FRONTIER CONSULTANCY · L200

CAIP Frontier Consultancy

Agents take the clicks. Consultants take the architecture.

Scroll or press ↓ to begin

ACT ONE

The Opportunity

Where your margin is hiding, and why it is recoverable now.

WHAT PARTNERS ARE TELLING US

Three things we keep hearing.

Different firms, different regions, same three patterns in every partner conversation this year.

🎯

Talent is stretched thin

"We are short on skilled consultants, and the ones we have spend a quarter of their week on the boring work."

📈

Demand is ahead of staffing

"Customers are not asking if agents fit. They are asking how many they can buy. We are selling agents faster than we can staff them."

⏳

Long projects carry risk

"Modernisation projects are long, expensive, and the outcomes are not guaranteed. Every long project is a bet."

If you recognise one of these, you are not alone. If you recognise all three, this session is for you.

THE HOOK

Reclaim 25% of your team's time

Three numbers every CAIP leader should know. One of them is already moving.

60–80h

Documentation overhead per engagement (Forrester)

15%

Consultant time on unbilled admin (McKinsey)

25–40%

Overhead reduction possible with AI (GitHub Octoverse)

Public proof points
Fellowmind launched Hive AI, an agentic platform for delivery. The flagship agent matches consultant CVs to RFP requirements, a problem every consultancy has. Built by a consultancy, productised for consultancies. Hive AI portfolio →

Accenture's published numbers. PR cycle time from 9.6 days to 2.4 days. 84% more builds passing CI. The productivity case for AI-first delivery is no longer theoretical. GitHub research →

The Partner Delivery Lifecycle

Five phases. AI compresses every one. The biggest compression sits at the front.

Pursue

→

Scope

→

Build

→

Ship

→

Run

60–70% of every project's consultancy overhead lives in Pursue + Scope. Workshops, proposals, requirements, architecture sketches. That is where AI-first compounds first.

THE MODEL

From phases to a delivery model

Every phase has a lead role, AI assets, a clean handoff, and a human gate. The shape is the value, the tools fill the cells.

Phase

Lead

AI Assets

Handoff

Human Gate

Pursue

Account Lead

M365 CopilotAzure CopilotCopilot StudioMicrosoft Learn MCP

Proposal · architecture sketch · estimate

SA sign-off

Scope

Solution Architect

WorkIQCopilot StudioFoundry templatesArchitecture Center

Plan · ADRs · reference design

Customer sign-off

Build

Engineer · Tech Lead

Copilot agent modeSpec KitAzure MCPADO MCPcustom MCP

Working code · tests · ADO updates

PR review · CI gate

Ship

Engineer · SRE

GH Actions + CopilotIaCAzure MCP

Deployed environment · release notes

Change advisory · deploy gate

Run

SRE · Knowledge Eng

Azure MCPContinuous AI workflowsevalsM365 Copilot Cowork

Telemetry · runbooks · captured learning

Incident review · eval thresholds

The cells change every six months as the stack evolves. The columns don't.

ACT TWO

The CAIP AI-First Toolbelt

Three lenses on the same toolbox, by who uses it, where speed compounds, and what holds the rest together.

PHASES 1 + 2 · PURSUE + SCOPE

Client-facing phases, where partners lead

Pursue and Scope. AI replaces the manual grind. Human judgment stays where it belongs.

💼

Pursue

Account research, proposals, architecture sketches, estimates. M365 Copilot drafts proposal content. Azure Copilot pulls reference architectures and cost models. Copilot Studio is where partners are building their own pre-sales agents (RFP analyzers, CV matching, win-loss). Pre-sales stops being a blank page.

🔍

Scope

Workshops, requirements, ADRs, reference design. M365 Copilot handles recaps and action tracking. WorkIQ surfaces how the customer's teams actually work. Copilot Studio prototypes the agent live in the workshop. Foundry templates and Architecture Center accelerate design decisions.

PHASES 3 + 4 + 5 · BUILD · SHIP · RUN

Delivery execution, where speed compounds

Build, Ship, Run. GitHub Copilot agent mode does the routine. Senior judgment stays on the gates.

⚡ Build ▾

GitHub Spec Kit turns plans into structured tasks. Copilot agent mode drafts code, runs tests, opens pull requests. Azure MCP and ADO MCP supply live cloud and project context. 30–50% faster on routine build work: boilerplate, tests, infra scaffolding. PR review and CI stay human.

