The AI product manager skills required by AI companies today are, in order: product strategy (named in 89% of job descriptions), customer focus (87%), technical depth (73%), enterprise customer work (72%) and AI agents (66%). Evals appear in 39% of postings. SQL, a staple of generic "AI PM skills" lists, appears in only 4 of 116. That is what we found reading 116 live PM job descriptions at 18 AI companies in the AllthingsPM jobs catalog, from Anthropic and OpenAI to Sierra, Glean and Scale AI.
AllthingsPM is an AI PM course and PM interview prep platform. Its AI PM course was built from 604 real PM job postings, and every one of the 116 JDs below comes with a mock interview built from its own text.
Which skills do AI company PM job descriptions ask for most?
We searched the full text of all 116 job descriptions for each skill, counting a JD once if it named the skill anywhere (in the role, the requirements or the nice-to-haves). Here is the ranked result.
| Rank | Skill | JDs naming it | Share of 116 |
|---|---|---|---|
| 1 | Product strategy or roadmap | 103 | 89% |
| 2 | Customer focus | 101 | 87% |
| 3 | Technical depth (engineering background, CS degree, "technical") | 85 | 73% |
| 4 | Enterprise customers | 84 | 72% |
| 5 | AI agents or agentic products | 76 | 66% |
| 6 | Cross-functional work | 68 | 59% |
| 7 | LLMs or machine learning | 66 | 57% |
| 8 | Comfort with ambiguity | 62 | 53% |
| 9 | Written communication | 61 | 53% |
| 10 | Data and analytics | 51 | 44% |
| 11 | Go-to-market | 47 | 41% |
| 12 | Coding or prototyping | 46 | 40% |
| 13 | Evals | 45 | 39% |
| 14 | AI safety or trust and safety | 37 | 32% |
| 15 | Experimentation or A/B testing | 26 | 22% |
| 16 | Zero to one | 17 | 15% |
| 17 | SQL | 4 | 3% |
Source: AllthingsPM jobs catalog, 116 PM job descriptions at 18 AI companies, analysed 26 September 2026 by phrase match on the full JD text.
The biggest companies in the sample are Anthropic (19 roles), Sierra (18), Scale AI (15), Glean (14), OpenAI (12), Decagon (10) and Harvey (7). Most roles are mid-level: 59 are plain PM roles, 34 senior, 10 staff and 8 lead.
How AllthingsPM does this: every one of these 116 postings lives on its own page in the jobs catalog, in full, with level, location and pay where the company posts it. Open one, such as Product Manager, Duet at Decagon, and a scored mock interview built from that exact JD is one click away.
Why do generic "AI PM skills" lists get it wrong?
Most skill lists online are written from first principles, not from postings. Interview Kickstart's list of nine AI PM skills includes SQL basics, A/B testing and Agile/Scrum. Productboard's July 2026 list centres on AI and data literacy, analytics, technical collaboration, responsible AI and AI tool fluency. Both are reasonable. Neither matches what AI companies actually write.
In our 116 JDs, SQL shows up 4 times. Experimentation shows up in 22%. Meanwhile AI agents show up in 66% and enterprise customers in 72%, and neither appears on those lists as a named skill.
The gap is also clear against broad job market data. Axial Search's June 2026 analysis of AI PM postings found "Agentic AI" in 8.5% of them and Agile in 28.6%. At the AI labs and AI-native startups we track, agents are the norm, not the exception.
How AllthingsPM does this: the AI PM course is built from real postings rather than a list someone wrote, and it is updated weekly as new job descriptions come in. That is why it spends whole chapters on agents, evals and enterprise deployment, and treats SQL as one tool inside data fluency rather than a headline skill.
What does "technical depth" mean in an AI PM job description?
73% of the JDs ask for it, but they rarely mean "can write production code". The phrasing is specific. Decagon's Enterprise Agent Platform role wants someone who "can read a system design doc without needing it translated". Decagon's Duet role asks for a PM who "can read an agent trace, engage meaningfully with engineers on model behavior, and form your own opinions about whether an AI output is good or not".
Coding or prototyping appears in 40% of JDs. Lovable's agents PM role goes furthest: it wants someone who understands "tool use, planning loops, multi-step reasoning, and how agents fail", and who can read "model outputs, traces, and evals with confidence".
