Why now: four findings that make the case for your own AI operating layer
Almost every company is trying out AI by now. The question is no longer whether, but how individual experiments become an operation. Four findings from McKinsey's State of Organizations 2026 show why now is the moment, and what follows for the architecture.
Complexity is the real problem
Two thirds of leaders say this about their own company. AI can reduce that complexity, but it can just as easily add to it: every department with its own AI subscription, every further SaaS tool for one sub-task adds an interface, a contract and a silo of data.
The answer is consolidation: one governed platform such as plugnpl.ai instead of a dozen AI tools. And for business applications, a few owned ones instead of many rented, built with the Launchpad. What that looks like in practice is shown by our own CRM, which replaced several tools.
Speed pays off
Fast organizations perform clearly better financially. A custom AI build that needs 12 to 24 months until first use does not fit that. Neither does a finished tool that cannot adapt to your own processes.
What matters is keeping two moments apart: when does the system run technically, and when does it save measurable time or money? The first should be done in days. The second depends on the first use case, for which the baseline, target and measurement period are agreed beforehand.
The model landscape keeps moving
Companies now use twice as many AI capabilities as a few years ago, and the models behind them change at short intervals. Anyone who ties their applications firmly to one model or vendor has to rebuild with every switch. According to Bitkom, 53 percent of German companies already feel locked into a vendor.
The architecture should therefore keep the model replaceable: a model layer through which OpenAI, Anthropic, Google or open-source models are connected without touching the applications. More under Every model. No lock-in.
No adoption, no transformation
Technology alone changes nothing. Almost three quarters of transformations get stuck when leadership does not carry them or the workforce does not take them up. For AI this means: it has to be accessible to everyone, not just a few licence holders, and governed so that IT and data protection can say yes.
Governance is therefore not a brake but a precondition for adoption: permissions through your existing Entra ID, an audit trail of every AI interaction and data that stays in your own tenant. How this is built in detail is described under Security.
What follows
- Consolidate instead of stacking: one governed AI platform and a few owned applications instead of ever more single-purpose tools.
- Deploy in days, measure on the use case: plan technical deployment and measurable benefit separately.
- Stay model-agnostic: keep the model layer replaceable so the next model switch is not a project.
- Plan adoption instead of hoping for it: make AI accessible to everyone and govern it so that IT and data protection agree.
This is exactly what plugnpl.ai is built for: a platform that runs in your own Azure tenant, with full code access and free choice of models.
Where does your company stand?
In one conversation we clarify your first use case, your requirements and what your own AI operating layer would cost you.
Book a meeting- McKinsey & Company, The State of Organizations 2026 (survey of 2,500+ business leaders at organizations with 1,000+ employees)
- Bitkom, 2025: share of companies that see themselves in vendor lock-in