🚀 Ship ▾

GitHub Actions with Copilot assist. IaC for environment parity. Deployment drift caught before it reaches production. Release notes drafted automatically. Change advisory and the deploy gate stay human.

📈 Run ▾

Azure Copilot for monitoring. Incidents summarised, root causes proposed, runbooks generated on the fly. Continuous AI workflows keep docs fresh and triage issues. Eval thresholds catch model drift before customers do.

Always on, alongside the lifecycle M365 Copilot Cowork. Flexible automation for repeating admin work that doesn't belong on a consultant's desk: status roll-ups, ticket triage, documentation nudges, recurring report generation. The lifecycle keeps moving while Cowork handles the noise.

PHASE 3 · CONNECTIVE TISSUE

The Project Management Backbone

The connective tissue. WorkIQ, ADO + GitHub Copilot, the MCP catalog, and Cowork, four surfaces working in parallel.

📋

ADO + GitHub Copilot · Remote MCP

Work items, sprints, PRs, managed conversationally. Both ADO MCP and GitHub MCP ship as remote endpoints in public preview. Pick the platform your customer runs, the pattern is the same. No more SDK plumbing.

📊

WorkIQ + Power BI

Workplace analytics on how your delivery team actually spends time. Power BI with Copilot surfaces risk across portfolios.

🧩

The Microsoft MCP Catalog

github.com/microsoft/mcp. Azure, ADO, Fabric, Learn, Power BI Modeling, SQL, GitHub, Clarity, Data Factory, one place.

🔁

M365 Copilot Cowork

Multi-step Copilot agents for the admin loop, recurring status, retro summaries, intake triage. Take repeating PM tasks off the delivery team.

2026 shift: the MCP catalog is now a first-class Microsoft surface, but it's one connector among several layers your team operates on.

ACT THREE

Live Demo

Copilot CLI. Azure MCP. ADO MCP. Customer transcript in. Landing zone out.

Pre-flight: managed identity · scoped access · audit trail · human review gates. The trust boundary is wrapped around what you're about to see. Full breakdown in act eight.

ACT THREE

Copilot CLI + MCP, grounded in customer context

From customer recap to landing zone, in one prompt.

Live Demo

Transcript

Prompt

Azure MCP

Bicep

PR + CI gates

ADO work item

Hand off

ACT FOUR

Four Delivery Paths

All partner-led. Start with one. Most engagements end up running two or three in parallel.

Four paths, all partner-led

Workshop-to-Plan is the universal entry point. The other three branch from it.

🗺️

Workshop-to-Plan

Your lowest-barrier entry. Any partner can run an AI-accelerated workshop this week. Turns a two-day scoping exercise into a half-day.

🔄

Agentic Modernization

App and data migration. You lead the TCO assessment. Microsoft's Cloud Accelerate Factory backs you on the execution.

🤖

AI Path to Production

The meta-play. Use AI tools to deliver AI projects. Microsoft Foundry Agent Service at the core.

⚡

AI-First Delivery

GitHub Copilot in agent mode across the full SDLC, with MCP, skills, evals, and Continuous AI workflows around it.

PATH 1

Workshop-to-Plan

Six steps. From customer research to delivered plan. Traditional 15–25h becomes 4–6h.

1. Research. AI-assisted customer context via MCP (30 min, was 2–4h manual)
2. Prep. Copilot-generated deck and agenda (1h, was 3–5h from scratch)
3. Workshop. Teams Recap + live transcription (no scribe needed)
4. Capture. Action items auto-extracted and assigned
5. Synthesis. AI drafts findings and recommendations
6. Deliverables. Plan delivered in 48 hours instead of two weeks

15–25h

Traditional workshop cycle

4–6h

AI-first workshop cycle

48h

Plan in customer's hands

Tools: M365 Copilot · Copilot Studio · Azure MCP · ADO MCP · Teams Recap

PATH 2

Agentic Modernization

App and data migration, accelerated end to end. You lead. Microsoft's Cloud Accelerate Factory backs you.

Discover

→

Plan

→

Execute

→

Validate

→

Operate

Azure Migrate · GitHub Copilot · Modernize Copilot · Cloud Accelerate Factory · Azure AI Foundry

PATH 3

AI Path to Production

The meta-play. Use AI tools to deliver AI projects. Grounded in Microsoft Foundry.

