Job titles are born in postings before they exist anywhere else. Before the bootcamps, before the LinkedIn headlines, someone in a hiring committee has to type a phrase into a job board for the first time. Because we read the boards daily, we get to watch the nursery. The July 2026 census of titles that were not a thing in 2023, then.
The census
Forward-deployed engineer: 337 postings, the clear leader, covered at length in its own piece. Agentic roles, engineers and product people building AI agents: 188 postings and climbing fast, the phrase spreading from labs into commerce and operations titles. Prompt engineer: 72 postings. Model evaluator: 13. AI red-teamer: 11. AI safety roles: 11. Responsible AI titles: a handful. Growth engineer, the title Cluely-adjacent startups love: 3, which suggests the attention economy generates more discourse than headcount.
Which ones consolidate, which ones vanish
New titles follow one of two paths. Some consolidate into a broader role once the skill stops being exotic. Prompt engineering is visibly on this path: 72 postings against 943 for AI engineer, and in our adjacency graph the two roles overlap so heavily that the smaller one reads as a feature of the bigger one. The US median gap, about 102,000 dollars for the specialist against 150,500 for the generalist, is the market pricing the consolidation in real time. Learn the skill, skip the title.
Others stay narrow because they answer to regulation or risk rather than fashion. Red-teaming and model evaluation are small but stubborn: 24 postings between them, mostly at companies with compliance exposure, and they behave in the data like early security engineering did, a niche that becomes an institution because someone has to sign the audit. If you want a small pond with a moat, that corner is worth a look before the certification industry finds it.
How to read a newborn title
Three questions separate a career from a costume. Does the title describe work someone was already doing under an older name, or genuinely new work? Does demand come from many industries or one hype cluster? And does the pay carry a premium over the nearest established role, or a discount? Forward-deployed engineering passes all three. Prompt engineering passes none of them anymore. The agentic cluster passes the first two and is still arguing with the third. Check again next quarter; the nursery updates daily.
Where this leaves you
Chasing new titles is a bad strategy, and ignoring them is a slightly worse one. The winning move is to watch which newborn titles describe your existing week better than your current title does, then apply while the applicant pool is still confused. That window is the entire value of reading the nursery. It closes the day the bootcamps open.
Title counts: regex over titles in the PivotHop raw corpus, July 2026, roughly 99,000 postings across thirteen sources. Salary medians from our blended engine (BLS OEWS (the Bureau of Labor Statistics wage survey) anchored). Counts are floors, not totals: our sources undersample some markets, and a title can exist in the wild before it reaches a board we read.