Nathan Labenz, host of The Cognitive Revolution, asked Woodson Martin whether AI was finally shortening the long development backlogs that corporate IT departments live with. Martin, chief executive of the software company OutSystems, did not answer with a number. Instead, he described a backlog whose contents are changing.
"Things are changing," Martin said. Companies are now starting projects they had long been afraid to take on, he said, while a wave of homemade AI tools is creating a new kind of cleanup work. And in his view, even though AI has made building software better, deciding what to build is still the hardest question, as it was before.
OutSystems sells a platform that large companies use to build and run their own business applications. It named Martin CEO in May 2025 after his 18 years at Salesforce. In his introduction to the episode, published October 7, 2026, Labenz said customers including Petrobras, Vodafone and Toyota are using the platform to speed up their product plans. Some, he said, are starting to clear backlogs that had only ever grown. According to Labenz, Martin says most of what those customers ship today is still traditional, predictable software that does not need the most advanced AI models.
The projects nobody wanted to start
The change Martin called "probably the most exciting" involves old legacy systems. His examples were COBOL, a business programming language that dates to the late 1950s; IBM's AS/400 business computers; and Lotus Notes, an aging collaboration platform. These projects are "finally getting their day," he said, because AI can now speed up every phase of a complicated job. First comes understanding what the old system actually does and translating that into modern requirements. Then laborious manual steps become work done by AI agents. Finally, the replacement is built, tested and rolled out.
His example came from insurance. Companies are taking on the modernization of "archaic 60-year-old case management systems," he said, and telling OutSystems: "Instead of planning it as a six-year thing, we're now going to do it in six months." Martin described these as plans, not finished projects, and he did not name the insurers.
OutSystems has built products around this pitch. On June 2, 2026, it announced Legacy Modernization Services that use Amazon's AWS Transform to migrate systems including COBOL and Lotus Notes. The services were released as previews for selected customers, and the company said it planned to expand the integration in the third quarter. OutSystems' case study on Axos Bank, a customer Martin brought up later in the conversation, reports two developers modernizing 12 business processes, averaging two months per project. Those figures come from the company.
A new backlog of homemade dashboards
The second change runs the other way. In most businesses, Martin said, a dashboard to track something new used to be a routine request: you asked someone, it went on their backlog and eventually somebody built it. Today, he said, in most organizations you can just tell Claude to "take this spreadsheet, make me a dashboard and boom."
That self-service has a cost. Companies are "losing standardization" and "losing integration," Martin said, and will need to clean up. But he saw an upside: the people who will use the tools have already worked out what the tools should do. "What do people really need? Well, they built it themselves. They kept talking to Claude until they got it."
What remains, in Martin's account, is connecting the behind-the-scenes plumbing of those tools so they follow the company's common data definitions. These are shared agreements on what something like a "customer" or an "order" means across every system. The tools also need to follow the company's standard role-based access control, the rules that decide who may see or change what. That, he argued, is where a platform like OutSystems makes a difference, handling "a very different kind of backlog than what it looked like even a few months ago."
Why productivity is slow to show up in profits
Labenz then raised a familiar puzzle. People see AI everywhere, as they once said about computers, but the economic growth statistics barely move. Were Martin's customers nearing a point where faster growth would become measurable?
Martin said it is "very much a customer by customer story." He pointed to Kevin Hearn, who he said runs software engineering at an all-digital bank based in San Diego. OutSystems' case study identifies the bank as Axos Bank and Hearn as its senior vice president and head of consumer bank development. Martin said the bank has "been at this for a long time" and is seeing huge productivity gains. Its goal, he said, is to deliver new features fast enough to compete digitally for deposits, customers and loans. "I'll let them speak to their own financial results from that," he added.
Martin did not cite figures, but the company's own case study does. It reports a 66% improvement in developer productivity using one-third of the resources on OutSystems projects, plus $150,000 a year saved by retiring old workflow platforms. A separate program of four AI agents reportedly cut the platform's technical debt by 10% in its first 60 days. It also compressed a three-year modernization plan into one year. Technical debt is the accumulated shortcuts and outdated code that make software harder to change.
Martin's second example was YESCO, which he said builds and operates signs around the world. He introduced it by mentioning the famous Welcome to Las Vegas sign. The company has "dramatically accelerated" its productivity, he said, so it can go after more new business than its existing staff could handle before. Again, he left the financial results to the company. YESCO's case study reports a 50% cut in development time since it adopted OutSystems. It also describes a shift from six-month delivery to two-week release cycles. One AI estimating application turns uploaded photos of buildings and logos into visualizations, prices and vector design files. These too are company-reported figures.
To Martin, the lag is normal. "It takes a while in an organization for progress in productivity to translate into P&L results," he said, referring to the profit-and-loss statement. "And that's true in my business." Not every AI investment will pay off, he said, because much of the current activity is experimentation with new things. Still, he said, "everybody's pretty convinced that the potential is high."
Who keeps the gains
Labenz pushed further. When AI makes a business more productive, does competition pass the savings on to customers, or does the industry's structure change? Martin said the answer will be "very different depending on industries." Discrete manufacturing, which makes individual products such as cars or appliances, was his extreme example. So much of its cost is tied up in physical materials that its margins are probably "a lot more closely related to tariff policy than it is to AI," he said. Consulting sits at the other end. Because that business deals in information, AI can take on more of the work. Martin expects that to affect both how much demand there is for consulting and the margins it earns.
He also expects big differences within industries. Some organizations are "leaning in, trying things," while others have chosen to wait and watch. That, he said, will help decide who gains on margin, who gains market share "and therefore, you know, who ultimately wins."
The hardest question: what to build
Labenz had tried Mentor, OutSystems' assistant for building applications. He asked a strategic question: if each new generation of models makes yesterday's hard-won work easy, how should a company decide what to tackle early and when patience pays off? Martin did not directly address the waiting question. He started instead with Mentor's history. OutSystems began the project in 2018, before he joined, when the available technology was machine learning, not generative AI. Today, he said, Mentor is "a set of MCP services." MCP, the Model Context Protocol, is a standard way for AI tools to connect to other software. Martin said those services can be used anywhere to build software that reliably meets requirements such as security and compliance. In Mentor's first quarter of general availability, OutSystems reported more than 2,500 applications generated with it, taking an average of three minutes each.
Better AI has not changed the central problem, Martin said: "What to do is still the hardest question." He said experimentation should focus on "where can I get the biggest bang for my buck" and on which business processes make sense to make agentic, meaning handed to AI agents that carry out multistep tasks.
OutSystems has "learned the hard way on this" with its own customers, some of whom run portfolios of thousands of applications on its platform, he said. Its answer is an agent that tells customers what to build. Martin said it had just launched and is currently in the hands of OutSystems' customer success teams. It uses telemetry, the usage and performance data, from the applications a customer has built. It also examines how those applications are structured and how data flows through them. From that, it suggests where an agent could turn a manual job into an automated one and forecasts the return on investment, such as time saved. Later in the conversation, Martin called it the agent foundry and described it as a tool that "inspects your architecture" to find opportunities for agents. OutSystems has separately described an Enterprise Context Graph, a structured map of a customer's applications, workflows, integrations and dependencies that agents can query before proposing changes.
The foundry goes one step further. "Press this button to build it," Martin said, and the AI builds a first version. The customer then runs that version through its own development, testing and validation process before trying it out. For most organizations attempting to turn their business over to agents, he said, "solving the problem of what to do is like a huge part of the challenge."