--- title: "40 Percent of AI Agent Projects Fail and Companies Are…" canonical: "https://www.metaintro.com/blog/ai-agent-projects-fail-companies-rehiring-grads-2026" language: "en" author: "drashtigarach" published: "2026-07-18T15:02:16.000Z" modified: "2026-07-18T16:15:14.102Z" --- [Back to Blog](/blog) [AI](/blog/tag/ai)[Hiring](/blog/tag/hiring)[Jobs](/blog/tag/jobs)[News](/blog/tag/news) # 40 Percent of AI Agent Projects Fail and Companies Are Rehiring Grads in 2026 Gartner says 40% of agentic AI projects will be canceled by 2027 as firms like Klarna rehire humans and founders train grads. Here is what it means for you. [![Drashti Garach](https://cdn.metaintro.com/rs:fill:40:40/q:72/plain/images/5719d740-e510-42bc-8017-e040d145f35f_1766029465094.png)Drashti Garach @DrashtiGarach](/blog/author/drashtigarach) [July 18, 2026](/blog/archive/2026/07)13 min read ![40 Percent of AI Agent Projects Fail and Companies Are Rehiring Grads in 2026](https://cdn.metaintro.com/rs:fill:1200:675/q:78/plain/images/5486-1.png) [https://x.com/intent/tweet?text=40%20Percent%20of%20AI%20Agent%20Projects%20Fail%20and%20Companies%20Are%20Rehiring%20Grads%20in%202026&url=https%3A%2F%2Fwww.metaintro.com%2Fblog%2Fai-agent-projects-fail-companies-rehiring-grads-2026](https://x.com/intent/tweet?text=40%20Percent%20of%20AI%20Agent%20Projects%20Fail%20and%20Companies%20Are%20Rehiring%20Grads%20in%202026&url=https%3A%2F%2Fwww.metaintro.com%2Fblog%2Fai-agent-projects-fail-companies-rehiring-grads-2026)[http://www.facebook.com/sharer.php?u=https%3A%2F%2Fwww.metaintro.com%2Fblog%2Fai-agent-projects-fail-companies-rehiring-grads-2026](http://www.facebook.com/sharer.php?u=https%3A%2F%2Fwww.metaintro.com%2Fblog%2Fai-agent-projects-fail-companies-rehiring-grads-2026)[https://www.linkedin.com/sharing/share-offsite/?url=https%3A%2F%2Fwww.metaintro.com%2Fblog%2Fai-agent-projects-fail-companies-rehiring-grads-2026&title=40%20Percent%20of%20AI%20Agent%20Projects%20Fail%20and%20Companies%20Are%20Rehiring%20Grads%20in%202026](https://www.linkedin.com/sharing/share-offsite/?url=https%3A%2F%2Fwww.metaintro.com%2Fblog%2Fai-agent-projects-fail-companies-rehiring-grads-2026&title=40%20Percent%20of%20AI%20Agent%20Projects%20Fail%20and%20Companies%20Are%20Rehiring%20Grads%20in%202026)[mailto:?subject=40%20Percent%20of%20AI%20Agent%20Projects%20Fail%20and%20Companies%20Are%20Rehiring%20Grads%20in%202026&body=https%3A%2F%2Fwww.metaintro.com%2Fblog%2Fai-agent-projects-fail-companies-rehiring-grads-2026](mailto:?subject=40%20Percent%20of%20AI%20Agent%20Projects%20Fail%20and%20Companies%20Are%20Rehiring%20Grads%20in%202026&body=https%3A%2F%2Fwww.metaintro.com%2Fblog%2Fai-agent-projects-fail-companies-rehiring-grads-2026) Two stories broke this month that job seekers should read together. [Inc](https://www.inc.com/daniel-robbins/it-was-faster-to-train-a-grad-an-ai-pioneer-explains-why-half-her-agents-failed/91374578) published an interview under the headline "It Was Faster to Train a Grad," in which an AI pioneer explained that roughly half of the AI agents she built failed and that coaching a recent graduate was quicker than repairing the software. Around the same time, research firm [Gartner](https://www.gartner.com/en/newsroom/press-releases/2025-06-25-gartner-predicts-over-40-percent-of-agentic-ai-projects-will-be-canceled-by-end-of-2027) put a number on the pattern, forecasting that more than 40% of agentic AI projects will be scrapped by 2027. At [Metaintro](https://www.metaintro.com), we track where the jobs actually are across tens of millions of live postings, and the signal is clear. The agent hype is colliding with reality, and human hires, especially cheap trainable ones, are quietly coming back into demand. If you are early in your career, this correction is your opening. ## Why are 40 percent of AI agent projects about to be canceled? The most concrete number in this story comes from [Gartner](https://www.gartner.com/en/newsroom/press-releases/2025-06-25-gartner-predicts-over-40-percent-of-agentic-ai-projects-will-be-canceled-by-end-of-2027), which surveyed more than 3,400 organizations that are actively investing in agentic AI. Its conclusion, [reported](https://martech.org/gartner-40-of-agentic-ai-projects-will-fail-making-humans-indispensable/) in detail by MarTech, is that over 40% of those projects will be canceled by the end of 2027 because of escalating costs, unclear business value, or inadequate risk controls. That is not a fringe prediction from an AI skeptic. It is the mainstream analyst view, drawn from the companies doing the spending. What