Topic hub
Copilot, Copilot Studio, Foundry, Fabric data agents and MCP servers: what fits, what it costs, what to fix first.
48 articles by the Solv Systems team
Microsoft's AI line-up changes monthly, and the names do not help: Microsoft 365 Copilot, Copilot Studio, Microsoft Foundry, Fabric data agents, Fabric IQ, Agent 365 and a growing set of MCP servers. Behind the naming there are a few durable questions. Which assistant or agent fits which job, what it costs to run, which model to choose, how to evaluate whether it works, and whether your data is in a fit state for any of it.
This hub brings together everything we have published on Copilot and AI agents across the Microsoft stack. It covers the platforms, with comparisons of Copilot Studio and Foundry and guides to building and evaluating agents, and the money, from Copilot licensing and credits to what an agent costs per month in GPT, Claude and Copilot terms. It covers the data side, including Fabric data agents, the Fabric and Power BI MCP servers, and giving agents secure access to OneLake. It also covers the foundations most AI projects skip: data quality, an AI-ready data estate and why pilots stall.
Governance of Copilot and agents, including Purview and Entra Agent ID, is in the data governance hub. Copilot features inside Power BI are also listed under Power BI.
Newest first.
Solv Systems is now a Microsoft Solutions Partner for Cloud & AI Platforms, qualifying in all three Azure pathways. What the designation means and how it is earned.
Solv Systems has earned the Microsoft Solutions Partner Digital & App Innovation (Azure) designation. What it covers, the certifications that count, and what it means for clients.
DP-800 earns the Microsoft SQL AI Developer Associate. Who it suits, the skills measured, how it compares with DP-600 and DP-700, a six-week plan and whether it is worth it.
Fabric IQ ontologies reuse your Power BI measures and bind business definitions to Fabric data for agents and Copilot. What they add, what is in preview, and how to start.
Three Claude models, two Microsoft routes. Where Fable 5.1, Opus 5.5 and Haiku 5.5 actually run, what they cost, and the data-processing settings UK and EU admins need to check first.
New Microsoft 365 Copilot Business licences now arrive with usage-based billing switched on and a 4,000-credit default limit. The dates, what it covers, and what admins should set first.
Agents fail quietly unless you test them. How to evaluate AI agents with Copilot Studio test sets and Foundry's rubric evaluators, traces-to-dataset and agent optimizer, and what to build first.
Fabric data agents in Copilot Studio are generally available through the Fabric IQ Data MCP tool. The setup, the User vs Maker authentication choice, and what to check before you publish.
Microsoft Copilot now has three parts: Home for Chat and Cowork, Code for building apps and Autopilot for always-on agents. What each does, where each stands, how it is billed and what to do first.
Routines in Microsoft Foundry are generally available: run agents on a schedule, once at a set time, or when a GitHub issue or Teams message arrives. How they work, the limits, and where they fit.
Microsoft Ignite 2026 runs 17-20 November in San Francisco, with a free digital pass. Dates, passes, the sessions worth watching, and what data and AI teams can reasonably expect.
GPT-6 Astra, Sol, Luna and now GPT-6.1 Sol are generally available in Microsoft Foundry. What each is for, what they cost by deployment type, and how to choose for production agents.
Microsoft Copilot pricing now pairs a per-user licence for everyday AI with Copilot Credits for Cowork, Code, Autopilot and frontier models. What each covers, the controls, and how to budget.
Which platforms offer reinforcement fine-tuning as a managed service? Microsoft Foundry, Amazon Bedrock and Google compared on models, graders, data and pricing - and when RFT is worth it at all.
AI did not create your data quality problems; it made them fluent. Why failures that hid in dashboards surface loudly through Copilot and agents, and the quality programme AI requires.
Before Copilot, agents or any AI over your data: the readiness checklist we run with clients across data foundations, semantic quality, security, governance and operations.
Copilot, Cowork, Foundry, Fabric, IQ, agents: the Microsoft AI portfolio in one readable map - what each layer is, what it costs in kind, and the order in which sensible organisations adopt it.
Foundry's model catalogue spans Microsoft's MAI family, OpenAI, Anthropic and open-source options. A selection framework built on task fit, cost per outcome and switchability, not leaderboard worship.
When a user asks Copilot a data question, a pipeline turns words into a governed query: matching, generation, execution, narration. How it works, where it breaks, and how to make it reliable.
Microsoft 365 Copilot, Copilot Chat, Copilot Studio, Copilot in Fabric, GitHub Copilot, Security Copilot: six products, four different charging models. A map of who pays what in 2026.
Both build AI agents; they are aimed at different builders. How Copilot Studio's low-code approach and Microsoft Foundry's pro-code platform divide the work, and when the right answer is both.
