Microsoft Developer Video Summaries

Gareth Bland explains how enterprises can make strategic decisions about token spending in agent-based systems. The discussion connects token costs and context caching with measures of agent efficiency and the value delivered by software features.

- Token cost dynamics: input, output, caching, and the effect of longer agent loops
- Agent efficiency measures: autonomous completion, production adoption, and user value
- Feature investment: using an S-curve to distinguish early development, leverage, and saturation

For enterprise teams building agentic software, this provides a framework for tracking token use and assessing where further investment may create value. It focuses on measurement and strategic decision-making rather than implementation instructions or a specific vendor’s tools.

Mahmood, a Microsoft MVP and Azure delivery lead, demonstrates how GitHub Copilot can help transform meeting transcripts and technical proposals into structured project delivery materials. The walkthrough covers the agent’s workflow, its use of reusable skills and scripts, and how Microsoft Learn MCP resources can inform recommendations.

- Agent workflow: converting PDF or transcript input into structured content and identifying project requirements
- Generated materials: project plans, sequenced tasks, kickoff questions, summaries, and recommendations
- Implementation details: VS Code agents, reusable templates and scripts, existing project context, and Microsoft Learn MCP
- Sharing and adaptation: organizing agents, prompts, skills, and scripts in a GitHub repository

Designed for delivery leads, architects, project managers, and teams exploring AI-assisted planning, this walkthrough offers practical ideas for adapting agent instructions and templates to their own projects. Viewers can review the shared repository and tailor its components to their workflows.

I demonstrate how a Foundry-hosted supply agent uses Microsoft IQ to connect business context with practical actions. A supply autopilot example shows how four IQ capabilities support information retrieval, supplier analysis, and email workflows.

■ IQ capabilities in Foundry
- Web IQ, Foundry IQ, Fabric IQ, and Work IQ in a supply-management scenario
■ Data and knowledge
- Knowledge sources, semantic models, ontologies, and data-agent traces
■ Agent actions and administration
- Mailbox identity, delegated permissions, licensing, and Teams setup

The walkthrough is aimed at developers and administrators exploring enterprise agents. It provides an overview of the demonstrated architecture and setup areas; use the resources at ak.ms/iqseries to learn more and begin building.

Azure Savings Plans for Databases offer a spend-based commitment intended to help teams manage costs as database workloads and architectures change. I explain how the hourly commitment works, how it differs from reservations, and how to purchase and review usage in the Azure portal.

■ How the savings plan works
- Eligible services, hourly commitments, discount application, and pay-as-you-go usage beyond the commitment
- Comparison with Azure reservations and considerations for combining the two
■ Purchasing and tracking
- Finding Azure Advisor recommendations and purchasing through the Azure portal
- Using Cost Management cost analysis to review resources receiving the benefit

For FinOps teams, architects, and Azure database administrators, this provides an overview of the commitment model and portal workflows. Use Azure Advisor to review recommendations for your historical usage, then consult Cost Management and billing to assess purchases and track benefits; detailed service-by-service rates are not covered.

No timestamps were provided in the transcript, so topic timestamps are omitted.

Learn how to set up Model Router Auto Evaluation to compare Microsoft Foundry Model Router with a baseline model on quality, cost, and latency. The walkthrough covers configuration, running an evaluation, and reviewing a sample results dashboard.

- Project setup, endpoints, API keys, model deployments, and pricing
- Evaluation data and running the command-line workflow
- Reviewing dashboard comparisons and planning follow-up tests

For developers choosing models for AI agents, this walkthrough explains how to run an evaluation and interpret its trade-offs. You can use the same process with your own prompts and agent priorities; the example uses a small sample set, so its results are not a general benchmark.

I discuss how Azure SQL brings approximate vector search into relational workloads, including how indexing works alongside SQL filters and data changes. Product and engineering guests use demos to explain the feature’s capabilities and query-planning behavior.

- Customer needs and approximate nearest-neighbor search with the DiskANN index
- Searches combined with relational filters and joins, plus insert, update, and delete operations
- How the optimizer chooses between exact and approximate search, and a large-scale search demo
- The discussion focuses on Azure SQL’s vector index rather than a general survey of vector-search systems; the feature is announced as generally available

For developers and data professionals evaluating vector search in SQL, this provides an overview of the design and practical query patterns. Try the generally available Azure SQL vector index with a workload of your own to assess its fit.

Learn how Work IQ Developer Tools and GitHub Copilot CLI can support the end-to-end development of plugins and declarative agents, from gathering work context to publishing and usage insights. A fictional insurance claims scenario demonstrates the workflow with an MCP server, skills, and web search.

■ Development and validation
- Connect to an MCP server, create a claims plugin and frontline agent, and test them in Copilot surfaces.
- Use colleague feedback to refine the agent and build an evaluation set.

■ Publishing and follow-up
- Prepare and publish to a tenant catalog, then review usage insights to guide further updates.
- Partner Center and marketplace publishing are noted as still in progress.

For developers building Copilot agents or plugins, this walkthrough provides a practical view of the tools and lifecycle. Use it to identify a project outcome, try the build-and-test workflow, and iterate based on user feedback and evaluations.

I explain how Copilot Cowork supports delegated, multi-step work and how Microsoft Scout is designed to identify and act on work proactively. A customer-feedback workflow provides a practical example of using AI and reusable skills to reduce repetitive effort.

■ Copilot Cowork and reusable skills
- How task delegation, context, planning, and approval for sensitive actions fit into a workflow
- Turning recurring instructions into skills; a customer-feedback and newsletter example
■ Microsoft Scout and choosing a tool
- Scout’s proactive assistant model and its relationship to chat and Cowork
- Early-access program and audience Q&A

For Microsoft 365 users looking to apply AI to recurring work, I outline an action plan: try one real task with Cowork, refine the instructions, and save them as a skill if the task repeats. The discussion focuses on these workflows and product concepts rather than a step-by-step technical setup.

I explain how AI is changing work through task automation, new roles, and ongoing skills development. A staged demonstration using Microsoft tools shows how to match AI capabilities to a real business task rather than choosing tools based on model hype.

■ AI at work and choosing tools
- Domain expertise, task support, drafting, automation, and delegated actions
- Copilot, proactive assistance, and background automation
■ Building and managing AI solutions
- Microsoft tool roles, reusable agents, and enterprise governance
- A practical plan: define the task and outcome, build a small solution, test it, and assign ownership

I also point to AI Skills Navigator as a resource for continued learning. This is aimed at information workers, builders, and IT enablers seeking a framework for selecting, testing, and governing AI tools; it focuses on practical workflows rather than predicting which model will prevail.