Inside the AI Revolution: LinkUp Job Data Shows Steady Hiring Overall, with a Different Mix of Jobs

• Independent confirmation, same data. A 2026 study by Khosravi & Liu, built on the same LinkUp dataset (77.8M postings, 186 countries), finds firms changed what they hire for more than how much.

• The shift is in the mix of jobs, not the total. AI isn't erasing jobs across the board, it's reallocating them: automating routine work while creating new roles built around deploying AI. Total daily active job listings in the US have held at a 7–8M band since early 2022; what's moving is the composition underneath.

• Routine roles are shrinking. The work that's easiest to automate is contracting first. Entry-level listings are down ~70% from their 2022 peak, and customer service rep listings are down ~30%.

• AI-implementation roles are growing. Postings for forward-deployed engineers—those hired to put AI to work across an organization—jumped 15x+ in the last year, from ~300 listings to over 5,000, under titles like "Senior AI Solutions Deployment Developer" and "Applied AI Engineer."

• The role isn't new, its use is. Palantir built the forward-deployed engineer in the mid-2000s to embed engineers inside hard-to-serve clients; AI companies revived it because AI doesn't plug in on its own across an organization, it has to be architected into a company's data, workflows, and broader strategic outlook.

In a March 2026 working paper, Generative AI and Firm Hiring Demand, economists Faezeh Khosravi and Elaine Liu (Georgia State University; Liu is also affiliated with NBER) analyze 77.8 million online job postings spanning 37,395 firm-country pairs across 186 countries from 2021 through mid-2025, using LinkUp job posting data.

Their headline finding, in plain terms: after the emergence of ChatGPT, firms changed what they hire for more than how much. The most AI-exposed employers didn't necessarily post fewer jobs; they cut the share of roles that are easiest to automate and shifted their remaining openings toward non-routine, communication-heavy, judgment-heavy work. 

Our Compass data visualization platform shows the same pattern at the occupation level. 

All Monthly Active Job Listings Since 1/1/2022

Across the entire job market, hiring hasn't fallen off a cliff since the advent of generative AI. Total active listings peaked near 9 million in the early 2022 tech hiring bonanza and have held in a 7-to-8 million band ever since—below the peak, but nowhere near a broad collapse. If AI were simply destroying jobs across the board, this line would be falling. For the time being, the change is happening within the total, in which kinds of jobs get posted: the volume is steady, the mix is changing.

Roles in decline

Entry-level Active Listings Down Roughly 70% from their 2022 Peak

Entry-level listings are the clearest casualty. Active postings tagged in our dataset as entry-level positions climbed above 60,000 at the early-2022 hiring peak, recovered to the mid-50,000s in 2024, and have since dropped below 20,000 by early 2026: roughly a 70% fall from peak.

Active Listings for Customer Service Reps: Down About a Third from 2022 high, then Flat

Customer service is a milder version of the same story. Daily active listings ran near 150,000 in early 2022, fell to around 100,000, and have stayed in a 100,000–115,000 band since. That’s about a 30% drop since an early-2022 peak, with the steepest decline coming after ChatGPT's launch on November 30, 2022. 

Both job types are what the Khosravi–Liu paper calls high-substitution occupations: routine, codifiable work that a tool producing text, code, and structured answers at near-zero cost can now do. The paper finds that across advanced economies, the probability that an exposed firm posts any vacancy in a quarter fell by more than 10 percentage points after ChatGPT, with the sharpest pullbacks concentrated in exactly these kinds of roles.

The decline isn't uniform. Entry-level postings keep sliding while customer service openings have found a plateau for now. That fits the paper's timing evidence: in advanced economies the volume cuts come first, and the deeper shift in the mix shows up later, over quarters rather than weeks. 

A role that's growing: Forward-Deployed Engineers: Jobs at the Vanguard of the AI Revolution

Forward-Deployed Engineer Listings Up More Than 1,500% Since 2022

While routine roles shrink, one role is climbing fast: the forward-deployed engineer for AI solutions. Active listings including “forward-deployed engineer/architect” in their titles or descriptions were flat at roughly 300 through 2022 and 2023, then rose sharply to more than 5,000 by early 2026, a more than fifteenfold increase, almost all of it in the last year. These are the people companies hire to actually put AI to work across their organizations, and the listings show up under titles like "Senior AI Solutions Deployment Developer" and "Applied AI" roles at Google, Accenture, and Palantir.

The role of a forward-deployed engineer began at Palantir in the mid-2000s. Palantir's early customers were intelligence and defense agencies that couldn't articulate exactly what they needed, couldn't share their data freely, and changed requirements constantly. Ordinary consultants couldn't write production code, and ordinary product engineers couldn't work inside a classified environment. So Palantir sent its own engineers to embed directly with the customer, learning the workflow on site and building the software in real time until it worked. By 2016 the company had more of these forward-deployed engineers than traditional software engineers.

AI companies have revived a similar model. A modern AI model is powerful but doesn't simply plug. It has to be wired into a company's messy data and specific workflows before it produces anything useful. So OpenAI, Anthropic, Google, Databricks, various consultants, and a growing list of startups now hire engineers to embed with customers and drive adoption across the organization: scoping the problem, writing the integration code, handling the edge cases, and feeding what they learn back to the product team. It's the same idea Palantir built, fixed on a new problem—instead of deploying hard-to-configure intelligence software into secretive agencies, these engineers deploy hard-to-configure AI into everyday enterprises. The work is non-routine, customer-facing, and heavy on the communication and judgment that don't compress into a prompt: the opposite of a high-substitution role.

In the appendix of the Khosravi–Liu paper, the study finds that among more AI-exposed firms, postings show rising communication and responsibility content and a measurable shift toward non-routine tasks.

Same data, two directions

Read the charts together and the story is simple. Total hiring is roughly steady. Underneath it, routine roles like entry-level and customer service are shrinking, and roles like the forward-deployed engineer are growing to help companies adopt AI. AI is lowering the value of routine, substitutable work and raising the premium on embedded, hands-on work—and the postings move before the headcount does. Job postings are a forward-looking read on hiring intent, not a lagging record of it.

One caution worth keeping in view: the growing role is not a replacement for the shrinking one. Forward-deployed roles require senior engineering skill plus client fluency, and pay well into six figures, which makes them a poor landing spot for the new graduates watching the bottom rung disappear. The mix of work is being rebuilt, but the entry point doesn’t have a clear path to restoration.

Data: LinkUp active job listings, visualized in Compass. Academic reference: Khosravi, F. & Liu, E. M., "Generative AI and Firm Hiring Demand: Evidence from Advanced and Developing Economies," March 2026.