Predicting the Future of Web Development and the Agency-Client Relationship

A Strategic Vision for the Next Generation of Digital Agencies. First published on Google Docs and then on LinkedIn and X.

The experience is fundamentally project-driven and client-driven. The agency manages the client’s entire project through an integrated project management system, providing a centralized environment for coordinating work, managing resources, tracking progress, and maintaining the ongoing client relationship.

Within this environment, clients have access to a conversational interface resembling a chat-based experience or command-line terminal, through which they can interact with their project and its connected systems. This interface is highly empowered by an extensive MCP architecture provisioned, configured, and managed by the agency. The agency determines which capabilities, integrations, and permissions are available to each client, enabling them to interact with and operate their digital infrastructure through a unified experience.

Traditional CMS administration interfaces remain accessible alongside these capabilities, allowing clients and developers to perform specialized tasks, make visual edits, and interact directly with the underlying systems when appropriate.

The agency’s responsibility evolves from performing individual development tasks to configuring, securing, maintaining, and expanding the client’s AI-enabled infrastructure.

The agency becomes the architect, operator, and steward of the client’s digital operating environment.

The client enters their digital environment through a unified application managed by the agency provider. The experience is fundamentally project-driven and permissions-based, with access to tools, integrations, administrative functions, and AI capabilities determined by each user’s role within the client’s organization.

The environment provides several primary interfaces operating at the same level:

Permissions extend beyond conventional CMS roles. Individual users within the same client organization may have substantially different experiences based on their responsibilities and authorization levels.

For example, a marketing manager may have access to Google Analytics, Google Search Console, SEO integrations, and content generation tools, while an ecommerce manager may have additional access to WooCommerce, subscription management, and shipping integrations. Other users may have access to only a limited selection of capabilities.

Some users may have no operational agent permissions whatsoever. Their experience could be limited to a conversational knowledge base tailored to the client’s website, documentation, policies, products, and internal resources. These users could ask questions and retrieve relevant information without possessing the ability to modify content, change configurations, or execute operations against connected systems.

The agency maintains overarching control of the capabilities made available to each client environment, while authorized client administrators manage access within the boundaries established by the agency.

Rather than requiring every user to navigate a collection of independent administration interfaces, the platform provides a unified, role-appropriate experience. Each user can access the management tools, traditional CMS interfaces, and agent-based capabilities necessary for their responsibilities, without being exposed to functionality outside their authorized scope.

The Model Context Protocol provides a standardized framework for connecting AI agents to tools, services, and business data.

Within the proposed SaaS environment, MCP integrations are presented as configurable modules that can be enabled, disabled, and managed at the client level.

The agency assists with their installation, authentication, configuration, permissions, and ongoing maintenance.

For narrative, we are imagining a scenario where WordPress is the underlying CMS. The follow items show opportunities where MCP servers can be modularized, permissioned, and served to client.

The objective is to allow a client to request presentation changes conversationally while retaining the ability to inspect and modify the result through a graphical interface.

Brand instructions, editorial standards, and established personas become persistent organizational resources rather than instructions that must be recreated for each individual task.

These capabilities enable clients to generate, edit, approve, and publish visual assets within the same operating environment.

The modules described above represent desired platform capabilities. Individual services may require custom MCP servers, API integrations, or alternative tool adapters depending on provider support.

The current artificial intelligence landscape is increasingly defined by the capabilities of frontier models developed by leading providers, including Anthropic’s Claude, OpenAI’s GPT models and ChatGPT, Google’s Gemini, xAI’s Grok, Moonshot AI’s Kimi, and DeepSeek.

These models are rapidly advancing in their ability to reason, write and execute code, interact with software tools, and perform increasingly complex operational tasks. Software development and code execution have emerged as particularly important applications, while reliable content creation, editorial management, and broader business workflow automation continue to improve.

For the immediate projected future, this proposal assumes that frontier models will remain central to delivering reliable, capable agentic experiences. Although open-source, open-weight, and self-hosted models may become increasingly competitive alternatives, the platform should initially prioritize rapid access to leading external model providers without becoming dependent on any single vendor.

