06 Do you Need a Lab Coat to Lead AI
Why strategy, solutioning, and delivery make you an AI expert
Recently, someone offered me a bit of well-meaning “advice” that I’ve been thinking about: If you brand yourself as an AI expert, you’ll get pushback because you’re not a data scientist.
I respect data scientists. I partner with them. But the idea that AI expertise = data science only or bust is outdated in my opinion.
In 2025, enterprise AI isn’t a science fair project. It’s a systems problem and a delivery problem: choose the right capability (often a reputable, existing model), embed it into the process, orchestrate it safely, and prove outcomes. That’s the job. That’s the expertise.
🎯 What I Mean by “AI Expertise”
AI expertise is the ability to translate a business outcome into a safe, integrated, measurable AI-enabled workflow. In practice: frame the problem, pick the minimal capable model, orchestrate it into the process with guardrails, and prove value with live evaluation.
🧩 Myth vs. Reality
Myth: An AI expert must build models from scratch.
Reality: Most enterprise value comes from composing proven capabilities—LLMs, vector search, RAG, rules, human-in-the-loop—into reliable workflows that move metrics. When hold times drop or risk reviews speed up, no one asks if you trained the model from scratch. They ask how fast you can scale.
🧱 What Enterprise AI Actually Requires
Think of modern AI as a stack. Data science is one vital layer—but not the only one.
🎯 Problem Framing & Value Cases
Tie capabilities to outcomes. Define success in real numbers: cycle time, quality thresholds, risk, cost, customer effort.🔐 Data Access & Governance
Bring intelligence to governed data. Minimize movement. Respect lineage, retention, and permissions.🔌 Orchestration & Integration
Low-code and integration platforms (yes, including Appian) + APIs (OpenAI, Anthropic, Google, etc.) + RAG + eventing = systems that actually run.▪️ Model portfolio + abstraction: pluggable layer to swap providers, support private/on-prem endpoints, and route by use case (cost/latency/sensitivity).
▪️ Scope boundaries: use low-code for human workflow, audit trails, and case management.
🛡️ Controls, Safety & Compliance
Guardrails, red-teaming, PII handling, auditability, fallback paths, and role-based access. Ship with confidence, not caveats.👥 Change Management & Adoption
Train the people. Update the SOPs. Align incentives. If it isn’t used, it isn’t value.📈 Evaluation & Monitoring
Evaluation harnesses before and after go-live; routing with pass/fail gates; dashboards for drift, cost, latency, and quality; rollback playbooks.
👩🔬 When You Do Need a Data Scientist
There are absolutely cases for bespoke modeling:
🔭 You need novel predictive power beyond off-the-shelf models.
⚙️ You’re optimizing at the algorithmic level for a unique domain (latency/cost/accuracy trade-offs).
🛰️ You’re building features from raw signals (IoT, specialized NLP/CV) that require experimentation.
In regulated or novel domains, data science leads; my role is ensuring their work is governed, integrated, and adopted at scale. AI is a team sport.
💡 Clear Take
You don’t need a lab coat to lead in AI. In the enterprise, expertise is the ability to turn AI into outcomes—by choosing the right capabilities, composing them into safe, usable systems, and delivering measurable value.
That’s strategy. That’s solutioning. That’s delivery. And yes—that’s AI expertise.
