Zendesk Layoffs 2026: CX Software Engineering Map

Last updated: August 4, 2026

By Robert Ardell, Co-Founder and Strategic Advisor, KORE1

Zendesk is not running a company-wide engineering layoff in 2026. Global headcount grew roughly 10 percent last year to about 6,821 people, and engineering is still the largest function at around 45 percent of the company. The cuts that did land were regional and support-side, led by roughly 350 Irish roles in January. What is moving is the skill mix underneath job titles that have not changed at all.

Two numbers, both true, published two weeks apart.

Zendesk cut about 350 jobs in Ireland this past January, roughly half of what it had in Dublin, first reported by the Business Post. Zendesk also closed 2025 with about 6,821 employees, up 10.3 percent year over year, according to workforce data from Revelio Labs. Nobody is lying in either direction. A company can shrink hard in one country and grow in aggregate, and when it does, the word “layoffs” stops describing anything useful.

So the search term is a blunt instrument. The instinct behind it is not.

Something genuinely did change in customer experience software over the last eighteen months, and it changed what a CX engineer is worth and which ones you can now hire. That is the part worth 2,000 words. If you want the cross-industry version first, our 2026 tech layoffs roundup tracks displaced talent across thirty-plus companies. This page is narrower. It is about reading a Zendesk, Intercom, Freshworks, or Salesforce Service Cloud resume correctly in 2026, and knowing which of those people your req actually needs.

Where I sit, stated plainly so you can discount it. I co-founded KORE1 in 2005 and we place engineers for a living, which means a hire you make from this bench can end up on our invoice. I have no financial stake in which of the six profiles below you pick, and one section argues that two of them are overpriced right now relative to what most buyers get out of them. That section stays in.

What Actually Happened, and What Did Not

Start with the paper trail, because it is short and it settles most of the argument.

Zendesk has filed three WARN notices in California since going private, per WARNTracker. November 2022, 84 workers. May 2023, 64 workers. February 2025, 51 workers. That is 199 people across two and a half years in the state where the company is headquartered. Not a bloodbath. The only broad, announced cut was the 8 percent reduction in May 2023, which CEO Tom Eggemeier attributed to hiring that had outpaced the business between 2020 and 2022.

Then January 2026, and Ireland. That one was not announced with a press release, and it was not proportional. Halving a country’s headcount while the global number climbs is a relocation, not a retrenchment.

Follow the geography and it gets obvious. Roughly 35.2 percent of Zendesk employees sit in the United States. India holds about 8.9 percent. The Philippines about 7.0. Both of those shares have been climbing for three years straight. Dublin used to be the EMEA support and go-to-market engine for a lot of American SaaS companies. In 2026, a growing share of tier-one support volume does not need a person in Dublin at all, because it is being resolved by software, and the human work that remains is being consolidated closer to the engineering centers that build the software.

Every SaaS company with a Dublin support floor is running some version of this play. Zendesk is just where you can watch it happen, because Zendesk sells the thing doing the replacing.

The Pivot That Reprices Everyone on the Org Chart

Here is the shift, and it is bigger than a headcount story.

At its Relate 2026 conference in May, Zendesk moved its AI Agents to outcome-based pricing. Customers pay per verified resolution instead of per seat. Forrester’s analysts covered the announcement in a detailed write-up of the event and noted the company has completed more than 15 acquisitions worth over 500 million dollars since the 10.2 billion-dollar take-private led by Hellman & Friedman and Permira in 2022. Klaus for quality assurance. Tymeshift for workforce management. Ultimate for AI agents. Local Measure for voice. HyperArc for analytics. Unleash for knowledge retrieval. Forethought for self-learning agents.

Read that list again as a hiring document rather than a press release, because that is what it is. Every one of those deals dropped a team, a codebase, an auth model, and a data pipeline into an organization that now has to make all of it behave like one product. Somebody writes all that glue. That integration burden is where a meaningful share of Zendesk engineering headcount currently sits, and it is the single most transferable skill coming off this bench.

Zendesk’s president of Product, Engineering and AI, Shashi Upadhyay, framed the endgame for TechCrunch in October 2025. “The world’s going to shift from software that’s built for human users, to a system where AI actually does most of the work.” The company claims its autonomous agent handles 80 percent of support issues without a human. Treat vendor resolution rates with the skepticism they deserve. Even discounted heavily, the direction is not ambiguous.

