Packaging the platform in Cortana and everyday surfaces

Video segment: 10:21–11:34

Learning goal

The earlier chapters introduced shared model access, tools, agent environments, reusable skills, and connected context. This chapter explains how those layers become usable through one assistant. By the end, you should be able to trace a question from Slack, a command-line interface (CLI), or the web through Cortana to those shared capabilities, and explain how a personalized setup adds team-specific behavior and context.

Cortana packages shared capabilities

Uday describes Cortana as Uber's AI assistant and the way to package the platform for people across the company. To understand its role, keep the earlier layers separate:

• The model gateway is the shared path for model requests.
• The MCP gateway is the shared path for tools. MCP is the tool-integration mechanism used in the talk; the speakers do not expand the acronym.
• A skill is a reusable agent capability or workflow.
• The context graph is the connected representation of software, organizational, documentation, incident, and data context that agents can query.

Cortana sits above these capabilities as an access layer. It does not replace the model gateway, MCP gateway, or context graph. It gives users one place from which to reach them instead of making every user assemble the platform's parts separately.

A surface is an interface through which the assistant is used. Uday names Slack, CLI, and web surfaces. The same shared capabilities can therefore be reached from a team conversation, a terminal, or a browser interface. The surface changes the way a person interacts with Cortana. It does not need to create a separate model, tool collection, or context system.

Click to enlarge

The slide shows Slack, Web UI, CLI, MCP, and SDK around a Cortana Platform. Its backend row names Auth, Harness Router, Skill Router, and Personalization. Its foundation row names Skills, MCPs, Context Graph, and DevPods. The diagram supports the role of Cortana as a common entry point to the earlier platform layers. Source at 10:30

The slide also frames practical needs that the platform is intended to address. Users should be able to reach AI without setting up a local terminal. Teams may need different bots with allowlisted skills and MCPs. Some workflows need to run unattended. These visible labels explain why a user-facing access layer matters, without specifying how authorization or unattended execution is implemented.

Trace a question through the assistant

Uday says that a person can ask a simple question from any of the named surfaces. Cortana can look up the context graph, invoke a skill, check code in a codebase, and provide an answer across those surfaces. The exact internal call order is not given, so the following is a teaching reconstruction of the capability path:

question from Slack, CLI, or web
              ↓
           Cortana
              ↓
 context graph, skill, MCP/tool,
 model access, or codebase inspection
              ↓
       answer through the chosen surface

The useful distinction is between the entry point and the capability path. A user chooses a surface based on the task. Cortana can then connect the question to the relevant shared context, skill, tool, model access, or codebase information. The platform does not have to be rebuilt for each interface.

Teacher-created example: one question, three surfaces

Suppose an engineer asks, “Which service owns this endpoint, and what code calls it?” This question is a teacher-created example, not a question shown in the talk.

1. In Slack, the engineer asks in a team channel, where other people can participate in the conversation.
2. In the CLI, the engineer asks from a terminal while working in a development environment.
3. In the web interface, the engineer asks through a browser-based interaction.

The entry point changes in each case. Cortana can use connected context, a relevant skill or tool, and codebase inspection to form the answer. This illustrates the speaker's claim that one assistant can expose shared capabilities across surfaces. It does not imply that every user can see every repository or that Cortana bypasses access controls. The talk does not specify its authorization or tenancy model.

Personalization adds team context

Shared access does not mean every team works in the same way. Uday says employees can personalize Cortana by connecting a custom skill, a custom prompt, and a team Slack channel.

A custom prompt is an instruction supplied by a user or team that shapes how the assistant operates in a chosen context. A persona is the customized assistant setup for a particular use or team. The transcript describes the relationship between a persona, a custom skill, and a custom prompt conversationally. It does not define a formal schema.

Personalization changes behavior and context, not only appearance:

• Cortana still provides the shared platform and access to common skills, MCPs, and context-graph queries.
• The persona adds a selected custom skill and custom prompt.
• The team Slack channel supplies conversation context for that team.

Uday says this setup lets Cortana know the team's information and work like a teammate. That is the speaker's description of the experience. It is not a claim that Cortana has unlimited knowledge or unrestricted access.

Click to enlarge

The diagram visibly connects Cortana's Personas capability to “Custom prompt + Skills + Slack.” It keeps that capability inside the broader platform, alongside shared backend components and foundations such as Skills, MCPs, Context Graph, and DevPods. This supports the distinction between a team-specific persona and the common platform beneath it. Source at 10:52

This gives two different levels of reuse. The shared assistant is the common access point and capability layer. A persona is a configuration built on top of that layer for a particular team or purpose. Using Slack, the CLI, and the web does not create a separate assistant for each surface. Personalizing a team setup does not require the talk to claim a separate knowledge system for every team.

What the usage numbers do and do not show

Uday reports that employees created 300 unique personas in the last month and that Cortana handled more than 20,000 sessions per day. These are speaker-reported activity measures. They show that people were creating configurations and using the assistant. They do not establish answer quality, successful task completion, or business value. The plan supplies no definitions, denominators, or measurement method for the counts.

The same caution applies to access. The slide mentions allowlisted skills and MCPs, but the talk does not explain the authorization or tenancy model. This chapter can describe how Cortana packages and exposes the platform. It cannot infer who may access a team's context, a tool, or a codebase.

Uday then hands the discussion to Adam, who uses these packaged capabilities in the end-to-end feature walkthrough in the next chapter.

Check your understanding

1. If a user asks the same question in Slack and in the CLI, what changes, and what can remain shared?

The interface and interaction style change. Cortana and the underlying skills, MCPs, context graph, model access, and codebase capabilities can remain shared. The talk does not present each surface as a separate knowledge system.

2. What makes a shared Cortana setup team-specific?

Uday describes connecting a custom skill and custom prompt to a team Slack channel. That adds team-specific behavior and conversation context. It is more than changing the assistant's appearance or giving it a new name.

Synthesis

Cortana is the sixth building block because it turns the preceding platform layers into a common user-facing path. A question can enter through Slack, the CLI, or the web, draw on shared skills, tools, context, model access, or codebase inspection, and return through the selected surface. A persona adds a custom skill, prompt, and team Slack context on top of that foundation. The reported counts show adoption activity, while the detailed authorization model and formal persona schema remain unspecified. This packaged path leads into Adam's feature walkthrough next.

Source boundary

This chapter follows Uday's description of Cortana, its Slack/CLI/web access pattern, shared skills, MCPs, and context graph, team Slack personalization, and the reported persona and session counts. The teacher-created question and flow diagram explain the model but do not claim an exact internal call order. The talk does not establish outcomes from the usage counts, Cortana's authorization or tenancy model, or a formal boundary between a persona, custom skill, and custom prompt.