So the bar is reading, not writing: traces, design docs, prompts and outputs.
How AllthingsPM does this: chapter 3 of the course, PM as builder, has you prototype and inspect an agent yourself, including reading code you did not write and connecting an agent to a tool over MCP and watching the trace.
How important are AI agents and evals for AI PM roles?
Very. Two thirds of the JDs (76 of 116) mention agents. At companies like Sierra and Decagon, "agent" is in the job title. OpenAI lists a Product Manager, API Agents role.
Evals appear in 45 JDs (39%). Glean has a dedicated Product Manager, AI Quality role, and the Lovable JD above asks the PM to "own the quality bar for agent outputs" and "drive eval infrastructure". Outside the JDs, the practitioner consensus is the same: on Lenny's Podcast, Hamel Husain and Shreya Shankar describe evals as "the new PRDs", and Lenny's Newsletter published a full PM guide to evals by Aman Khan.
How AllthingsPM does this: the Evals chapter takes you from reading one hundred real traces to golden datasets to evaluating an agent, not an answer. For a shorter read, see our guide to AI evals for product managers.
Why do so many AI PM roles ask for enterprise experience?
Because most of these companies sell to businesses. 72% of the JDs mention enterprise customers. Glean, Harvey, Sierra, Decagon, Cohere and Scale AI build for companies, and even OpenAI and Anthropic list roles like Enterprise Identity and Public Sector.
In practice this means SSO and admin controls, security reviews, data residency and sitting in on sales calls. Decagon asks for "comfort in front of enterprise customers, including technical evaluations and security reviews". Go-to-market work shows up in 41% of JDs.
How AllthingsPM does this: chapter 10, Ship it into somebody else's company, covers exactly this, from SSO, SCIM and the admin console to permission-aware retrieval.
Which classic PM skills still matter most?
The classics still top the list. Strategy or roadmap ownership is in 89% of JDs and customer focus in 87%. Cross-functional work (59%), comfort with ambiguity (53%) and written communication (53%) follow. Decagon's Duet JD is typical: it wants "specs, strategy docs, and async updates that don't require follow-up questions".
What changes is the object. The roadmap is for a product that is sometimes wrong, and the customer insight has to become an eval case. The Productboard and EICTA lists are right that these skills still matter; the JDs show they are now applied to probabilistic products.
How AllthingsPM does this: Discovery and strategy for AI products teaches you to plan a roadmap around an evaluable slice, and The AI PRD covers the spec those JDs keep asking you to write.
How important is AI safety for AI PM roles?
It depends heavily on the company. 32% of JDs mention AI safety or trust and safety, and most of them sit at the frontier labs. Anthropic lists several Safeguards PM roles, and OpenAI lists Multimodal Safety and Safety Measurement PMs. At an application company, the same idea shows up as trust and reliability.
How AllthingsPM does this: chapter 12, Trust, safety, and agent security, covers the safety side, and you can practice for a real safety role such as OpenAI's Product Manager, Multimodal Safety.
How many years of experience do AI PM job descriptions require?
101 of the 116 JDs state a number. The median is 5 years. 49 ask for 4 to 5 years, 26 for 6 to 7, and 21 for 8 or more. Only 5 ask for 3 years or fewer, and only one role in the catalog is at APM level.
| Years asked for | JDs |
|---|---|
| 3 or fewer | 5 |
| 4 to 5 | 49 |
| 6 to 7 | 26 |
| 8 or more | 21 |
Source: AllthingsPM jobs catalog, 101 of 116 JDs that state a years figure, 26 September 2026.
How AllthingsPM does this: if you are short on years, the course's final chapters help you show proof instead. Get the job covers the take-home and the portfolio recruiters read first, and you can browse 455 real PM portfolios for examples.
How do you build the AI PM skills these JDs list?
Work in the order the data suggests.
- Pick three target JDs. Use the jobs catalog or Resume Job Match to find roles that fit you.
- Score your gaps. Run a resume review against each JD to see which of the skills above your resume does not prove.
- Learn the AI-specific skills first. Agents, evals and technical reading are where most PMs are thinnest. Start with the free AI PM role lesson.