Forrester put the workforce consequence more precisely than anyone at the conference did. Human agents, they wrote, are becoming service engineers who design and curate the work that AI agents then execute. Which is a polite way of saying a large category of jobs is being converted from doing the work to specifying the work, and the people who can make that jump are worth considerably more than they were paid last year.

The macro data agrees, quietly and without drama. The Bureau of Labor Statistics projects customer service representative employment to decline 5 percent from 2024 to 2034, against a median wage of 20.59 dollars an hour as of May 2024. Around 2.4 million Americans hold that job today. That is the scale of it. A 5 percent decline sounds mild until you sit with the fact that it is the first sustained contraction in a job category that absorbed entry-level workers for forty years.

Six Engineering Profiles Coming Off the CX Software Bench

A generic recruiter files all of these under “backend engineer, SaaS.” That is how you end up with four weeks of wrong resumes. Inside the sector the lines are sharp, and they map to six genuinely different buyer markets.

1. Ticketing-Core and Rails Platform Engineers

Zendesk is one of the largest production Ruby on Rails deployments on earth, sitting on MySQL, Redis, and Kafka, running on Kubernetes over AWS. The engineers who maintain that core do multi-tenant work at a scale almost nobody gets to practice on. Sharding strategies. Noisy-neighbor throttling. Schema migrations against tables with tens of billions of rows and a customer base that notices a 200-millisecond regression.

They are undervalued right now and I will say so directly. Rails carries an unfair reputation as a legacy skill, which means a senior Zendesk platform engineer often prices 15 to 20 percent below an equivalent Go or Java engineer with strictly less operational scar tissue. Same work. Lower number. If your product runs on Rails, this is the cheapest genuinely senior talent available in 2026.

2. Conversational AI and Agent Engineers

The Ultimate and Forethought lineage. Intent classification, LLM orchestration, retrieval over a customer’s knowledge base, and the part nobody puts on a job posting, which is building the evaluation harness that proves the agent did not confidently invent a refund policy. This profile is the most expensive on the list and the most likely to be mis-hired, because the title is three years old and half the resumes claiming it describe someone who wired up an API key and shipped a chatbot.

How you tell them apart. Ask what their containment rate was, then ask how it was measured, then ask what the false-resolution rate looked like. Someone who has done this work has a strong opinion about the second question and a painful story about the third.

3. Knowledge and Retrieval Engineers

Scarce. Genuinely scarce, in a way the market has not caught up to.

Eggemeier put a number on the problem at Relate that I have quoted in probably a dozen client calls since. Across the customer base Zendesk studied, only 27 percent of knowledge artifacts were actually deployment-ready, and another 28 percent were what he called unlockable resolutions, blocked by gaps that required converting narrative documentation into encoded form. Every company buying an AI agent right now is discovering the same thing about its own help center. The engineers who fix it work on permission-aware search, embedding pipelines, chunking strategy, and content governance. Six people I would consider truly expert at it have crossed my desk in the last year. All six closed in under three weeks.

4. Voice and Contact-Center Engineers

Local Measure brought Zendesk a CCaaS product built on Amazon Connect, and with it a bench that does SIP, WebRTC, real-time streaming transcription, and telephony at a level web engineers find genuinely alien. Narrow receiving market. Intense within it. Every company that decided its AI agent needs to answer a phone call rather than a chat widget is competing for a few hundred people nationally, and the ones with both Amazon Connect and streaming-ASR experience can name a number.

5. Integration and API Platform Engineers

Seven acquisitions in roughly 28 months. Somebody unifies the auth.

This profile owns webhooks, marketplace apps, OAuth flows, event streams, and the migration paths that let an acquired product’s customers keep working while their backend is rebuilt underneath them. It is unglamorous and it transfers to almost everything. A CX integration engineer walks into a fintech, a healthcare platform, or an ERP shop and is useful in week two, which is why they are the profile most likely to leave the CX sector entirely. If you have a messy multi-product roadmap and a title you can flex, this is where I would spend the req.

