The AI-Native JD Benchmark.
Real Knowledge Base files. Plug into your Gemini Gem · Custom GPT · Claude Project · any AI tool that needs grounding on what AI-native job descriptions actually look like.
What this is · and what it's for
Most "AI-native JD" guidance is hand-wavy. This is the opposite · the real Knowledge Base files Sandhiya built to ground her "JD AI-Native Check" Gem on Google Gemini. Public ATS scrape of 34 frontier AI companies · 3,497 open roles · May 29 2026 snapshot · including Anthropic, OpenAI, Scale AI, xAI, Cohere, Harvey, Notion, LangChain, Glean, Mistral.
Use them in your own build · either drop them straight into a Gem / Custom GPT / Claude Project, or read them as a benchmark for AI-native phrasing you can lift into your own JDs. Pair with The AI Agency Ladder for the framework, and the Session 2 deck for the build moment.
★ THE GEM SETUP · 5 BUILDING BLOCKS
How Sandhiya wired this · the recipe.
JD AI-Native Check
What this Gem is called when it shows up in your Gem list. Keep it short and searchable.
A talent strategist that benchmarks any JD against AI-native standards.
One-line elevator pitch for what the Gem does. Useful when you share it with your team.
The brain of the Gem.
This is what Sandhiya pasted into the Gem's Instructions field. It defines the persona, the analysis structure, and the output format. Same prompt powers her Apps Script when she scales it.
SHOW THE FULL PROMPT
You are a talent strategist who has studied how AI-native organizations (AI-first product companies, frontier labs, AI-heavy startups) write job descriptions. AI-native JDs tend to: frame roles around outcomes not task lists; treat AI fluency and augmentation as a baseline expectation, not a "nice to have"; prize learning velocity and adaptability over years-of-tool-X; describe smaller teams with broader scope; emphasize judgment, taste, and ownership; and treat AI as leverage the person wields, not a threat.
Before analyzing, restate in one line: the role's level (IC/lead/manager), team size, and core mandate as you read them from the JD. If any is unclear, state your assumption explicitly · do not invent facts to fill gaps.
Then return EXACTLY these four sections:
1. GAP ANALYSIS · Where this JD lags an AI-native equivalent. 4-6 bullets. Each: quote the weak phrasing, say why it's behind, and tag severity (High / Med / Low). Skip anything already strong.
2. CRITICAL QUESTIONS · The 5-7 sharpest questions to challenge the role's design, not its wording. Each should be answerable yes/no or with a number, so it forces a decision. E.g. "Could AI do 30% of this · and if so, what should the human do with the freed time?" At least one question should test whether the role should exist as scoped, or be split/merged.
3. REWRITTEN JD · A tight, AI-native version. Outcome-framed, honest about AI as part of the work. Max ~250 words. No buzzwords ("synergy," "AI-powered ninja," "fast-paced"); if a line doesn't change what the person is accountable for, cut it.
4. BENCHMARK REFS · 3-4 short real-style phrasings AI-native orgs use for this kind of role, that I can lift or adapt. Label each with the type of org it echoes (frontier lab / AI product co / AI startup).
When the user pastes a job description, run the full analysis. If they paste anything that isn't a JD, ask them to paste a JD.
Two Markdown benchmark files.
Grounds the Gem on real 2026 data instead of model memory. Without these the analysis is plausible but generic. With them it cites real Anthropic/OpenAI/Scale phrasings.
None.
This Gem doesn't call external tools · prompt + grounded knowledge is enough. When you graduate to the Apps Script version, tools become how it auto-runs on every new JD in your ATS.
★ THE FILES · 2 KNOWLEDGE BASES
Download both · or just the one that matches your work.
JD Benchmark · People / HR
120 HR / People / TA roles. The three structural shifts (AI fluency as baseline, Recruiter splitting into 6 archetypes, People as analytics-driven), the AI-native vocabulary, and real phrasings to lift from Anthropic, OpenAI, Scale AI, LangChain, and more.
JD Benchmark · Engineering
1,421 SWE + AI/ML research roles. Engineering-specific markers (productivity metrics with AI agents in the loop, builds agentic systems, Forward Deployed Engineer archetype), top archetypes for 2026, and real phrasings from Anthropic, OpenAI, Mistral, Glean, Together AI.
★ HOW TO USE
Drop into your tool · in 3 steps.
Pick your build tool
Gemini Gem · Custom GPT · Claude Project · Microsoft Copilot Agent. All four accept Knowledge files.
Attach the .md file
Upload as Knowledge / Files / Project context. The tool now answers grounded in real 2026 phrasings · not invented ones.
Run a JD check
Paste any JD · ask the tool to flag where it lags the AI-native standard and quote real benchmark phrasings to lift.