1. Readiness. AI maturity assessment, use-case prioritisation, data readiness audit
2. Foundations. AI Landing Zone, governance, Responsible AI framework, security baseline
3. Build. Microsoft Foundry Agent Service. Prompt agents, workflow agents, hosted agents, manifests
4. Govern. Production monitoring, cost management, agent identity / OBO, evaluation in CI

🧠

Prompt Agents

GA. Reasoning + built-in tools (web search, file search, memory, MCP).

🔀

Workflow Agents

Preview. Deterministic multi-step agent orchestration.

🛰️

Hosted Agents

Preview. Managed compute, private networking, agent identity.

Microsoft Foundry, learn.microsoft.com/azure/foundry · rebranded from Azure AI Foundry in late 2025

PATH 4

AI-First Delivery with GitHub Copilot

Four surfaces. GitHub Copilot powers them all. Partner IP compounds across all four, and custom MCP is only one of them.

⚙️

GitHub Copilot, the engine

Agent mode, CLI, VS Code, code review. Across the full SDLC, not a productivity tool on a laptop.

🔌

MCP, the connector

Microsoft catalog (Azure, ADO, Fabric, Learn, SQL…) plus your own custom servers. One protocol for tools and partner data.

📚

Skills · prompts · evals

Custom modes, prompt libraries, eval suites. Your methodology codified, your quality bar measurable before deploy.

🔁

Continuous AI workflows

GitHub's repo-level agents, continuous documentation, triage, fault analysis, quality. Team-scale AI on every PR.

MCP connects. Skills capture method. Evals measure quality. Continuous AI runs on every PR.

ACT FIVE · SYNTHESIS

Build your AI delivery toolbox

Six roles every CAIP partner has. Six agent patterns. What is each person doing on Monday that they were not doing on Friday?

🏔️

Solution Architect

Reference designs and ADRs in hours, not days. Sizing drawn from live Azure inventory. Tools: Azure MCP, Architecture Center, Foundry templates.

🧭

Account Lead

Proposal first drafts grounded in customer context. Past wins and competitive intel surfaced before the call. Tools: M365 Copilot, Copilot Studio, Microsoft Learn MCP, SharePoint.

📊

Delivery Manager

Monday status report written for her, she reviews and forwards. Sprint health and risk flags summarised. Tools: ADO MCP, Power BI MCP, WorkIQ.

🛠️

Tech Lead

PR reviews with full repo context. Agent is scribe in design reviews. Spec-to-PR loop on routine work. Tools: GitHub Copilot, Spec Kit, ADO MCP.

🤖

AI Engineer

Foundry templates scaffolded. Eval harnesses generated from spec. Prompt versioning in CI, not in a notebook. Tools: Azure AI Foundry, Copilot Studio, GitHub Actions.

🗄️

Data Engineer

Pipeline drafts from sample data. Schema sketches from source docs. RAG indexes review-ready before the standup. Tools: Fabric, Azure AI Search, SQL MCP.

Start with the Azure Skills Plugin (25 skills · 200+ MCP tools · one install). Layer your delivery IP on top. 2–4 weeks per partner-specific agent. No model training required.

ACT SIX · ACTION

What to do Monday morning

Three tiers. Pick the one that matches where your delivery team is today, and the metric line that proves you moved.

🌱

Just Starting

Turn on Teams transcription for every client meeting. Activate M365 Copilot for your delivery team. Start a GitHub Copilot trial and install the Azure Skills Plugin (one command, 25 Azure skills, 200+ MCP tools). Run one Workshop-to-Plan this month.

Measure (adoption)
% meetings transcribed · weekly active Copilot users · workshop hours per engagement

🚶

Have the Basics

Adopt the Microsoft MCP catalog across your delivery team. Ship your first AI-first engagement. Build your first Copilot Studio agent. Stand up a prompt library for your top five plays.

Measure (flow)
time-to-first-PR · rework rate on AI-drafted artefacts · documentation freshness

🏃

Ready to Scale

Build custom MCP servers for your IP. Publish prompt libraries by practice area. Wire AI into your methodology, not just your tooling. Measure and publish the productivity gains. This is your CoE, standards, repeatable assets, KPIs.

Measure (economics)
gross margin per engagement · billable utilisation · defect escape rate

If the thesis is economic, the metrics have to be too. Pick three from your tier and report them quarterly.

ACT SEVEN · HANDS-ON

One prompt. Copilot did all of this.

