Digital platforms give smaller businesses access to customers, infrastructure and capabilities that once required considerable scale. Large language models now extend that opportunity by helping people navigate services and coordinate work across systems. The strategic challenge is to capture these benefits while retaining control over customer relationships, business decisions and data.
The early promise of the platform economy was compelling: connect local businesses and individuals, reduce transaction costs and distribute economic opportunity more widely. In 2026, that promise remains relevant. It also requires a more demanding assessment of who sets the rules and captures the value.
Platforms rarely eliminate intermediation. They reorganise it. A marketplace can replace several distribution relationships with one powerful intermediary. As AI assistants begin to influence which products customers consider and which services they use, another intermediary enters the relationship. For management teams, this creates opportunities for growth alongside dependencies that need active governance.
The benefits of access and shared infrastructure
A specialist manufacturer can reach international buyers without establishing a sales office in every market. An independent professional can combine payment processing, scheduling and digital distribution without building each capability internally. Platforms can also help businesses use spare capacity, match supply with demand and test new offerings at relatively low initial cost.
Their economic value comes partly from sharing infrastructure. Participants gain access to transaction systems, customer discovery and mechanisms that support trust, such as reviews and identity checks. Where these mechanisms work well, they reduce the effort required to find a supplier and assess whether a transaction is credible.
For customers, the benefits include broader choice and easier comparison. For smaller firms, they include access to demand and capabilities that would otherwise be expensive to develop. The commercial test is whether the additional contribution margin remains attractive after fees, advertising, fulfilment, returns and internal administration.
The dependencies behind platform growth
Access to customers can become dependence on a platform’s rules. Changes in ranking, fees or account eligibility may affect revenue quickly. A business can achieve substantial sales while having limited ability to contact customers directly or move its reputation to another service.
Network effects can reinforce this imbalance. More buyers attract more sellers; more sellers increase the platform’s usefulness to buyers. Once participation becomes commercially necessary, individual businesses may have little bargaining power. Selling through several channels can reduce exposure, although it also creates additional operating costs.
The distribution of risk deserves equal attention. Depending on the model, participants may retain inventory, delivery, asset or income risk while the platform controls discovery and transaction conditions. Reviews can support trust, but misleading feedback and opaque dispute processes can undermine it. Wider participation therefore does not automatically produce a fairer distribution of economic power.
How LLMs connect fragmented services
Large language models can interpret requests expressed in ordinary language and help translate them into structured tasks. When combined with retrieval, software tools and appropriate permissions, they can support work across several applications. The model provides interpretation and planning; connected systems supply current records and execute authorised actions.
Consider an illustrative procurement workflow. A manager asks for suppliers that meet a technical specification and delivery deadline. An AI assistant retrieves approved supplier records, compares quotations and prepares a purchase request. A person reviews the proposed order before commitment. The potential benefit comes from coordinating information that previously required repeated searches and manual copying.
This connectivity still requires integration. The Model Context Protocol, or MCP, provides a common approach for connecting AI applications to external data and tools. It can reduce the need for separate custom connections, but it does not by itself make incompatible data consistent or grant permission to use it. [1]
Commerce is developing similar infrastructure. In January 2026, Google introduced the Universal Commerce Protocol with partners including Shopify, Etsy and Walmart. It is designed to support interactions between AI services, merchants and payment providers, including discovery and checkout. It supports integration through APIs, MCP and Agent2Agent. [2] These initiatives demonstrate a concrete direction of development; adoption and dependable execution still vary across services.
Lower barriers to knowledge and coordination
For smaller organisations, LLMs can make existing capabilities easier to use. Employees can ask questions across authorised documents, draft supplier communications and work with multilingual information. A company may be able to serve more markets or manage more complex operations without increasing administrative effort at the same rate.
The opportunity also extends to collaboration. Businesses could coordinate capacity, exchange specifications and identify complementary services through connected systems. Clear data structures and reliable records remain essential: language models can help interpret different descriptions, but a plausible interpretation is not proof that two products or contractual terms are equivalent.
These benefits depend on process design. A faster draft or search result creates value when it improves a business outcome. Management should assess completed work, correction effort and service quality rather than treating the number of AI interactions as a productivity measure.
AI assistants can become powerful gatekeepers
A customer who delegates comparison to an assistant may see a short recommendation instead of visiting several marketplaces. This creates a strategic possibility: control over the assistant could influence which suppliers gain visibility. Recommendation criteria, commercial partnerships and sponsored placement would then matter to market access.
For businesses, this suggests preparing reliable product information and maintaining direct customer relationships alongside platform distribution. It also raises a governance question for buyers: does the assistant evaluate the options against the customer’s criteria, and can the basis for its recommendation be inspected?
Concentration at the infrastructure level adds another dependency. On 25 June 2026, the European Commission announced its preliminary view that AWS and Microsoft Azure should be designated as gatekeepers under the Digital Markets Act. It highlighted switching costs, lock-in and the influence of AI offerings on cloud procurement. [3] That announcement was a preliminary position. Its strategic relevance is the connection between AI choices and broader infrastructure dependence.
Connectivity can turn errors into business actions
An incorrect answer becomes more consequential when an agent can place orders, modify records or transmit confidential information. A model might misunderstand a specification, rely on an outdated price or select an inappropriate tool. Connecting additional systems increases the potential reach of such mistakes.
There is also a security risk from prompt injection: malicious instructions embedded in a webpage, document or tool response may attempt to redirect an agent. Anthropic’s published research describes this threat and acknowledges that it remains unresolved despite improvements in defensive measures. [4]
Businesses should therefore enforce permissions in the systems that execute actions. A model’s instruction to respect a limit should be supported by an actual technical restriction. Spending thresholds, restricted data access, approval requirements and transaction logs help make responsibilities explicit. Duplicate-action protection and recovery procedures matter when workflows involve payments or irreversible changes.
LLMs can also lower the cost of producing deceptive messages or misleading content. Platforms and their participants need to preserve the provenance of important information and provide practical routes for human review when authenticity or accuracy is disputed.
Management priorities for the next phase
A useful starting point is one workflow with a measurable business problem, reliable source data and a clearly accountable owner. Establish the current completion time, error rate and cost. Introduce AI within defined permissions, then evaluate successful outcomes and the effort needed to correct failures.
Commercial resilience should be part of the design. Review whether customer records, transaction histories and workflow configurations can be exported. Test the practical effort of changing a provider. Include model usage, integration, supervision and incident handling when assessing total cost.
The central management task is to decide which activities can be delegated, under what conditions and with what evidence of successful completion. Platforms and LLMs can widen participation in economic activity and improve coordination across organisations. Achieving that value requires businesses to understand the dependencies they accept and retain the ability to question, override and change the systems on which they rely.
STRATECTA supports management teams in assessing AI opportunities, designing governance and translating connected AI capabilities into practical business processes.
Sources
[1] Model Context Protocol. What is the Model Context Protocol. Official documentation, accessed 3 November 2026. https://modelcontextprotocol.io/introduction
[2] Google Developers Blog. Under the Hood Universal Commerce Protocol. 11 January 2026. https://developers.googleblog.com/under-the-hood-universal-commerce-protocol-ucp/
[3] European Commission. Preliminary position on designation of Amazon and Microsoft cloud services under the DMA. 25 June 2026. https://digital-markets-act.ec.europa.eu/commission-reaches-preliminary-position-amazons-and-microsofts-market-leading-cloud-services-should-2026-06-25_en
[4] Anthropic. Mitigating the risk of prompt injections in browser use. 24 November 2025. https://www.anthropic.com/research/prompt-injection-defenses
Related Links:
