Software Engineer

If you missed the Microsoft Build and Google IO conferences, this one's for you. Don't get left behind with the next major step for digital products. Two new acronyms that you likely haven't heard before but will no doubt start to over the coming years: MCP (Model Context Protocol) and NLWeb (Natural Language Web). These will be the open layers that will set up the paradigm for the new agentic web; MCP being the protocol (communication) layer akin to HTTP on the open web, and NLWeb being the markup (interface) layer akin to HTML. Agentic represents the next evolutionary layer of the internet just like web, search, social, and cloud did before it.

However, unlike these previous layers that enhanced human capabilities, the agentic web will fundamentally transform how software interacts with users and other systems autonomously. Instead of you manually following hyperlinks and consuming information, the software will do all this instead, acting autonomously - perceiving environments, making independent decisions, and taking action with minimal human intervention. This is the beginning of the “agentic economy" where these autonomous AI agents will be fully interconnected and transact in real time.

“Since the 90s, I haven’t felt this excited about being a software developer or technologist…I can comprehend how each piece operates and how they fit together, allowing me to explore freely” - Kevin Scott, Microsoft CTO on the agentic web at Microsoft Build 2025

____________________________________________________________________________

A new product architecture for an agentic web

I am a software engineer, so I will dive deeper into what this means for engineers and will touch on product owners and designers a bit less. For people like me, this will fundamentally shift focus away from building static interfaces to creating dynamic, intelligent systems that self optimise, anticipate needs and execute complex workflows without explicit human direction.

Engineers will be adopting this new paradigm I mentioned earlier, using MCP for a standardised way of communicating between different AI systems and data repositories. From Anthrophic’s specification, MCP will:

Share contextual information with language models
Expose tools and capabilities to AI systems
Build composable integrations and workflows

NLWeb is a newly announced, open, technology agnostic project from Microsoft that simplifies creating natural language interfaces for websites usable for both humans and AI agents. Taking existing semi-structured formats like Schema.org, RSS and other data that websites publish that then is enhanced with LLM-powered tools to incorporate external knowledge (like layering geographic insights to a restaurant query) to enrich the user experience. NLWeb enables developers to turn any website into an AI app where users can query site content through natural language.

Every instance of NLWeb also functions as an MCP server, supporting the fundamental ask method that enables both human-site interactions and agent-site interactions through the same protocol. The key idea of NLWeb is to allow developers to integrate MCP with the minimal amount of code to their existing client side code.

How NLWeb works

‍____________________________________________________________________________

Digital products of the close future

These new technologies I have talked about so far are mainly what the web will start to look like over the coming years (or months!) as web apps turn into agentic workflows, but what about native apps? Apple and Google have been implementing similar protocols into their ecosystems for a while now, like with App Intents (in the case of Apple), so both are setting up the agentic highway to enable this new paradigm to unfold. Although Apple has a lot of catching up to do with its model capabilities, they are integrating with 3rd party models like Claude and ChatGPT and doubled down on that at WWDC.

AI has been hot the past few years, but I do believe that what is going on now does warrant a lot of hype: we are reaching a point where there is a clear foundation being laid in the form of these new open standards for what will be the next major iteration of how we interact with the digital world. Microsoft CTO Kevin Scott's quote captured this excitement perfectly. The agentic web presents similar opportunities for exploration and innovation as software engineers shape the next evolution of the internet.

Furthermore, these new intelligent products will automate a lot of what product teams spend significant amounts of time trying to plan and optimise. For instance:

Self-optimising interfaces: an agentic app could dynamically adjust UI/UX based on real time user behaviour analysis and environmental factors (e.g. switching to voice-first mode when driving)
Predictive resource management: automatically scales cloud infrastructure, adjusts API rate limits, and allocates computing resources based on usage patterns
Continuous A/B testing: where AI agents autonomously deploy micro-variants and analyse performance without human intervention
Self correction: dynamically reroute processes when detecting API failures
Intelligent sourcing: switch data sources based on freshness/accuracy metrics
More data made availability: combine structured and unstructured data processing

____________________________________________________________________________

The transformed product team

A new product team structure that is adapted to this new paradigm:

Traditional Team → Agentic Team

PM → Product Architect

Designer → Agent UX Specialist

Engineer → Workflow Orchestrator

Data Scientist → Ethical Alignment Engineer

The new core responsibilities for a product team in this agentic economy:

Intent Architecture Design
1. Defining decision boundaries and success metrics for AI agents
2. Creating failure mode playbooks (e.g., “If conversion drops 15%, revert to last stable version”)
Ethical Alignment Engineering
1. Implementing constitutional AI principles
2. Designing audit trails for autonomous decisions
Agent UX Design
1. Creating dual interfaces:
  1. Human-facing: Traditional GUI
  2. Agent-facing: Structured data endpoints and semantic markup
2. Developing cross-agent communication protocols

Strategic Implications

Product teams must master:

Agentic system debugging - Tracing decisions through multi-agent chains
Dynamic workflow monitoring - Using tools like LangSmith for AI ops
Cross-domain optimisation - Balancing human/AI interaction points

____________________________________________________________________________

The bottom line

These recent announcements at these conferences are a canary for the fast approaching agentic economy. Large companies like Microsoft and Google are already laying the groundwork for this new infrastructure that will shape the next iteration of the internet and the new product opportunities it will enable. It is vital to be thinking about how this will disrupt our industry and how things will look moving forward, and this post is designed to give an awareness of what is soon to come.

Get the latest news and views from Loomery directly to your inbox

Stay ahead of the curve with our monthly newsletter, The Weave.

Newsletter sign up

The internet of the next decade is autonomous, what should your product team do to prepare?

The emergence of new open standards and foundations for the coming agentic web

June 2025

Get the latest news and views from Loomery directly to your inbox

Company news

Innovation

Move fast

Let's talk →

The internet of the next decade is autonomous, what should your product team do to prepare?

The emergence of new open standards and foundations for the coming agentic web

June 2025

Get the latest news and views from Loomery directly to your inbox

Score your team against the 8Cs

Company news

Innovation

Move fast

Let's talk →