makes the finding land harder is the reason behind it. According to Anushree Verma, a senior director analyst at Gartner, "Most agentic AI projects right now are early-stage experiments or proof of concepts that are mostly driven by hype and are often misapplied." In plain language, a lot of the software was deployed before anyone knew what problem it was supposed to solve. Gartner even coined a term for the marketing around it, "agent washing," and estimates that of the thousands of vendors claiming to offer agentic products, only around 130 deliver genuine autonomous features. For workers, the lesson is that a headline about AI replacing a function is often a proof of concept, not a finished replacement. The gap between the two is where jobs still live. ## What does the Klarna reversal tell job seekers? The clearest real-world example of an agent rollout hitting a wall is [Klarna](https://www.entrepreneur.com/business-news/klarna-ceo-reverses-course-by-hiring-more-humans-not-ai/491396), the buy now, pay later company valued at $14.6 billion. In February 2024 Klarna [claimed](https://www.entrepreneur.com/business-news/klarna-ceo-reverses-course-by-hiring-more-humans-not-ai/491396) that its AI assistant could do the work of 700 customer service agents and had captured 75% of customer chats, roughly 2.3 million conversations in a single month. It became the poster child for the idea that agents could gut a support team overnight. Then the quality problems showed up. By 2025, chief executive Sebastian Siemiatkowski was reversing course. He told the press the AI chatbots were cheaper than humans but produced "lower quality" output, and that independent testing showed the bot mostly filtered customers toward human support rather than resolving their issues. Klarna reduced its headcount from about 5,500 workers to roughly 3,000, then began [rehiring people](https://techcrunch.com/2025/06/04/klarna-ceo-says-company-will-use-humans-to-offer-vip-customer-service/) for the work the agents could not do well. "We think offering human customer service is always going to be a VIP thing," Siemiatkowski said, and to [Entrepreneur](https://www.entrepreneur.com/business-news/klarna-ceo-reverses-course-by-hiring-more-humans-not-ai/491396) he framed the fix bluntly, saying investing in the quality of human support is the way of the future. For anyone applying to support, operations, or client-facing roles, Klarna is proof that the pendulum swings back, and that the [companies rehiring workers replaced by AI](https://www.metaintro.com/blog/companies-rehiring-workers-replaced-by-ai) are not rare exceptions. ## Why was it faster to train a grad than fix the agents? The founder featured by [Inc](https://www.inc.com/daniel-robbins/it-was-faster-to-train-a-grad-an-ai-pioneer-explains-why-half-her-agents-failed/91374578) landed on the same conclusion from the builder side rather than the buyer side. After watching about half of the agents fail, the calculation flipped. A recent graduate could be onboarded, corrected, and made productive faster than a brittle software agent could be debugged, re-prompted, and monitored. That is not a sentimental argument about the human spirit. It is an economics argument about total cost, and it is one job seekers can use. Think about what actually goes into a working AI agent. Someone has to scope the task, connect the systems, write the guardrails, test the edge cases, catch the failures, and keep watching it after launch so it does not quietly drift off the rails. A new hire, by contrast, absorbs context, asks questions, flags when something looks wrong, and improves without a specialist rebuilding them every week. When a task is fuzzy or changes often, the trainable human wins on speed to competence. There is a hidden cost to the failed agents, too. Every one of them still needs a person to notice it broke, apologize to the customer it annoyed, and clean up the mess before the next attempt. Gartner warns that in 2026, one-third of companies will damage customer experiences by deploying AI prematurely, which means the reputational bill for a bad agent often lands right back on a human team. Once you price in that supervision, monitoring, and cleanup, the case for a graduate who simply does the work correctly the first time gets stronger, not weaker. This is why understanding [the real impact of AI on jobs](https://www.metaintro.com/blog/read-ai-real-impact-jobs-seven-signals) matters more than reacting to the scariest headline. The founders closest to the technology are the ones quietly