Neuronify

How we work

Editorial standards

We publish data and writing about the New York AI economy, and we also do business in it. These are the rules that keep the first honest despite the second.

Someone is accountable for everything here

Neuronify was founded by Vik Chadha and Rob May. Every newsletter issue and article carries the name of the person who wrote it, and Vik Chadha is responsible for what this site publishes. Pages built from data, such as the Hiring Index and the skills glossary, are produced by code we wrote and can explain, and the methodology page explains it.

We are operators and investors, not career journalists, and we do not pretend otherwise. What we offer in place of a newsroom is a narrow beat, first-hand data, and these rules.

The business cannot buy the coverage

We run a business alongside this publication: a vetted network that matches companies with AI engineers and consultancies, for a fee paid when an engagement happens. That business has no say in what we publish, and the two are kept apart by rule:

  • No company can pay to be included in, ranked higher in, or removed from any list or dataset.
  • Whether a company is a client, a prospect, a show guest, or none of those has no effect on how it is counted or described.
  • Nobody we write about sees a piece before it is published, and nobody gets a veto.
  • We do not sell links, and we do not publish paid material without labeling it as paid.

The relationships that could put pressure on those rules, including our founders’ other roles and investments, are set out in our disclosures.

Primary sources, linked

When we cite a number, a law, or an enforcement action, we read it in the original: the regulator’s order, the statute, the company’s filing, the survey’s own report. We link to that document so you can check us. If we could only find a claim in secondary coverage, we either say so or leave it out.

We keep the base of every statistic attached to it. “97% lacked access controls” means something different when the 97% is of all companies than when it is of the few that suffered a breach, and most misquoted numbers went wrong by losing that detail. We name the tracker when trackers disagree.

We distinguish allegations from findings. A person who has been charged is described as charged, and a settled matter is described as settled. When a well-known story turns out to be false, we say that rather than repeat it, as we did with two of them in our guide to AI washing.

Rules for our own data

Most of what is original here comes from reading the public job boards of New York’s AI companies every morning. The full method is on the methodology page. The rules that matter most:

  • Companies before postings. One employer with two hundred openings must not set the market, so we count how many companies do something before how many postings do.
  • Like-for-like change only. When a company enters our coverage its whole board arrives at once. We report change across the companies present at both dates, and show the raw difference only so you can see the gap.
  • No number from thin data. We do not publish a pay median from fewer than eight posted ranges. A blank is more honest than a precise-looking guess.
  • Say what the number is. A tool named in a job posting is a tool an employer asks for. It is not proof of what runs in production, and we never describe it that way.
  • Dates as they are. If the latest figure for a company is two years old, we print the date next to it instead of implying it is current.

How we use AI

We cover AI and we use it. AI tools help us research, draft, and write the code behind this site, including the pipelines that produce the data. We think a publication on this subject should say so plainly.

What AI does not do here is decide what is true. Every figure is checked against its source or against our own data before it is published. We do not publish quotes, sources, or statistics generated by a model, and when research assistance surfaces a claim, a claim is all it is until we have read the primary document. A named person reviews everything before it goes out and answers for it afterward. If a piece is wrong, “the model wrote it” is not a defense we will offer.

Naming companies and people

Our company data comes from what companies publish themselves: job boards, filings, announcements. We do not publish information about private individuals, and we do not collect personal data from job postings. When we describe what a person said on a show or in a filing, we describe it as closely as we can and link to it.

If we have written about you or your company and you believe a fact is wrong, tell us. We will check it and, if you are right, fix it and log the correction.

Dates and updates

A date on this site means the content was written or materially revised on that date. We do not refresh dates to look current. Pages built from daily data show the date of the data. Guides that depend on laws or figures that change carry an “updated” line, and when we revise one for substance we change that line and nothing earlier.

Hold us to it

If something here falls short of these standards, we want to hear it.

Report an errorRead our disclosures