO/IEC 42001, privacy, bias, auditability and incident response.
The curriculum also needs human-centered design and domain depth: workflow integration, uncertainty in interfaces, escalation, accessibility, trust calibration and applied work in healthcare, finance, manufacturing, education, cybersecurity or the public sector.
The capstone should go beyond a demo. Students should specify, build, evaluate, red-team, secure, deploy and govern an AI product or workflow.
Industry, academia and policy leaders should define the AI engineer role together. Start with a shared graduate and early-career profile. Map existing courses to that profile. Launch certificates where needed. Build AI systems labs with model access, cloud credits, vector databases, evaluation tools and red-team environments.
Employers should bring workflows, anonymized datasets, mentors and production constraints. They should also redesign entry-level work. If AI automates old junior tasks, early-career talent still needs a path to build judgment. Apprenticeship should move toward supervised agent orchestration, evaluation, security review and production monitoring.
Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?
此内容由惯性聚合(RSS阅读器)自动聚合整理，仅供阅读参考。 原文来自 — 版权归原作者所有。