6. Analytics and Data Engineers

The HyperArc line of work. Semantic layers, generative analytics over service data, and the reporting infrastructure that has to answer a question like “which of these resolutions actually stuck” when the pricing model now bills against that answer. Outcome-based pricing turns an analytics team into a revenue system. That has not fully priced in yet. Wait twelve months.

There is a seventh group and it does not have a title anyone agrees on. Former support leads who now write the decision trees, curate the knowledge base, and own the AI agent’s behavior. Forrester calls them service engineers. Zendesk does not consistently call them anything. They are not software engineers and should not be paid as though they were, but they are also not tier-two support and paying them like that is how you lose them in month five. Most of the companies I talk to have not built the job description yet.

Where This Bench Is Actually Landing

What follows is what our desk has watched over the past nine months across CX and SaaS platform searches. It is not a public dataset and it should be read as directional, not as a survey.

The spread in that last column is the whole strategic point. Two of these profiles clear in under three weeks and four do not, which means a buyer who runs one uniform interview loop across all six is going to lose the two that matter most and win the four that were never contested.

What the Money Looks Like

Anchor on the company itself first, because it is the cleanest public data available. Levels.fyi puts median total compensation for a Zendesk senior software engineer at 176,866 dollars and a staff engineer at 234,258 dollars, with the full software engineer range running from about 152,000 to north of 401,000 at the top of the distribution.

Now the messy part. Salary aggregators disagree violently about the AI-adjacent titles, and the disagreement is not noise. It is a sampling problem you need to understand before you set a band. Glassdoor reports an average of 123,167 dollars for “conversational AI engineer” with a 25th-to-75th range of 92,375 to 168,031. ZipRecruiter reports 101,752 dollars for “AI engineer” as of July 2026. Both are pulling heavily from postings that use the title loosely, including plenty of prompt-tuning and support-automation roles that are not engineering jobs at all. The senior people who actually built and shipped a production agent do not appear in those averages in any meaningful number, and if you build a band off them you will lose every finalist you like. Every one.

Our own observed bands for the six profiles, nationwide, base salary only, excluding equity. Directional.

Two honest caveats on that table. San Francisco, New York, and Seattle run 10 to 18 percent above these numbers and most of the Midwest runs below them, so treat the bands as a national midpoint rather than a quote. Not a quote. And the conversational AI row is the one I would argue is overpriced today, along with the top of the knowledge row, because a chunk of that premium is buying a resume keyword rather than shipped production experience. If you can interview well enough to tell the difference, you can hire at the bottom of that band and be happier than the buyer paying the top of it. If you cannot, pay the premium, because the failure mode on the other side is an agent that hallucinates policy to your customers.

For a broader view of what AI-titled engineering roles pay outside the CX category, our AI engineer salary guide breaks down real offer data by level, and the salary benchmark assistant will pull a band for a specific title and market.

If You Are Hiring Against This Bench

Fix the job title before you fix anything else. “Senior Software Engineer, Customer Experience” gets you nobody from the two profiles you actually want, because the people in those profiles have stopped reading generic postings entirely. “AI Agent Engineer, retrieval and evaluation” gets you the right inbound and screens out the wrong one in the first sixty seconds. That is not a copywriting note. It is most of the difference between a req that fills in three weeks and a req that sits until Q4.

Second thing. Interview for the failure mode, not the happy path. Anyone can demo an agent answering a question it has an article for. Ask what happens when the knowledge base contradicts itself, when the customer is angry and the policy says no, when the model is confident and wrong. Candidates who have shipped this will light up at the question. Candidates who have not will describe the happy path again, louder. Listen for that.

Third one. Almost nobody does this. If you are hiring a service engineer, the person who curates the knowledge base and owns the agent’s behavior, write that job description before you post the engineering req. Most companies do it backwards, hire the engineer first, and discover four months later that they built a very good agent on top of documentation nobody owns. The 27 percent readiness number is not a Zendesk problem. It is going to be your problem.

The last note is about timing, and it is the only genuinely urgent thing on this page. Conversational AI and knowledge-retrieval people are closing in eight to twenty-one days. If your loop takes five weeks and requires four onsites and a take-home, you are not going to hire one, and no amount of budget fixes a calendar problem. Compress the loop for those two profiles specifically. Leave it alone for the other four.

