Agentic AI: Trading one lock-in for another

The two sides of lock-in

Organisations experience lock-in in two forms, often simultaneously. Commercial lock-in is the more visible kind: pricing subject to unilateral revision, with renewal cycles the point at which that power imbalance is felt most acutely – licensing terms that change after acquisition and support bundles that expand in scope and cost with each contract cycle. Technical lock-in is more insidious. It accumulates through proprietary data formats, platform-specific integrations, and architectural dependencies until the cost of leaving exceeds anything a contract negotiation can fix. Most enterprise software providers engineer both, deliberately.

The organisations currently weighing agentic AI adoption are, in many cases, being invited to do so by the same IT suppliers who engineered their existing dependencies. SAP's Business AI and Oracle's Fusion Agentic Applications both position agentic capability as the logical next layer of the platforms their customers already run. The data is already there, the workflows are ready to connect, and the agents are native to the system. The word those IT suppliers are reaching for, again, is transformation. It was the pitch in the 1990s. The dependency model it creates – proprietary data, platform-specific integrations, the cost of leaving that compounds until it exceeds the cost of staying – is the same one their customers have spent years navigating.

There is a second risk here. Agentic AI platforms are immature in ways that ERP platforms, for all their faults, are not. Building an agent is straightforward. Running one in production – with real data, real consequences, and real integration requirements – is where the complexity accumulates. Custom model fine-tuning, platform-specific orchestration layers, proprietary data pipelines: each one is a dependency that compounds over time and becomes progressively harder to unwind. The compounding problem rarely features in IT supplier conversations. AI amplifies whatever environment it enters – the strengths of a well-governed system, and equally, the dysfunctions of a fragmented one. That fragmentation operates at two levels: unresolved technical debt in the infrastructure, and accumulated data quality problems that have been deferred rather than addressed. Introduced into that environment, AI doesn’t resolve either. It accelerates both. Organisations that commit to a platform without first stabilising their underlying environment – both architecturally and at the data level – may find they have traded one set of constraints for something considerably more difficult to manage.

Urgency is not the same as readiness

This is not an argument against agentic AI adoption. The capabilities are real and, in the right context, genuinely transformative. The question is whether organisations are making the choice on their own terms or on the IT provider’s– and whether anyone in the room has explicitly named what is being traded.

The concern many CIOs are acting on is legitimate. Competitors are investing. Automation capabilities are being built. Customer expectations are evolving, and the gap between organisations that have embedded intelligent workflows and those that haven't will, over time, become visible in operational performance and market position. Nobody wants to be the organisation that watched that gap open and did nothing.

But that competitive reality is precisely what makes the IT supplier urgency narrative so effective – and so worth examining. There is a difference between moving because the business case is clear and moving because a technology provider has made delay feel dangerous. Adopting a platform whose commercial and technical dependencies haven't been properly stress-tested doesn't close the competitive gap. It trades one vulnerability for another, on a timeline someone else has set.

There is a third option

The choice technology providers present is not the only one available. Stay and pay more or transform on our terms: it is a commercial construction, and it only holds for as long as organisations accept that their existing environment must remain tethered to the original software provider’s roadmap. 

It doesn't have to. When ERP environments are supported independently of the original software providers, the deadline disappears. There is no end-of-support date creating artificial urgency, no forced transformation to justify a platform decision that hasn't been properly examined. The pressure – and it is pressure, not necessity – dissipates. What's left is a genuine decision, made on the organisation's timeline, against its own criteria. That is a different conversation entirely from the one most IT suppliers want to have.

Third-party software support means security vulnerabilities continue to be addressed; compliance coverage remains in place, and incident response continues – without any requirement to transform. Customisations built over years of operational use – the integrations, the configurations, the workflows that make a system fit the business it runs – are maintained and supported rather than treated as obstacles to an IT provider’s preferred migration path. Pricing and terms cannot be unilaterally revised by the IT support provider at renewal. The commercial exposure that defines IT supplier lock-in – the repricing, the bundling, the end-of-life pressure – ceases to be the condition under which every subsequent decision gets made.

For organisations whose core systems are stable, well understood, and performing, this is not a defensive posture. It is the conditions under which a genuinely strategic AI decision becomes possible. The ERP environment that software providers characterise as legacy is, in many cases, a stable, deeply configured platform that already holds the data, the business logic, and the process context that agentic AI needs to deliver real value. The transformation isn't the prerequisite. It was always the upsell.

Agentic AI can be pursued on top of existing ERP implementations, through an organisation's own choice of supplier, scoped to the workflows where the business case is clearest. That is a faster path to value than a platform-wide migration, at a fraction of the cost, without the implementation risk that has historically turned IT supplier transformation promises into multi-year ordeals. It is also, notably, the path that doesn't require handing the negotiating position back to the same IT providers who created the dependency in the first place.

Getting that foundation right matters more than moving quickly. The value of agentic AI, when it is real, does not expire. The only thing that actually expires is the ability to negotiate on your own terms – and IT suppliers, for obvious reasons, would prefer you didn't notice that.

Martin Biggs is vice president of product management at  Spinnaker Support