The legacy reckoning

Legacy systems were always a drag on efficiency but now the liability is existential, as organizations discover that the ceiling on what they can do with AI is set by the floor of what their infrastructure can support.

Technical debt used to be a slow tax on efficiency, absorbed quietly into IT budgets and rarely surfaced at the board level. AI has changed what is at stake. According to the AlixPartners Disruption Index, a third or more of developers' time is now consumed by workarounds caused by technical debt rather than value-creating work. Among companies running on outdated technology, two-thirds view new technologies as a threat to revenue, while three-quarters of those with well-maintained infrastructure see the same technologies as opportunities. The difference is not ambition. It is whether the underlying estate can support what the business is trying to build on top of it.

The organizations closing that gap are not treating modernization as an IT cleanup exercise. They are quantifying technical debt in terms a CFO already tracks, which brings visibility to the true cost. Leading teams then decide, system by system, whether to remediate, wrap, or rebuild based on total cost over a realistic horizon, rather than whichever number looks smallest in year one.

Talking to the CFO about tech debt 

Technical debt has a tendency to become background noise: a known cost organizations learn to absorb rather than address. Debt doesn't just inflate run costs by 10-20%. It consumes 30-40% of the total cost of every change program, a far larger and far less visible number. AlixPartners recommends a board-level metric, the technical debt index, that measures rework cost against new capability build and puts the problem in the same reporting environment as revenue and EBITDA. Until that translation happens, the IT modernization argument will keep losing to competing priorities in the CFO's inbox.

ERP modernization in an AI world

What looks like an IT project is often a business transformation in disguise, and ERP migrations are where that gap between expectation and reality is most costly. The crucial data questions — how it's structured, who owns it, how it moves between functions — tend to surface every long-standing disagreement in the organization that process design and reporting accountability have papered over for years. Integration gets treated as a technical afterthought rather than a design principle, customization complexity is built on unstable foundations, and leadership misalignment gets reflected directly in the architecture. The stakes are higher than before. An ERP that can't flow clean, consistent data end to end doesn't just create operational friction; it sets the ceiling on what AI can do. The choice between rebuilding and patching should be made on total cost over a realistic horizon, not whichever number looks smaller in year one.

The next-generation stack

The IT architecture most large enterprises run today didn't arrive by design. It accumulated, layer by layer, until the original logic was buried under decades of customization nobody wanted to disturb. The next generation risks the same fate, only the accumulation will happen faster. A thin, standardized ERP base topped by a proliferating agent layer built by multiple parties sounds like progress, and in many respects it is. But the governance model for that agent layer doesn't yet exist at scale. Unlike legacy ERP customizations, which at least sat within a managed portfolio with known owners, the agent layer will be diffuse, multi-authored, and difficult to audit. The organizations that build governance in from the start are avoiding a debt that will otherwise compound in exactly the way legacy customization did before it. 

SaaS and ERP in the AI era

The case for replacing SaaS platforms with AI agents is logically coherent yet practically premature for most organizations. Switching costs, embedded institutional knowledge, and vendor R&D budgets that dwarf what most customers could build independently are structural advantages that don't dissolve quickly. The deeper competitive moat, however, has always been feature differentiation through the capabilities that required years of proprietary development to build and that locked customers into a specific platform. AI is commoditizing that. The same capabilities can now be replicated by any vendor using the same underlying models, and vendors are in the middle of their own AI transitions rather than ahead of them. The build-versus-buy equation is shifting as custom software costs fall. The death of SaaS is overstated. The erosion of its moat is not.