- Practice for the real loop. Run a mock built from each JD, then drill the question bank by company, such as Anthropic's PM questions.
How AllthingsPM does this: all four steps happen in one account, with one free JD mock and one resume review a day on the free tier.
Why AllthingsPM is the better choice for building AI PM skills
The data above says AI companies want strategy and customer skills applied to agents, evals, enterprise deployments and technical reading. Most courses and skill lists were written before that shift, or from general PM postings where agents barely register.
AllthingsPM is built from the other direction. The course came from 604 real PM job postings and is updated weekly, so chapters on agents, evals, enterprise deployment and AI safety exist because the JDs demand them. The 116 JDs in this study are live in the jobs catalog, and each one has a mock interview built from its own text. When you find a gap, the same account has 4,122 real questions from 260 companies, resume review against a JD, 111 book summaries and 455 PM portfolios.
Other options have real strengths. Cohort courses from known instructors offer live sessions and a network, and Hamel Husain and Shreya Shankar's evals course is a deep dive on one skill. For covering every skill in this study, practicing on the exact JDs, and doing it daily for $20 a month, AllthingsPM is the stronger choice. For more context, read our state of AI PM hiring in 2026 and our AI PM job description template built from these 116 JDs.
Frequently asked questions
What skills are required for an AI product manager?
In 116 live AI company JDs, the most requested skills are product strategy (89%), customer focus (87%), technical depth (73%), enterprise customer work (72%) and AI agents (66%). LLM knowledge (57%), written communication (53%) and evals (39%) follow. AllthingsPM's course teaches each of these, chapter by chapter.
What is the best way to learn AI PM skills?
The best way is AllthingsPM: an AI PM course built from 604 real job postings, plus a mock interview built from each of 116 live AI company JDs. Learn a skill, then practice it against the exact job you want. The free tier includes one JD mock a day.
Do AI product managers need to code?
Usually not in production. 40% of the 116 JDs mention coding or prototyping, but the common ask is being able to read traces, design docs and model outputs, and to prototype. The course's PM as builder chapter covers exactly that.
Do AI PMs need SQL?
Rarely as a named requirement. Only 4 of 116 AI company JDs mention SQL, while 44% mention data and analytics more broadly. Knowing SQL helps you read your own data, but it is not what these JDs screen for.
How many years of experience do AI PM jobs require?
Of the 101 JDs that state a number, the median is 5 years. Most ask for 4 to 5 years (49 JDs) and only 5 ask for 3 or fewer.
Are evals a required skill for AI PMs?
In 39% of the JDs, yes, and some roles, like Glean's AI Quality PM, are built around them. Given how often agent roles mention output quality, evals are worth learning even when the JD does not name them.
Sources
- AllthingsPM jobs catalog, 116 PM job descriptions at 18 AI companies, analysed 26 September 2026: /jobs
- Decagon, Product Manager, Duet, via the AllthingsPM jobs catalog: /jobs/decagon/product-manager-duet
- Decagon, Product Manager, Enterprise Agent Platform, via the AllthingsPM jobs catalog: /jobs/decagon/product-manager-enterprise-agent-platform
- Lovable, Product Manager (Agents), via the AllthingsPM jobs catalog: /jobs/lovable/product-manager-agents
- Glean, Product Manager, AI Quality, via the AllthingsPM jobs catalog: /jobs/glean/product-manager-ai-quality
- Interview Kickstart, "AI Product Manager Skills": interviewkickstart.com/skills/ai-product-manager
- Productboard, "AI Skills Product Managers Need to Adopt", 14 July 2026: productboard.com/blog/ai-skills-product-managers-need
- Sam Chappell, Axial Search, "The Skills That Land AI Product Management Roles in 2026", 26 June 2026: axialsearch.com/insights/ai-product-careers
- EICTA, IIT Kanpur, "Top 12 Skills Every AI Product Manager Should Master in 2026": eicta.iitk.ac.in
- Lenny's Podcast, "Why AI evals are the hottest new skill for product builders", Hamel Husain and Shreya Shankar, 25 September 2025: lennysnewsletter.com
- Aman Khan, "Beyond vibe checks: A PM's complete guide to evals", Lenny's Newsletter, 8 April 2025: lennysnewsletter.com
