Who should own your business operating system: your business, or the company that supplies its AI model? The distinction used to sound academic. It is becoming a commercial decision.
For years, businesses accepted a fragmented collection of applications because switching was expensive and integration was painful. Generative AI promised to break that trade-off. But a new one is forming: the companies selling intelligence increasingly want to own the applications, workflows and interfaces through which that intelligence is used.
My argument is simple: bring your AI; we bring the tools. Your choice of model should be an operating decision, not a permanent bet on which technology company will control your business.
The model provider is becoming a software provider
On 10 November 2026, Anthropic announced Claude Meetings, Whiteboards, Sheets and Data, extending Claude deeper into meetings, collaboration, spreadsheets and reporting. In September, it launched tools for financial advisers. These are not proof that traditional software has been displaced. They show how quickly the boundaries between model, assistant and business application are blurring.
The competition is useful. Incumbents need pressure. Better models and simpler interfaces can eliminate work that software used to make us do. Yet customers should notice the incentive change: the vendor powering the assistant may also compete to become the owner of the workflow, system of record and customer relationship.
If your business builds its processes inside one model company's proprietary application layer, the future cost of switching may be much greater than the difference between this year's model prices.
Reliability is another reason not to concentrate every dependency
On 12–13 November 2026, a Microsoft 365 disruption affected Outlook, Teams and other services. Separately, on 13 November the FAA grounded departing US flights after an air-traffic communications failure. Officials attributed the FAA event to overloaded legacy equipment. The two incidents should not be conflated.
Neither incident proves that one large platform is inherently less reliable than a collection of smaller ones. They are reminders that connected systems create shared points of failure. A sensible operating design needs clear fallback processes, recoverable records and an intelligible answer to one question: what happens when a supplier is unavailable?
Integrated should not mean inseparable
Integration is valuable; dependence is a cost. Business applications should share identity, permissions, data relationships and a dependable action layer. But the model performing the reasoning should be separable from the records and rules it operates on.
That lets an organization route an appropriate task to a closed model from OpenAI or Anthropic, an open-weight model hosted by a trusted provider, or—where its infrastructure and use case support it—a model running privately. It also lets the organization change that choice as quality, latency, price, regulation and data sensitivity evolve. A truly model-agnostic platform must test those transitions rather than merely advertise an array of logos.
Different models will not always produce identical outputs. Portability means preserving the business process, permissions and source-of-truth records while accommodating model differences—not pretending those differences have disappeared.
The business is the asset. The model is a supplier.
A company's competitive advantage often sits in the unglamorous details: its customer relationships, pricing logic, judgement, supplier terms, operational history, specialized processes and accumulated experience. Together, those make up its institutional knowledge.
It is reasonable to use a powerful AI service to analyze that knowledge. It is less reasonable to assume that moving all of it into one vendor's proprietary working environment carries no strategic consequences. Enterprise privacy and contractual protections matter, and providers can make meaningful commitments about data use. The issue is not an accusation of misuse. It is negotiating power, portability, resilience and who ultimately controls the operating layer.
If a model vendor changes prices, restricts a feature, introduces a directly competing product or becomes unavailable, your organization should still be able to operate.
Systems of record will compete on more than features
I expect the functional gap between many business applications to narrow. AI-assisted development can make forms, CRMs, content tools and workflow products faster to build. That does not make every system interchangeable: data quality, security, regulation, reliability, distribution and operational depth remain hard.
But the lasting advantage will be less about whether one app has a particular button and more about whether the whole business can work coherently: consistent identity, permissions, linked records, reliable actions, observable results and a practical exit route.
That is the bet behind TopoloOne: a growing, integrated collection of focused business applications sharing an operating foundation, designed to let people and AI agents work through the same governed tools. The published Topolo agent and automation contracts describe how applications and actions can be discovered, permissioned and verified. Broad provider interchangeability—including every privately hosted model—is a direction to validate workload by workload, not a blanket compatibility claim.
Five tests to run before committing your business to an AI ecosystem
- Model exit: Can a critical workflow be run with a different model without rewriting its business logic or moving its source-of-truth data?
- Data exit: Can you export records, identifiers, relationships, attachments and operational history in usable formats?
- Permission boundary: Are agent actions scoped by identity, organization, resource and approval—regardless of the chosen model?
- Operational fallback: What still works if the model endpoint, the application suite or the network is unavailable?
- Commercial independence: What happens to your economics and roadmap when the supplier starts selling software that competes with yours?
Demand a real end-to-end test for each question. A diagram and a sales promise are not enough. If the answer is unknown, record it as unknown—not as compliance, portability or resilience.
Optionality is the strategy
The best AI model next quarter may not be the model that is best for your business today. The price-performance leader may change; an open model may be preferable for a sensitive local task; a closed model may remain unmatched for another. We should welcome the competition without requiring every company to gamble its operations on the winner.
Bring your AI. We'll bring the tools. Your business remains yours.
Ashley Lane, Founder, Topolo