Blog
Accelerators, prototypes, and build notes. Personal projects; views are my own.
Personal project notes, separated by lane.
14 posts
Sales assistance
What to turn on for sellers: the Sales agent in Microsoft 365 Copilot, Dynamics 365 Sales skills in Copilot Cowork, and an optional accelerator that makes briefs, follow-ups and handoffs consistent.
Agent patterns
Microsoft renamed the GitHub Copilot harness to the Microsoft Copilot harness. Here's what each Copilot Studio harness is for, which to pick, and how to think about agents you've already built.
Decision support
Size an AI agent's monthly running cost as an honest low / likely / high range, pick the right harness, and weigh cost against business value. Unconfirmed rates are flagged on purpose, with the reason why.
Agent patterns
A private, issue-first pattern where every build request becomes a spec'd GitHub issue, a coding agent opens a small pull request, and I review every one before merging.
IT administration
A practical administrator preview turns performance questions into sourced evidence, visible gaps and controlled change planning.
Personal systems
A React reference build combines Home Assistant cards and custom estate controls without publishing real household configuration or treating self-monitoring as a supported service.
Business applications
A control-tower prototype reframes shipment signals as visible, aging, assignable work instead of another thread in an inbox.
Information design
An unpublished static demo makes the filter and digest interaction tangible while clearly separating sample content from live product news.
Business applications
An adaptive form component assembles related Dataverse context without turning context gathering into another application.
Agents and AI
A public starter kit uses GitHub Issues as the control surface for specialist agents, explicit handoffs, and human-reviewed outcomes.
Business applications
A Power Platform accelerator structures intake and review while preserving the human decision boundary that matters.
Agents and AI
A multiplayer escape room makes instructions, constraints, iteration, and reflection part of the learning loop.
Agents and AI
Deal to Glory makes a five-stage workflow visible while its source reveals the important limits: deterministic extraction, illustrative calculations, simulated notification, and an ungated optional CRM write path.
Developer labs
A deliberately small joke server makes the protocol boundary, deployment step, client connection, and user invocation easy to inspect.
No blog posts match this filter.