Everybody is building AI agents. But the enterprises actually shipping them to production are learning that an agent on its own is rarely the answer.Woodson Martin, CEO of the low-code AI development platform OutSystems, tells The New Stack that “an agent is rarely the whole story.”“What you need is actually a blend of agents, data, workflow, APIs, and applications — user experiences for humans,” Martin says on this latest episode of The New Stack Agents podcast.Success = agents + humansThe often-cited stat that 95% of agent pilots fail has become a mainstay of discussions about the technology. But Martin thinks it misses the point. In his view, it has become so easy to experiment with agents that a lot of those “failures” were simply low-commitment trials to begin with.“We’re seeing more agents get into production,” Martin says. “Some of them are delivering real ROI, and some are helping a little around the edges. It depends on the business model you’re running and things like what the gross margin structure is you’ve got, how much value you can add in a process, and deliver ROI through automation. Most things differ by industry, but everybody’s making progress.”The more interesting lesson, he says, is in what those successful deployments look like. They’re not standalone agents making autonomous decisions. They’re systems that combine AI prompts and model calls with traditional data structures, APIs, existing business logic, and human oversight.“Most of our customers are at the point where they want humans in the loop, observing, maybe confirming recommendations made by agents, turning those into decisions made by humans, at least at the beginning until they get comfortable,” Martin says.There are some common patterns in how his customers use agents today: document processing, where LLMs excel at extracting structured data from messy inputs; decision support, where agents make recommendations that humans approve; and personalization, where AI tailors outputs based on customer context. He argues that the enterprises that combine those are seeing meaningful productivity gains.Dutch travel company Travel Essence, one of OutSystem’s reference customers, built an agentic system on its platform that compressed a two-hour-per-customer planning process into three minutes.Dutch travel company Travel Essence, one of OutSystem’s reference customers, built an agentic system on its platform that compressed a two-hour-per-customer planning process into three minutes, for example. The result, Martin argues: travel advisors now spend more time selling and less time on back-office admin, which has helped the company accelerate its top-line growth by 20%.The SaaSpocalypse is real, but not in the way you thinkNot everybody is excited about AI agents, and there is some real anxiety about this technology as well. That anxiety centers on a single metric: the per-seat pricing model that fueled two decades of SaaS growth.If agents are doing more of the work, the argument goes, companies need fewer seats. Martin doesn’t subscribe to the most dramatic version of this thesis. He pushes back on what he calls “the human labor apocalypse,” but he does think the per-seat premium is unwinding.“I do think that for a lot of SaaS businesses based purely on the seat, there is real risk,” Martin said. The risk isn’t that all those seats disappear overnight. It’s that investors will start asking whether seat-based businesses deserve the growth premium they’ve commanded.”The other source of anxiety may actually play out in OutSystem’s favor. If enterprises can more easily write custom software tailored exactly to their own needs, they’ll buy fewer SaaS licenses.That trend, he argues, will only accelerate, and Martin believes we’ll see more custom software built in the next five years than in the five before — something OutSystems would love to capitalize on, of course.Low code in the age of vibe codingIt’s no secret that vibe coding has exploded in popularity among individual developers, but enterprise IT leaders are watching nervously. But that’s also an opportunity for a company like OutSystems. Those enterprises want to move faster, but while they need the creative speed of agentic coding, they also need the guardrails of a governed platform: model agnosticism, lifecycle management, evals, role-based data access, and compliance controls.That’s where a company like OutSystems, with its full-stack approach, can obviously help. But Martin is also clear-eyed that not every enterprise will move to its solution.“Most of our customers have Salesforce. […] I think all of our customers also have Microsoft’s Power Apps. I think 100% of our customers also write stuff in code by hand and always have,” he says. “We’re a part of everybody’s infrastructure and system for software development. Of course, if you were writing code by hand yesterday or a year ago, today you’re using Cursor, you’re using Claude, you’re using all these — and some people are just using chat. […] Our job is to be there and be your best partner and make it obvious why you want to do that with OutSystems.”The question, he argues, isn’t whether an enterprise should use agentic coding or OutSystems. “The question is, how can I best do vibe coding and take advantage of all the power of a platform like OutSystems to keep my enterprise safe, and organized, and manageable, monitorable, and compliant.”The post OutSystems CEO on how enterprises can successfully adopt vibe coding appeared first on The New Stack.