Personal research · Tampa, FL
Notes from the agent layer.
A personal research notebook on agentic AI inside Microsoft 365. Not currently consulting. Happy to trade notes if you're working on something similar.
- No hype, just results
- Prompting that holds up
- Agents past the demo
- Human-centered AI
- Audit-ready by default
- Microsoft 365 native
- Bounded autonomy
- AI enablement
- Evidence over vibes
Approach
What I'm working through, in public.
I write the briefs I wish were on every architect's desk. Posting them publicly so other people working on the same problems can find them.
Research notes you can read in an afternoon
No fluffy maturity models. Concrete briefs on what's shipping, what's leaking, and where to start.
Microsoft 365 as the substrate
Identity, endpoint, and Purview already do the unglamorous work. Most of what I read about starts here.
Bounded autonomy by default
Audit-ready, sensitivity-aware, human-in-the-loop where it matters. Trust earned through evidence.
About
Hi, I'm Chase.
By day I architect Microsoft 365 environments: identity, endpoint, and the quiet plumbing that everything else in the modern workplace rides on. By night I'm increasingly fascinated by the layer above. How AI agents are starting to live inside that workplace and actually do work.
I'm based in Tampa. I read more papers than I sleep some weeks, run experiments in a sandbox tenant most evenings, and post the ones that survive contact with reality. This site is where the survivors land.
If anything here resonates, or you violently disagree, I'd love to hear from you.
- Region
- Tampa, FL
- Day job
- M365 architecture
- Off hours
- Reading, testing, writing
- Mode
- Independent · curious
Method
How I work through a new topic.
Step 01
Read
Primary sources first. Vendor docs, engineering blogs, the people actually shipping. Skim the analysts after.
Step 02
Test
Run the patterns against my own M365 tenant or sandbox. Where do they leak? What breaks under permissions? Notes go in a folder.
Step 03
Write
Publish what survived contact with reality, with sources cited. If it didn't survive, write that too.
What I'm researching
The parts I keep coming back to.
Prompting, configuration, evals, and governance: the four corners I read about most. Each links to the brief where I put my notes.
Prompting that survives the model swap
System prompts, tool descriptions, and stage instructions: the load-bearing parts. What stays the same when models change underneath you.
Read the briefConfiguring agents that don't embarrass you
Identity, scope, grounding, tools, exit conditions, observability. The unglamorous setup that decides whether an agent ever ships.
Read the briefEvals that catch regressions before users do
Twenty representative inputs, weekly runs, side-by-side outputs. The cheapest version of the discipline that pays for itself.
Read the briefGovernance you can show a regulator
Sensitivity-aware grounding, audit trails that read like English, bounded autonomy by default, decision provenance from day one.
Read the brief
From the lab
Six briefs on the agent layer,
read in any order.
01 · Agentic AI
Agentic AI: from demo to delegated work
What people actually mean by 'agentic AI' in 2026, why the demo-to-production gap is wider than the keynotes admit, and what the early shipping pattern looks like.
Read brief02 · Work IQ
Work IQ: Microsoft's intelligence layer for delegated work
Work IQ is the part of Microsoft's 2026 Copilot story that's actually new. Here's what it does, why it matters more than the model behind it, and how it changes the M365 architecture conversation.
Read brief03 · Multi-Agent
Multi-agent orchestration: the five patterns that ship
Sequential, concurrent, group-chat, dynamic handoff, magentic. The orchestration patterns that actually appear in production agent systems, when each one fits, and the failure mode each one quietly invites.
Read brief04 · MCP & Tools
MCP and tool use: the protocol the agent era settled on
The Model Context Protocol went from niche Anthropic spec to industry default in eighteen months. Why that happened, what it actually solves, and where the rough edges are in 2026.
Read brief05 · Trust & Governance
Trust, governance, and the unglamorous half of agent work
Identity for non-human actors, decision provenance, bounded autonomy, audit. The work that determines whether an agent leaves the pilot, plus the 2026 frameworks worth borrowing from.
Read brief06 · The Stack
The 2026 stack: providers, models, and choosing what's load-bearing
A working architect's read on the Anthropic / OpenAI / Microsoft landscape in 2026. Where each one is strong, where each one is overhyped, and how to think about the choice when it's actually yours to make.
Read brief07 · Prompting
Prompting in 2026: writing instructions that systems can run
The prompt is no longer something you whisper to a model. It's the load-bearing config of an agent. This is the practical guide: anatomy of a real system prompt, tool descriptions side-by-side, ten best practices in order of impact, and a worked example you can lift.
Read brief08 · Agent Setup
Configuring agents: scope, tools, exit conditions
Most agents that fail in production fail at configuration, not capability. Here's the unglamorous starter pack: what to set, what to default, and what to instrument before you let an agent loose in your tenant.
Read brief
Get in touch
Working on something similar? I'd love to trade notes.
Send me what you're reading or building. I read everything, even if I can't reply right away. This site is a personal research notebook, not a consulting offering.