concluding that people are still the faster path, and they are backing that conclusion with payroll. ## Why do AI agents keep failing at real work? There is now hard research explaining why agents stumble on exactly the collaborative, messy work that fills most jobs. A [Stanford](https://hai.stanford.edu/news/ai-coding-agents-fail-at-teamwork) study called CooperBench, presented at an April 2026 workshop, tested more than 650 real software engineering tasks across four languages, Python, TypeScript, Go, and Rust. The finding was counterintuitive. When two AI agents were asked to split the work, they performed significantly worse than a single model working alone. "Today's best coding agents lose nearly half their capability when paired up," said Diyi Yang, an assistant professor of computer science, who added that "social intelligence, not coding skill, is the key bottleneck for AI collaboration." First author Hao Zhu put it even more simply, saying "a single model is better than two agents sharing the work." The researchers described a "coordination gap," where agents could not reliably figure out where to make changes versus what changes were needed, and giving them the ability to message each other barely helped. They ignored warnings, left questions unanswered, and dropped tasks they had promised to do. That is a precise description of the human coordination work, the standups, handoffs, and clarifying questions, that entry-level employees learn to do naturally. The skills that make agents fail are the ones that make junior humans valuable. ## Which employers are betting on inexperienced hires right now? Some founders are not waiting for the correction to arrive. [Fortune](https://fortune.com/2026/04/06/while-other-ceos-freeze-entry-level-roles-this-ai-founder-is-hiring-gen-z-with-zero-experience/) profiled Alon Chen, founder and chief executive of the generative AI platform [Tastewise](https://fortune.com/2026/04/06/while-other-ceos-freeze-entry-level-roles-this-ai-founder-is-hiring-gen-z-with-zero-experience/), which has raised $71.6 million and works with more than half of the Fortune 100 food and beverage companies, including [PepsiCo](https://fortune.com/2026/04/06/while-other-ceos-freeze-entry-level-roles-this-ai-founder-is-hiring-gen-z-with-zero-experience/), [Nestle](https://fortune.com/2026/04/06/while-other-ceos-freeze-entry-level-roles-this-ai-founder-is-hiring-gen-z-with-zero-experience/), and [Mars](https://fortune.com/2026/04/06/while-other-ceos-freeze-entry-level-roles-this-ai-founder-is-hiring-gen-z-with-zero-experience/). While other chief executives freeze entry-level roles, Chen actively recruits Gen Z candidates with no experience and no degree requirement. His reasoning is worth quoting. "There are some positions where you actually want people that do not have the prejudice or the old way of working," Chen told Fortune. "When you come as someone who just sees the problem and finds the best way to solve it, it's sometimes better than someone who has been doing the same job for so long." He describes young hires as "AI natives themselves, having been born and raised in this new realm of opportunities," and points them at pivotal roles like customer insights, where they move fluidly between technology, business, and client work. Chen is not a soft touch on talent, either. Fortune notes he founded a computer sales business in Israel at 15 and later became [Google](https://fortune.com/2026/04/06/while-other-ceos-freeze-entry-level-roles-this-ai-founder-is-hiring-gen-z-with-zero-experience/)'s chief marketing officer at 28 without a marketing degree. His bet on the untrained is a bet on adaptability, and it is a live example of the [employers who value skill over pedigree](https://www.metaintro.com/blog/ai-skills-gap-grade-vs-skill-employers-2026). ## What skills actually make you the trainable hire? If trainability is the asset, the practical question is how to signal it. The good news is that the very traits that break AI agents map to teachable human behaviors. Agents fail at coordination, at flagging uncertainty, and at following through on promised tasks. So the candidate who communicates clearly, asks the right clarifying question, and reliably closes loops is offering exactly what the software cannot. Those are learnable habits, not innate gifts, and they show up in how you handle an application, an interview, and a first project. Being AI-native helps too, but not in the way people assume. Employers are not looking for prompt-engineering party tricks. They want people who can use AI