AI product management
What is AI product management?
AI product management is the work of deciding what an AI product should do, proving that it works, and keeping it safe once people rely on it. It adds evaluation, guardrails and agent design to ordinary product work. Adnan Khan, a product leader at Vantage Data Centers, writes about it here.
A traditional feature behaves the same way every time, so a fixed test can pass or fail it. An AI feature gives different answers to similar inputs. The product manager's requirements therefore include a set of real examples, a way to score the results, and thresholds that decide whether to launch.
The job also covers what the system is allowed to do on its own. A system that only explains needs few controls. One that proposes drafts or takes actions needs permissions, review steps and a way to stop it.
The skills
Core product skills come first. These are what AI adds on top, each with the Field Notes topic and the how-tos that cover it.
Scoping the AI job
Turn a vague wish into one bounded job with named inputs, tools and human decisions.
Writing requirements with evals
Put the test set, the scores and the launch thresholds in the PRD.
Evaluating models and agents
Score the path an agent took and its tool use, and check any LLM judge against human labels.
Guardrails and governance
Separate explaining, proposing and acting, set permissions per tool, and keep a stop control.
Tools, context and memory
Decide what an agent can connect to, what it holds in mind and what it remembers.
Workflows and multi-agent systems
Turn a repeated process into steps an agent can take, and know when more than one agent helps.
Enablement and adoption
Roll AI tools out to teams with a roadmap, KPIs, training and shared practices.
What are AI agents?
An AI agent is a system that uses a language model to plan steps and call tools to finish a job, within limits a person sets. A chatbot answers inside a conversation, and an agent can change things outside it, so it needs more controls.
How Adnan works
Adnan builds AI systems and the operating models that help enterprises trust, adopt and scale them. His through-line is turning ambiguous business problems into working products, clear controls and repeatable operating systems.
At Equinix, Adnan built a production reporting agent on Claude and Model Context Protocol (MCP) for the Quote to Cash initiative. It automated status reporting from days to minutes across 22 products and 273 stakeholders. The agent used an existing Atlassian MCP integration; Adnan used that integration and did not author it.
As Senior Manager, Product Operations at Equinix, Adnan owned AI enablement across Claude and Microsoft Copilot 365: roadmap, KPIs, training and agentic workflows. He re-engineered Equinix Internet Access ordering, reducing sales ordering and provisioning from days to hours and related support tickets by 27%.
Adnan is developing a supply and demand visibility platform and agent at Vantage Data Centers to connect site selection, due diligence, test-fit and prefab procurement. He is also implementing a test-fitting tool designed to reduce site selection from 2 weeks to days. These time savings and the 8-week ready-for-service reduction are targets, not verified achieved outcomes.
Outside work, Adnan builds agentic products end to end. They include OM5, a governed operating system with builder and judge separation and earned autonomy; Centurian, an agent-operations platform for regulated industries; and SherHaul, an AI iOS app for owner-operator truckers, live on the App Store.
Adnan is an AWS Certified AI Practitioner. His Anthropic certifications are Claude 101, Model Context Protocol: Advanced Topics and Introduction to Agent Skills. He completed the UC Berkeley Executive Education program Agentic AI: Strategy, Applications and Organizational Impact at Haas, and lists Communication: Public Speaking.
The full record is on the About page.
Field Notes on these topics
Adnan keeps research notes from his reading on AI, 1,406 so far. Each topic page lists its notes with the source linked.
- AI agents 154 notes
- AI evaluation and benchmarks 257 notes
- Multi-agent systems 55 notes
- Memory and context 32 notes
- Tool use and MCP 13 notes
- AI safety and alignment 109 notes
- AI governance and ethics 68 notes
- Enterprise AI and automation 56 notes
- Coding agents and software engineering 45 notes
- AI security 51 notes
Browse every topic or search the notes.
Guides
Short how-tos that open with a direct answer and cite the Field Notes behind them.
How to build an enterprise AI agent around a real workflow
Start with a job people already need to finish. Map the steps, identify trusted information and decide who owns each decision. Give the agent a small role, connect only the tools it needs, test realistic failures and keep human approval for consequential actions. Expand after a pilot shows that the workflow works.
Frequently asked questions
What does an AI product manager do?
An AI product manager decides what an AI system should do, for whom, and how the team will know it works. They write requirements that include evaluations, set limits on what the system may do on its own, and watch quality after launch. The rest of the job looks like product management anywhere: customers, priorities and launches.
How is AI product management different from regular product management?
The product's behavior varies from one run to the next, so a fixed test script can't certify it. Requirements include evaluation sets and thresholds, and quality needs ongoing monitoring. The product manager also owns new risks, such as wrong answers, data leaks and actions taken without approval.
Do AI product managers need to know how to code?
They don't need to be engineers, but they should be able to read code, run a notebook and build a small prototype. That's how they judge what's feasible and find failures early. Adnan researches, prototypes and builds with AI coding agents and tools himself.
What skills does an AI product manager need?
Core product skills come first: finding the problem, setting priorities and working with engineers. On top of those sit evaluation, guardrails and governance, working with tools and memory, and rolling AI out to teams. The skills section on this page links to a how-to for each.
What is an AI agent, and when should a product use one?
An AI agent is a system that uses a language model to plan steps and call tools to finish a job, within limits a person sets. Use one when the job needs action in other systems, and use a simpler chatbot when an answer is enough. Start an agent with drafts and reversible actions.
How do you know whether an AI feature works?
Test it against a set of real examples, including messy and adversarial ones, and score the results against thresholds you set in advance. Score the path the system took as well as its final answer. Keep checking after launch, because inputs and models change.
What is the Model Context Protocol, and why does a product manager care?
The Model Context Protocol, or MCP, is an open standard that lets an AI agent discover and call tools and data on a server. A product manager decides which systems connect, what each connection may do and who the agent acts as. Adnan holds Anthropic's Model Context Protocol: Advanced Topics certification.
How does Adnan Khan approach AI product management?
Adnan builds AI systems and the operating models that help enterprises trust, adopt and scale them. At Equinix he built a production reporting agent on Claude and MCP that took status reporting from days to minutes across 22 products and 273 stakeholders. He now leads North America product enablement at Vantage Data Centers.
