# Don’t Layer AI Onto Legacy Systems > Source: > Published: 2026-08-31 20:35:01+00:00 **To truly improve efficiency and eliminate bottlenecks with AI agents, organizations must first modernize the environments on which those agents depend.** When it comes to AI, many organizations are essentially running in place. Enormous effort is going into adding agents to legacy systems, rather than modernizing how work is actually done. This approach often creates productivity improvements, but these gains are typically isolated and sometimes short-lived. That’s because the organization has done nothing to remove the outdated applications, fragmented data, and other factors that continue to slow the overall business. We’re all familiar with the hurdles that organizations have encountered when trying to create real, lasting business value with AI tools. Last summer, MIT’s NANDA initiative reported that [95% of organizations](https://fortune.com/2025/08/18/mit-report-95-percent-generative-ai-pilots-at-companies-failing-cfo/) in its research pool were getting essentially zero value from their AI pilots. And this year, [Deloitte found](https://www.deloitte.com/us/en/what-we-do/capabilities/applied-artificial-intelligence/content/state-of-ai-in-the-enterprise.html) that only around one-third (34%) of organizations are using AI to “deeply transform” their operations with new products, services, or business models. There is certainly no shortage of investment or enthusiasm for AI, but many leaders are left wondering why they have not found a “silver bullet” application that dramatically improves productivity or creates new revenue. Most commonly, people tend to blame factors such as hardware shortages, poor data governance, unclear ownership, skills gaps, and disconnected experimentation across business units. One of the most overlooked issues, however, may also be one of the most obvious. Too often, organizations add AI agents on top of the same applications, data stores, and infrastructure that were already creating inefficiencies before they began experimenting with AI. In other words, many of us are expecting AI agents to transform our business, even though we’re not giving those agents modern systems that are conducive to success. To create value at scale with AI, we often need to look beyond the agents themselves and first examine our own IT environments. **Inherited Bottlenecks Limit AI Agents’ Effectiveness** AI agents don’t operate in a vacuum. To perform useful tasks at scale, they need access to enterprise data, applications, and workflows. But if these systems are fragmented or difficult to integrate, agents will inherit these limitations. Consider a customer service workflow, for example. An AI agent might be able to interpret the intent behind a customer’s question instantly. However, a fully successful agentic workflow (including providing a useful answer or taking action on behalf of the customer) will likely require additional information from several business tools that may be largely disconnected. Existing workflows may depend on human employees to transfer data between applications, obtain approvals, or even reconcile conflicting records. In this type of fragmented environment, adding an intelligent agent will only make one portion of the process more efficient, while the remaining bottlenecks delay or derail the overall workflow. This same dynamic applies to virtually any use case in which data stores are fragmented. AI tools, of course, can only work with the information they can access, and legacy warehouses and disconnected databases can prevent agents from obtaining the context they need to take appropriate action. At Mactores, we talk about agentic AI in terms of [three pillars](https://mactores.com/what-we-ship): - Data platform modernization (data infrastructure is consolidated on Amazon Web Services [AWS]) - Application and database modernization (the retirement of legacy database and ERP debt, as well as reallocation of license and maintenance spend) - AI agents for apps (agents that go beyond proof-of-concept stage to perform auditable work in real operations, with human time reallocated to edge cases) This approach recognizes that in practice, IT modernization and agentic AI cannot be treated as separate initiatives. Rather, modernization has become a foundational component of the AI strategy itself. Organizations that bolt agents onto existing processes and infrastructure may notch short-term wins, but they will rarely achieve the lasting business transformation that agentic AI promises. **Modernizing for AI, with AI** Of course, modernizing infrastructure for agentic AI is easier said than done. Organizations have spent years implementing applications, standing up databases, and establishing integrations and business rules. Simply mapping the existing environment, let alone modernizing it, can be a massive undertaking. Fortunately, AI tools themselves are invaluable in assessing and preparing environments for agentic AI. Discovery is the first step in any modernization effort. Leaders need to understand which applications and databases support various workflows, where the critical data lives, how systems exchange that