KORE1 has been placing technology talent since 2005, across more than 30 U.S. metros, with recruiters who average over 15 years on the desk. Our 2026 IT and engineering placements are averaging 17 days to hire, and 92 percent of them are still in seat at the twelve-month mark. Most of this work runs through our software engineer staffing practice, with the model-specific searches handled by the AI and ML engineer staffing desk and the majority placed as direct hire rather than contract. If you want to pressure-test a CX engineering req before it goes live, talk to a recruiter and we will tell you which of the six profiles you are actually describing. Sometimes the answer is that you already employ the person and should promote them.

Questions Hiring Managers Keep Asking

So did Zendesk actually lay people off in 2026 or not?

Regionally, yes. Roughly 350 Irish roles went in January 2026, about half of the Dublin workforce, while global headcount grew about 10.3 percent to 6,821. The company has filed only three California WARN notices since 2022, covering 199 workers total.

The distinction matters because it tells you what kind of candidate is available. A company-wide engineering cut floods the market with a broad mix. A geographic consolidation plus an AI pivot releases a specific slice, mostly support and support-adjacent, while the engineering bench stays employed and only moves when somebody outbids.

Is the AI agent thing real, or is it a vendor slide?

Both, and the ratio is improving. Zendesk claims 80 percent autonomous resolution and reported roughly 200 million dollars in AI ARR at the close of 2025 with a 2026 target as high as 500 million. Discount the resolution rate. The revenue is harder to argue with.

The tell that this is not purely marketing is the pricing change. A vendor that genuinely doubts its own automation does not move to billing per verified resolution, because that model only works if the software actually resolves things. Forrester’s own critique was different and worth reading. Their analysts flagged that Zendesk risks looking like a collection of acquired tools rather than one integrated platform. That is an integration problem, not a demand problem.

We run on Rails. Is a Zendesk platform engineer a good fit for us?

Best value on this list, and it is not close. Zendesk runs one of the largest production Rails deployments anywhere, on MySQL, Redis, and Kafka over Kubernetes. Engineers from that core bring multi-tenant scaling experience most Rails shops cannot develop internally.

The catch is expectations. Someone who has spent six years inside a platform where a bad migration pages three teams will ask hard questions about your deploy process and your on-call rotation, and if the honest answers are embarrassing, they will pass. Have those answers ready. It is a good filter for you too.

How do I tell a real conversational AI engineer from someone with the title?

Ask for their containment rate, how it was measured, and what the false-resolution rate was. Real practitioners have a strong opinion on the measurement question and an uncomfortable story about the false positives. Keyword candidates restate the happy path.

A second filter that works well. Ask how they handled evaluation before shipping, and listen for whether there was a golden dataset, a regression suite, or a human review loop, or whether the answer is essentially that they tried it a few times and it seemed fine. That one question has saved several of our clients a full quarter.

What about the support people, not the engineers?

The strong ones become service engineers, designing and curating what the AI executes. BLS projects a 5 percent decline in customer service representative employment from 2024 to 2034, with about 341,700 annual openings still turning over. The floor is eroding while the ceiling rises.

If you have a tier-two lead who already writes your macros, owns your help center, and argues with product about policy edge cases, you are looking at the person who should own your AI agent. Promote them. It costs less than a search, they already know your customers, and the alternative is watching a competitor hire them into exactly this role in about eight months. I would rather tell you that than sell you a req.

Is now a good window, or should we wait?

Now, for four of the six profiles. Rails platform, voice, integration, and analytics engineers are all reachable at reasonable numbers today. The conversational AI and knowledge-retrieval profiles are not a window at all. They are a permanent seller’s market, and it keeps getting worse.

Waiting on the first four costs you a little. Waiting on the last two costs you a lot, because that price curve has gone one direction for two straight years and nothing in the acquisition activity or the pricing shift suggests it flattens in 2027.

Does KORE1 work with candidates coming out of this market, or only with employers?

Both sides. We fill buyer-side reqs across all six profiles and we represent CX and SaaS platform engineers at no cost to the candidate. The hiring company pays the fee, structured against a refund clause if the placement does not hold.

Engineers reading this from inside Zendesk, Intercom, Freshworks, or a Salesforce Service Cloud team should send a note through the contact form with the profile you fit and what you have actually shipped. We route to the right desk within a day. For the adjacent SaaS picture, our HubSpot layoffs breakdown maps where that bench has been going.