Copilot is the assistant Microsoft built; a Fabric data agent is one you configure. What each is, how they overlap, and a clean rule for deciding which answers your data questions.
Agentic analytics is the shift from reading dashboards to delegating analytical work to AI agents that plan, query and act. What it means in the Microsoft stack and what it demands of your data.
Copilot in Power BI writes DAX, drafts report pages, summarises visuals and answers data questions. What works, what disappoints, and what your model needs before any of it is trustworthy.
Copilot writes DAX, agents run analyses, and executives are asking the question out loud. What AI actually replaces in an analyst's week, what it cannot, and how the role is genuinely changing.
Most stalled AI initiatives fail the same way: the data foundation wasn't ready, the use case was vague, and nobody planned for production. How AI readiness turns spend into results.
Microsoft's 2026 AI strategy has five pillars: intelligence layers on governed data, its own frontier models, custom silicon, agents as a platform primitive, and the data foundation underneath.
Microsoft Copilot Studio is a low-code platform for building custom AI assistants connected to your own data and systems. Here is what it does, how it connects to Fabric, and when to use it.
MAI-Thinking-1, Claude Opus 4.6, and GPT-5 are all frontier-capable. This guide helps you choose the right model for your specific AI application, use case, and cost constraints.
The six-layer agent stack is a framework for mapping where your organisation sits on the AI agent maturity journey - from unified data foundation to fully autonomous AI operations.
Fabric IQ surfaces Fabric data intelligence directly in Microsoft 365 Copilot in Teams and Outlook. Here is what it means for Power BI users, report builders, and BI architects.
Agent skills for Fabric let engineers query Power BI datasets, trigger Fabric pipelines, and manage workspace assets from GitHub Copilot CLI using natural language - without leaving the terminal.
Maia 200 is Microsoft's AI accelerator chip and Cobalt 200 is its ARM compute chip. Here is what Microsoft's custom silicon strategy means for your AI and Fabric cost model over the next three years.
Frontier Tuning, announced at Build 2026, applies reinforcement learning to Microsoft's MAI models inside your own compliance boundary so agents learn your workflows and standards - with your data and the tuned model staying in your environment.
Windows PCs can now execute AI agent workflows locally using on-device NPUs through Windows Copilot Runtime. Here is what Microsoft's Build 2026 bet means for engineering teams and enterprise architects.
The bottleneck on enterprise AI is no longer model capability - it is context. The ability to give AI agents accurate, governed, sufficiently rich context at inference time now determines how much AI capability translates into business value.
Microsoft Foundry Agent Service is the platform for building, deploying, and operating production AI agents. This guide covers the architecture, the key decisions, and what it takes to go from prototype to production.
What an AI agent really costs to run in 2026: current GPT-6, Claude and Copilot Credit prices, the tool and licence charges teams miss, and one workload priced from $200 to $19,800 a month.
The narrative that AI makes junior data engineers obsolete is wrong. What AI compresses is rote mid-level work. Firms cutting grad programmes in 2026 are creating a capability crisis.
Microsoft has shipped four distinct MCP capabilities for Fabric. Each solves a different problem. Together, they make Fabric an agentic data platform. Here's how they fit together.
A published Fabric Data Agent can now act as an MCP server, letting Foundry agents, Copilot Studio, and custom AI systems call it as a tool. Here's what that changes for enterprise AI architecture.
The Fabric RTI MCP Server bridges AI agents and live data in Microsoft Fabric - exposing Eventhouse, Eventstreams, and Activator as MCP tools with natural-language KQL generation.
The Fabric Pro-Dev MCP Server grounds AI code generation in the current Fabric API spec and item schemas. Here's what it changes for developers building on Fabric - and why it matters.
The Fabric Core MCP Server lets AI agents perform real, authenticated operations against your Fabric tenant - workspaces, permissions, capacities - through natural language conversation.
The Power BI Authoring MCP server, formerly the Modeling MCP server, lets AI agents build, refactor and document semantic models. What it changes for BI developers - and what it doesn't.
Microsoft Copilot Cowork turns M365 Copilot from an answer engine into an execution engine. You describe an outcome; Cowork plans, executes, and checks in until the work is done.
Microsoft Foundry is the production platform for enterprise AI agents on Azure. For organisations running Microsoft Fabric, it's the missing layer above your data estate.
A Fabric Data Agent lets users ask questions in plain English and get governed answers from your Lakehouse or semantic model. Here's how to build one that actually works.
Articles filed under another topic that also cover Copilot and AI agents.
We build Copilot and agent solutions on data that is actually ready for them. Book a free 30-minute call to find the first use case that will pay off.