information, and which dependencies could create risk during a modernization-related migration or cutover. When those relationships are invisible, it is impossible to determine what needs to change within an environment to support agentic AI at scale. [Boston Consulting Group](https://www.bcg.com/publications/2026/agentic-ai-power-core-insurance-ai-modernization) (BCG) notes that AI agents can analyze legacy applications (including those based on COBOL) to automatically extract business rules, generate documentation, and create process maps. “This capability accelerates one of the toughest, most time-consuming steps in modernization: understanding how legacy systems actually work,” BCG writes. “AI agents reduce the reliance on subject matter experts, whose time is at a premium.” Modernization teams can also use AI tools to assist with validation, test generation, and repetitive modernization work. Mactores uses the Aedeon AI agent platform for enterprise modernization, analyzing, changing, testing, and migrating complex enterprise systems, especially data platforms, legacy applications, databases, and AI workloads on AWS. This AI-led approach dramatically improves the economics of modernization. Historically, modernization efforts have been built on droves of junior engineers working under a few senior consultants incentivized to rack up billable hours. With Aedeon, Mactores breaks that model. A task that may have previously taken 50 analyst hours can now be done in an afternoon. **Moving from Assessment to Production ** To progress from initial assessment to a sequenced modernization plan, to production-ready agentic AI, organizations must follow a structured plan. At Mactores, we follow a five-step process to help organizations move from assessment to production: Assess, Design, Build, Test, and Deploy. During assessment, teams map the existing environment and its dependencies, using the Aedeon platform. The design phase establishes a target for infrastructure modernization, with leaders carefully considering technical and cost tradeoffs. Then, AI can accelerate portions of the build and testing phases, including code translation, pipeline scaffolding, regression testing, and validation. Finally, experienced engineers retain ownership of cutover planning and production acceptance. In a typical modernization effort, around 60 to 70% of the engagement work is absorbed by the Aedeon platform, with engineering teams shifting much of their focus to architecture decisions and sign-offs that require human expertise and accountability. The effort is led by forward-deployed engineers who have all shipped production agentic AI on AWS, and the median time to production is 12 weeks. We used this process to help the semiconductor manufacturer [Synaptics](https://mactores.com/stories/synaptics-oracle) migrate its legacy Oracle environment to Amazon Relational Database Service. It was a migration that the company’s engineers had desired for some time, but (somewhat ironically) the company could never spare the engineering cycles needed to make the move happen. The company relied on Aedeon for dependency mapping, code conversion automation, and schema analysis, while forward-deployed engineers were responsible for architecture decisions, risk-weighted cutover sequencing, and the ownership of production cutover. We also worked with a [branded-payments leader](https://mactores.com/stories/branded-payments) that had stalled out on two previous attempts to modernize its infrastructure. Aedeon ran the code analysis, dependency mapping, test generation, and validation that had consumed the budgets of previous modernization attempts. Then, our forward-deployed engineers used this budget on cutover risk management. Company leaders estimated the project’s ROI at between $2.5 million and $4 million per year. **Modernization That Ships** Agentic AI holds enormous potential to transform enterprise operations. Organizations are already using AI agents to improve customer service, simplify IT service management, and automate revenue-related workflows. And we’re only just beginning to see what sorts of outcomes this emerging technology will make possible in the very near future. But organizations will not achieve these benefits by deploying sophisticated AI tools atop aging technology. Instead, leaders need to take the time to understand which legacy applications, data stores, and dependencies are hindering their highest-value AI use cases and then establish a path to rapid modernization. On Tuesday, October 13 at 1 p.m. Eastern, I’ll be discussing these critically important issues with The Futurum Group’s Tech Field Day lead, Tom Hollingsworth. During the webinar, we will discuss: - Why adding agents on top of legacy systems often only preserves the very bottlenecks organizations are trying to eliminate. - How to identify the applications, data, workflows, and dependencies that must be modernized before agentic AI can scale. - How AI agents can accelerate modernization, and where experienced engineers must retain ownership. - How to progress from assessment to execution. Register [here](https://webinars.techstronglearning.com/modernization-that-ships-with-ai) to join us.