All of the growth in New York AI hiring is actually in New York

Issue one was published on August 18 and called itself weekly. This is issue two, five weeks later. The data never stopped: every company's board has been read every morning since. The writing did. It is back on Mondays from here, and the long gap has one benefit, which is that five weeks shows things one week cannot.

Here is what moved.

1,567 open roles

Across 47 New York AI companies. 759 of those sit in the city itself and 264 are remote-eligible.

Comparing the same 47 companies at both dates, open roles went from 1,547 to 1,567. Twenty roles in five weeks, or 1.3%. (The raw totals say +17, because one small company left our coverage. We give you the like-for-like number, always.)

Flat in total, but not still. 39 of the 47 companies changed their count, and the median move was four roles. The market is turning over underneath a level surface.

The finding: the growth is in the city

Split those same 47 companies by where the job is.

Roles based in New York: 724 to 759. Up 35. Roles anywhere else: 823 to 808. Down 15.

Every bit of net growth in New York's AI companies over the past five weeks was a job in New York. Everything else shrank.

You can see it company by company. Ramp added 12 roles and all 12 are in the city. EliseAI added 11, nine of them here. ElevenLabs, the largest board we track and the least local, cut 20, and its New York count did not move from 10.

It is not universal. Rillet added 13 roles and one of them is in New York. But the direction across the whole set is clear, and it cuts against the assumption that AI companies hire wherever the talent is cheapest. The ones growing right now are growing where their offices are.

If you want an in-person AI job in this city, the market got better for you this month while the headline number went nowhere.

The go-to-market lead widened

Last issue ended on a question: was go-to-market out-hiring engineering a blip?

It was not. 552 open go-to-market roles against 478 in engineering. Five weeks ago it was 527 against 474. The lead went from 53 roles to 74.

Go-to-market added 25 roles. Engineering added four. Every other function combined lost 12.

The constraint at these companies is still distribution, not construction. The engineering title that lives on that boundary, forward deployed engineer, stands at 45 open roles, twice the size of AI engineer, solutions architect, and developer relations put together.

Who is hiring hardest

The two largest boards both shrank. The two most local large boards both grew. Further down, Rillet went from 24 roles to 37 and Spade from 4 to 10.

Every open role at these companies is on the live jobs board, refreshed daily.

What it pays

From salary ranges posted on New York listings. 92% of NYC postings now carry one, up from 85% five weeks ago, so this table rests on more data than the last one did.

Engineering is flat: $215,000 against $217,500. By level, staff and principal engineers post a $250,000 median against $207,500 for senior.

Some medians moved more than that, and we would not read a raise or a cut into any of them. Operations is up $15,000 and go-to-market is down $7,500, but these are medians over whichever roles happen to be open. When a company posts six senior roles, the median rises and nobody got paid more. Watch engineering, which has the most data and barely moved.

Base salary only. The NYC AI engineer salary guide has the bands by level.

The top of the ladder thinned

Companies added mid-level and senior roles and pulled back on staff, management, and leadership: 33 fewer openings across those three. One reading is that organizations made their senior hires over the summer and are now filling in beneath them. Five weeks is too short to be sure.

Entry level did not benefit. 49 entry-level roles against 174 at staff and principal, still three and a half to one. The honest routes in are still lateral ones, and they are in how to get into AI.

What they are building with

Tools named in job descriptions, counted by how many of the 47 companies mention each:

Anthropic still leads OpenAI, 24 companies to 20, but the gap closed from six to four. The mover is further down: LangChain went from 5 companies to 9, and its sibling LangGraph from 4 to 6. Agent frameworks are entering job descriptions quickly.

New this issue: the skills, not just the tools

As of this week we read each posting for what it asks people to do, not only which products it names. The first count:

• AI agents: 37 of 47 companies
• Evals: 25
• RAG: 20
• Fine-tuning: 12
• PyTorch: 11

Three times as many companies want someone to build agents as want someone to train a model. That is the same story the go-to-market numbers tell, from the engineering side: the work is in applying models, not making them. Every term, with who is asking for it and what those roles pay, is in the new AI skills glossary.

A correction to our own counting

While building that, we found a flaw in how we counted tools. Several companies name a vendor in the introduction they paste into every job post, as a customer or an investor. We were counting that once per posting. On the day we fixed it, OpenAI went from 138 postings to 37, and Anthropic from 117 to 64.

The number we publish, companies rather than postings, barely moved: four tools lost one company each, which is why Anthropic reads 24 above rather than 25. We now strip each company's repeated text before reading a posting, and the older posting counts are marked as not comparable. No figure in issue one depended on them. We would rather tell you than have you find it, and it is the first entry in our public corrections log.

Time to fill: a first honest read

Last issue we had two closures. We now have more than five hundred, which is enough to say something, carefully.

Take every role that first appeared in the week of August 16, at companies we were already tracking. There were 115. Four weeks later, 25% of them had come down. 13% were gone within two weeks. As of today, 82 of the 115 are still open.

So roughly seven in ten AI roles posted in New York five weeks ago are still looking. If you are told a role is about to close, the base rate says you have more time than that.

Two cautions. "Came down" means the posting disappeared, which covers filled, cancelled, and reposted under a new link, and we cannot tell those apart. And this is one week's cohort. We will add a cohort every week, and the number will firm up or move.

---

Next week: whether the in-city growth holds, and the second cohort.