to move faster while still catching its mistakes, which is why the [AI skills employers actually want](https://www.metaintro.com/blog/ai-skills-employers-actually-want) tend to be judgment plus fluency rather than raw technical depth. Pair that with the durable [human skills that get more valuable as AI spreads](https://www.metaintro.com/blog/human-skills-that-get-more-valuable-as-ai-spreads), and you become the hire who is cheaper to train than an agent is to maintain. If you want a structured path, our guides on [staying employable through 2026](https://www.metaintro.com/blog/ai-skills-that-keep-you-employable-2026) and [making your role AI-resilient](https://www.metaintro.com/blog/making-your-role-ai-resilient) turn these ideas into weekly steps you can act on. ## Does this mean the job market is suddenly easy again? No, and it is important to be honest about that. The correction in agent hype does not erase a genuinely tough entry-level market, where many [graduates are taking any paycheck](https://www.metaintro.com/blog/entry-level-grads-any-paycheck-brutal-market-2026) they can get and [young adults are competing hard](https://www.metaintro.com/blog/young-adults-entry-level-jobs-2026-economy) for a shrinking number of first jobs. What is changing is the narrative that AI has permanently closed the door on inexperienced workers. Gartner still expects at least 15% of day-to-day work decisions to be made autonomously by 2028, up from essentially zero in 2024, so automation is real and advancing. The point is that it is advancing unevenly, and it is leaving large gaps where humans are still the better option. Gartner also flagged a second-order effect worth planning around. It forecasts that heavy reliance on generative AI will erode critical thinking to the point that 50% of global organizations will start requiring AI-free competency evaluations, essentially testing whether a candidate can still reason without the tool. If that becomes common, the workers who kept their own judgment sharp instead of outsourcing every decision will have a measurable edge in hiring. Being AI-native does not mean being AI-dependent, and employers are beginning to screen for the difference. For a job seeker, the takeaway is strategic, not just reassuring. Target the roles and companies where agents are failing or being walked back, support, operations, client insights, coordination-heavy work, and the roles that require judgment. Read past the layoff headline to the [signals that show AI's real effect on jobs](https://www.metaintro.com/blog/read-ai-real-impact-jobs-seven-signals), and remember that the [current wave of entry-level hiring is not entirely AI's doing](https://www.metaintro.com/blog/gen-z-job-market-not-ai-fault). The employers who over-automated are learning the same lesson the [bosses who pushed AI on their teams learned when it backfired](https://www.metaintro.com/blog/bosses-pushed-workers-use-ai-backfired). That lesson is opening doors, and the candidates who understand it will walk through first. The smartest move right now is to stop treating AI as the thing that took your job and start treating it as the thing that keeps proving why your job is hard to automate. Position yourself as the trainable, judgment-carrying human that a failed agent leaves behind, and let the market's own correction do some of the work for you. --- ## Related Articles - [Gartner Says 40 Percent of AI Projects Will Be Scrapped by 2027 and What It Means for Tech Jobs](https://www.metaintro.com/blog/gartner-40-percent-ai-projects-scrapped-2027-tech-jobs) - [Companies Are Rehiring the Workers They Replaced With AI](https://www.metaintro.com/blog/companies-rehiring-workers-replaced-by-ai) - [When Bosses Pushed Workers to Use AI and It Backfired](https://www.metaintro.com/blog/bosses-pushed-workers-use-ai-backfired) - [AI Is Doing the Work and Leaders Want Thinking Skills](https://www.metaintro.com/blog/ai-doing-work-leaders-thinking-skills-2026) - [Entry-Level Grads Are Taking Any Paycheck in a Brutal Market](https://www.metaintro.com/blog/entry-level-grads-any-paycheck-brutal-market-2026) - [Young Adults and Entry-Level Jobs in the 2026 Economy](https://www.metaintro.com/blog/young-adults-entry-level-jobs-2026-economy) - [The Gen Z Job Market Is Not AI's Fault](https://www.metaintro.com/blog/gen-z-job-market-not-ai-fault) - [The AI Skills Employers Actually Want](https://www.metaintro.com/blog/ai-skills-employers-actually-want) - [Human Skills That Get More Valuable as AI Spreads](https://www.metaintro.com/blog/human-skills-that-get-more-valuable-as-ai-spreads) - [AI