Paste a single prompt into Copilot CLI. It fetches pages it has never seen, downloads code, edits your config, and reboots itself. You never opened a file.

📥

Downloaded the exercise

Picked a sensible folder. Cloned the repo. Moved into it. You never typed git clone.

🌐

Read a docs URL

Fetched the Microsoft Learn MCP setup page and figured out the right config by reading prose. No cryptic JSON you had to decode.

📝

Edited your config

Wrote the correct JSON into ~/.copilot/mcp-config.json alongside whatever was already there. You never opened the file.

🔁

Reloaded itself

Restarted its MCP stack. Verified the new tools are live. Told you how to confirm. No 'now restart' step you had to remember.

Delivery stops being clicks. Delivery starts being instructions.

ACT SEVEN · HANDS-ON

Now it's your turn

Twenty minutes. One client. One tool. A brief worth sending.

🛰️

The client

Nordwind Aerospace, small-sat operator, just landed a reinsurance contract. Their CTO wants a one-pager: "Can we just use ChatGPT, or do we need something more?"

🧭

The tool

Copilot CLI + Microsoft Learn MCP. You'll prompt Copilot to install the MCP itself, then use grounded Microsoft guidance to draft the brief. Not vibes. Cited.

📝

The deliverable

brief.md, one page, honest, specific to Nordwind. Azure Landing Zone for AI + Microsoft Agent Framework as the backbone. Three next steps.

Tiered setup, fresh account, baseline, stretch, super-stretch. Nobody sits out.

WHAT'S NEXT

Four ways to go deeper

Pick what fits. One of these becomes your next session.

⚡

Hackathon

Build something real together over 1-2 days. Your team, our hands on the keyboard alongside. Walk out with a working agentic asset you own.

🎯

Product deep dives

60-90 minute masterclasses, modular. MCP design, Agent Framework, Azure AI Landing Zone, grounding and evaluation, agent observability. Pick the ones that fit.

🤝

Co-delivery

We embed on a live customer engagement. AI-first delivery applied in your client context, then handed back. One referenceable engagement out the other side.

👁️

Shadow + debrief

Watch us run an AI-first discovery with a real customer. 90-minute debrief after. Lowest-effort way to see the loop in live context.

More options on the table▾

📚

Pattern library

Joint sessions to build your delivery IP. Agentic discovery script, AI-augmented SOW templates, agent-enabled project plans.

🎓

Train-the-Trainer

Certify 3-5 of your people to run this workshop internally. Scales beyond Microsoft involvement.

💬

Office hours

Recurring 60-min clinic for problems from live engagements. On-tap advice, low overhead, runs in parallel with anything else.

"Which one would you want to run first, and what would have to be true to make it happen?"

THE CLOSE

The Trust Boundary

Why this works at enterprise scale. Platform-grade trust on the data, the model, and the agent itself.

🔒

Data Security

Customer data stays inside the Microsoft trust boundary. No training on your customer data. Entra ID authentication. SOC 2, ISO 27001, GDPR.

✅

AI Quality

Production-grade guardrails. Content filtering, grounding in verified data, evaluation frameworks in CI, continuous monitoring.

⚖️

Responsible AI

Microsoft's six Responsible AI principles. Fairness, reliability, privacy, inclusiveness, transparency, accountability, built into the platform.

🛡️

Agent Operations

Tool allowlists scope what each agent can touch. Approval gates on every write action. Every action logged, every change reversible. The rails are what let the agent move fast safely.

Johan Wallquist · Partner Solution Architect, Microsoft · aka.ms/caip-frontier

APPENDIX · IN YOUR POCKET

From individual productivity to delivery factory

Multi-agent topology that lives in your repo.

▶

Demo · Squad in action

Pre-recorded. One direction in, four agents fan out in parallel, file-write guards hold, you review the merge.

Pending capture · assets/demo-squad.mp4

🚀 Specialists in parallel

Roles fan out together on one direction you set. You review the merge.

🛡️ Allowed actions

File-write guards, PII scrubbing, reviewer lockout. Rules in code, not prompts.

🧠 Persistent memory

Decisions and conventions stored in .squad/. Inspectable with your code.

🐙 GitHub-native, SDK-first

TypeScript squad.config.ts. copilot --agent squad. Alpha · MIT.

Brady Gaster · bradygaster.github.io/squad · Surfaced from Juan Manuel Servera Bondroit's review