The agency-managed project environment should therefore provide a model-agnostic execution layer, allowing agencies to quickly enable, configure, replace, and manage different frontier models across their clients’ projects.

Importantly, different models may be better suited to different operational requirements. A model selected for software development and technical troubleshooting may not be the preferred model for editorial writing, research, image generation, or routine administrative tasks.

The platform should support explicit model-to-capability associations, allowing particular models to be assigned to specific tools, integrations, MCP services, and operational workflows.

These associations may be administered at the agency level, delegated to authorized client administrators, or configured automatically according to established rules and performance requirements.

These assignments should remain configurable as model capabilities, pricing, and availability evolve. The system should support agency-defined defaults, client-specific overrides where permitted, model-specific permissions, usage limits, and fallback routing when a preferred model becomes unavailable.

The agency should ultimately control not only which MCP capabilities are available within a client environment, but also which AI models power those capabilities and under what circumstances they may be used.

The long-term value of the agency-managed AI harness lies not in dependence on any particular frontier model, but in providing a persistent, governed, and extensible framework through which agencies can apply evolving AI capabilities to their clients’ projects.

Permission management is one of the most important elements of the proposed operating environment, particularly as agencies begin managing projects that involve multiple client stakeholders, team members, and increasingly capable AI agents.

While controlling what individual agents and AI models can access or modify is an important consideration, the primary function of permission management is to govern the experience and operational authority of the human users accessing the client environment.

Different members of a client’s organization will have different responsibilities, and the platform should provide granular control over the tools, integrations, project management functions, and agentic capabilities available to each individual.

Permissions should also distinguish between staging and production environments. A user may be authorized to experiment, develop, or implement changes in staging without possessing the authority to deploy those changes to production.

The same principles extend to the AI agents operating on behalf of these users.

An agent should not automatically inherit unrestricted access to every integration or MCP capability provisioned within the client environment. Its available operations should reflect both the permissions assigned to the human user and any additional restrictions established for the model, agent, or connected service.

For example, a content strategist may be permitted to use an agent to analyze Google Search Console data and prepare content recommendations, while being prohibited from using that same agent to modify theme files or publish changes directly to production.

These permissions should be configurable across individual users, organizational roles, project management functions, MCP integrations, agent capabilities, and deployment environments.

The agency maintains overarching governance of the client environment, while designated client administrators may be granted the authority to manage their own teams and operational permissions.

The platform should also support authentication, credential management, activity logging, approval workflows, and reversible changes where technically feasible.

Ultimately, permission management must treat human users and AI agents as interconnected but distinct participants in the same operating environment. The system should prioritize clearly defined human responsibilities and organizational authority, while ensuring that AI agents operate strictly within the permissions and capabilities made available to them.

The value exchange between clients and web development agencies is rapidly changing as artificial intelligence assumes an increasing share of the organizational, technical, and executional work traditionally performed by service providers.

Tasks that once required specialized expertise, substantial development time, or direct agency involvement are increasingly accessible through agent-assisted workflows. However, this does not eliminate the need for web development agencies or the value they provide.

Clients will continue to rely on agencies for technical guidance, infrastructure setup, maintenance, specialized development, and operational management. At the same time, they will increasingly want to determine which responsibilities they retain internally and which activities can be performed independently through their own agentic experiences.

The agency-client relationship must therefore evolve toward providing the infrastructure, operational knowledge, and technical capabilities that enable clients to accomplish more independently.

Rather than relying exclusively on individual development engagements, agencies can provision and maintain a persistent operating environment, introducing new capabilities as client requirements evolve.

As the relationship develops, the agency might introduce:

Each additional integration expands what the client and their authorized agents can accomplish independently.

The agency’s role increasingly centers on configuring, maintaining, and expanding these capabilities, while helping clients establish effective workflows and determine where human expertise remains necessary.

These activities create new opportunities for monetization through infrastructure management, integration services, training, specialized toolkits, and ongoing technical support.

The agency’s value is increasingly measured by the capabilities, infrastructure, and operational knowledge it makes available to clients, rather than the hours it spends executing individual tasks.

As