Practical AI for enterprise architects
45:00

Start presenter mode

People on this site will follow your slide and interactive choices.

Practical AI for
enterprise architects

01Designing a skill
02Connect MCP
QR code linking to this workshop presentation Scan to follow along
Scroll or use ↓

One company. Two possible first transitions.

Inherited from PRO-FIT / Legally YoursShared applicationsProposed change

Your prompt enters the harness.

Source · Anthropic Skills overview ↗

The model chooses what to load, one step at a time.

Source · How Skills work ↗

Every token has a price.

Turn 1 · selected tokens0
Source · Context windows ↗

Context window

Turn 1 · input + output17,221 / 100,000
System contextInputOutput
Source · Context windows ↗

What would you change in this skill?

architecture-change-assessment / SKILL.md

          

Session 1 · 17 minutes

Improve the skill, then test what changed.

Read, define & run
Improve the skill
Rerun & evaluate
Share result
Every group: what does good look like?

“We’ll know it works when ; we’ll check this by .”

Session 2

Simulate the impact of your recommendation.

Harness
MCP server
Enterprise architecture repository

An MCP server makes tools available.

Source · MCP architecture ↗

One harness. Multiple MCP servers.

HarnessManages connections, permissions and model context
MCP client 01
MCP client 02
MCP client 03
MCP serverCMDBe.g. ServiceNow
MCP serverEnterprise architecture repositorye.g. Caplo
MCP serverKnowledge basee.g. Confluence
Harness ⇄ModelContext in · requested actions and answers out
Source · MCP architecture ↗

MCP is more than tools.

Source · MCP server capabilities ↗

The same agent loop · now with MCP tools

The model chooses when to use a tool.

→ → → → →
Source · MCP tools ↗ Examples · Caplo tool catalogue ↗

Does everything over MCP use tokens?

Tool names, descriptions & schemasMade available to the model
The model requests a toolTool name + arguments
Results, resources & prompt templatesContent returned to the harness
Protocol traffic & server executionConnection messages, API calls, database queries
Source · Tool-use token accounting ↗

Session 2 · Group work

Connect. Simulate. Share what you discover.

01

Connect your harness

Add https://app.caplo.ai/mcp to your client. Sign in and select the workshop workspace. Find the ArchiSurance applications and relationships.

02

Draft the impact

Use recommendation.md to model the change in your group’s own Shared draft, in the shared repository. Inspect affected applications and dependencies.

03

Share your findings

Show the proposed change. What did the repository reveal that your original recommendation missed?

Each group sharesOur recommendation → the simulated change → one new finding → what we would revise

The architect’s contribution

Method. Evidence. Trade-off. Decision.

The AI can make the work faster and more repeatable. The architect sets the evidence threshold, handles uncertainty and accepts the change.

SkillA reusable way of working, loaded only when relevant.

MCPA standard connection to tools and context.

DraftA proposed architecture state that people can inspect.

Prepared for Bram Verswalm · ArchiSurance workshop material adapted from The Open Group case study and supplied model.

Continue the conversation

Let’s stay in touch.

Bram Verswalm

QR code linking to Bram Verswalm’s LinkedIn profile Scan to connect on LinkedIn
1 / 11