Skills That Keep You Employable in 2026](https://www.metaintro.com/blog/ai-skills-that-keep-you-employable-2026) - [How to Make Your Role AI-Resilient](https://www.metaintro.com/blog/making-your-role-ai-resilient) - [Grade vs Skill, What Employers Screen For in the AI Era](https://www.metaintro.com/blog/ai-skills-gap-grade-vs-skill-employers-2026) --- ## People Also Asked ### Q: Are AI agents actually replacing entry-level jobs in 2026? A: Less than the headlines suggest. Gartner projects that more than 40% of agentic AI projects will be canceled by the end of 2027, and companies like Klarna have reversed course after AI produced lower quality work, cutting their headcount from about 5,500 to 3,000 before rehiring humans. Automation is real, but it is uneven, and many employers are finding that trainable people beat brittle software on the fuzzy, coordination-heavy tasks that fill most entry-level roles. ### Q: Why do companies say it is faster to train a grad than build an AI agent? A: Because the total cost of a working agent is high. Someone has to scope the task, connect the systems, write guardrails, test edge cases, and keep monitoring it after launch. Inc reported on an AI pioneer who found roughly half of her agents failed and concluded a recent graduate could be made productive faster. A new hire absorbs context, flags problems, and improves without a specialist rebuilding them every week. ### Q: What skills make me the hire a company picks over automation? A: The traits that break AI agents. Stanford's CooperBench research found agents lose nearly half their capability when they have to coordinate, and that social intelligence, not raw skill, is the bottleneck. So clear communication, asking the right clarifying questions, reliably closing out tasks, and using AI tools while catching their mistakes are exactly what employers cannot get from software. Those are learnable habits you can demonstrate in an interview and a first project. Ready to future-proof your career? [Metaintro](https://www.metaintro.com) reads the live job market so you can target the roles where humans still win, spot the employers walking back over-automation, and land the first job that actually trains you. [Create your free Metaintro profile](https://www.metaintro.com/signup) and get matched to entry-level and early-career roles built for people who are ready to learn fast. ### Share this article [https://x.com/intent/tweet?text=40%20Percent%20of%20AI%20Agent%20Projects%20Fail%20and%20Companies%20Are%20Rehiring%20Grads%20in%202026&url=https%3A%2F%2Fwww.metaintro.com%2Fblog%2Fai-agent-projects-fail-companies-rehiring-grads-2026](https://x.com/intent/tweet?text=40%20Percent%20of%20AI%20Agent%20Projects%20Fail%20and%20Companies%20Are%20Rehiring%20Grads%20in%202026&url=https%3A%2F%2Fwww.metaintro.com%2Fblog%2Fai-agent-projects-fail-companies-rehiring-grads-2026)[http://www.facebook.com/sharer.php?u=https%3A%2F%2Fwww.metaintro.com%2Fblog%2Fai-agent-projects-fail-companies-rehiring-grads-2026](http://www.facebook.com/sharer.php?u=https%3A%2F%2Fwww.metaintro.com%2Fblog%2Fai-agent-projects-fail-companies-rehiring-grads-2026)[https://www.linkedin.com/sharing/share-offsite/?url=https%3A%2F%2Fwww.metaintro.com%2Fblog%2Fai-agent-projects-fail-companies-rehiring-grads-2026&title=40%20Percent%20of%20AI%20Agent%20Projects%20Fail%20and%20Companies%20Are%20Rehiring%20Grads%20in%202026](https://www.linkedin.com/sharing/share-offsite/?url=https%3A%2F%2Fwww.metaintro.com%2Fblog%2Fai-agent-projects-fail-companies-rehiring-grads-2026&title=40%20Percent%20of%20AI%20Agent%20Projects%20Fail%20and%20Companies%20Are%20Rehiring%20Grads%20in%202026)[mailto:?subject=40%20Percent%20of%20AI%20Agent%20Projects%20Fail%20and%20Companies%20Are%20Rehiring%20Grads%20in%202026&body=https%3A%2F%2Fwww.metaintro.com%2Fblog%2Fai-agent-projects-fail-companies-rehiring-grads-2026](mailto:?subject=40%20Percent%20of%20AI%20Agent%20Projects%20Fail%20and%20Companies%20Are%20Rehiring%20Grads%20in%202026&body=https%3A%2F%2Fwww.metaintro.com%2Fblog%2Fai-agent-projects-fail-companies-rehiring-grads-2026) For job seekers ## The work you want, without the hunt. Metaintro reads every new posting each day and texts you the few that fit your skills, pay and commute. [Get Started Free](/signup)[Search matching jobs](/jobs/search) - MatchCompare live roles against your current evidence. - PositionTurn proof projects into role-specific applications. - ImproveUse market feedback to keep the skill plan current. [Return to navigation](#main-navigation)