---
title: "AI Product Management: Strategy to Shipped, With the Evidence Attached"
description: "Aleksandr Filippov practices AI product management end to end — choosing the problems worth an AI product, setting a measurable quality bar, running evaluation before wide release, and driving adoption — across a localization system, a retrieval platform, and agent-facing tooling."
date: "2024-03-17T07:00:00Z"
last_updated: "2026-08-28"
build_time: "2026-09-07T11:25:46Z"
skill_name: "AI Product Management"
skill_description: "AI products taken from problem selection to adopted production systems — a measurable quality bar set up front, evaluation run before wide release, platform framing that compounds, and outcomes stated in numbers the products themselves published."
skill_category: ["ai-product-strategy"]
since: "2025-02-01"
last_used: "2026-07-27"
related_skills: ["artificial-intelligence","product-technology-leadership","business-analysis","stakeholder-management","system-analysis","agile-methodologies","scrum"]
related_projects: ["air-api","aila","youtrack-mcp-server","peopleforce-mcp-server"]
related_experience: ["ai-product-manager-at-spotware"]
certifications: ["psm-one-by-scrum-dot-org"]
keywords: ["AI Product Management","AI Product Strategy","Product Discovery","Quality Bar","Blind Evaluation","Product Adoption","Platform Product Management","Stakeholder Alignment","Roadmap"]
license: "https://creativecommons.org/licenses/by/4.0/"
canonical: "https://www.alexfeel.info/skills/product-management/"
---

My AI product management starts one step earlier than most definitions of the job: **choosing which problems deserve an AI product at all**. An expensive, slow, high-volume workflow with a checkable quality standard is a candidate; a workflow where nobody can say what "good" means is not, yet. The products under my management came out of that filter — a localization system aimed at a cost and turnaround problem the whole company could feel, a retrieval platform built once so every team's AI product inherits the same grounded answers, and agent-facing tool servers that turn systems of record into surfaces agents can safely use.

The method is evaluation before conviction. AILA went wide only after a **blind quality comparison against professional human translation**, and its first year is stated in measured terms — volume, cost per unit, turnaround — published on its own page. The retrieval platform logs per-stage score distributions and percentile bands, so retrieval quality is calibrated from production evidence rather than demos. Alongside the bar-setting sits the unglamorous majority of the role: aligning support, marketing, engineering, and executives on what ships and what waits, sequencing a roadmap so each release earns trust for the next, and saying no to AI features whose success could not be measured. I also build a large share of what I manage — pipeline code, serving surfaces, tests — which keeps the roadmap grounded in what the systems can actually do; the split between the leading and the building is documented on the [AI Product Manager at Spotware](/experience/ai-product-manager-at-spotware/) page.


## Site index

- [llms.txt](https://www.alexfeel.info/llms.txt)
- [Complete index](https://www.alexfeel.info/llms-index.txt)