The problems that bring people here
Most readers arrive from one of our Copilot or AI agent guides with a specific problem. The engagement usually starts with one of these:
Copilot agents are spreading and security has not signed off
Makers are building agents in Copilot Studio, SharePoint oversharing is about to become Copilot oversharing, and the security team wants a control list before anything reaches a regulated workload. Our Copilot agents security checklist is the public version of the review we run; the Project Lobster and ClawPilot security guide is where many readers start.
The data governance work has to happen before the Copilot rollout
Sensitivity labels, DLP for Copilot, restricted content discovery and retention are prerequisites, not follow-ups. We sequence the Purview and SharePoint work so the rollout does not surface documents that should never have been findable.
Every vendor says AI will transform the business and none will say how
You do not need another strategy deck. You need a practitioner who has run agents in production and can say what works, what it costs and what goes wrong. We evaluate Copilot Studio, Microsoft Foundry Agent Service, OpenClaw and custom agents against the workflow in front of you and recommend one, including when the answer is the cheapest option.
Agents need identities, audit trails and an off switch
Governed production means agent identities in Microsoft Entra, Conditional Access, Graph permissions scoped to the task, every action logged separately from human activity, human approval for sensitive operations and token budgets with alerts. We design that layer first.
How an engagement runs
Every engagement is fixed-fee and remote by default, with on-site days available in Alberta and by arrangement across the United States and Canada. Hours cover Pacific, Mountain, Central and Eastern time. US clients are invoiced in USD and Canadian clients in CAD.
Assessment (one to two weeks)
Inventory of current automation and AI use, identification of high-value opportunities, platform evaluation and a prioritized roadmap.
You get: a written assessment, a platform recommendation with the reasoning, a cost model and a phased plan.
Copilot and agent security review
Identity and access for agents, Purview labels and DLP, SharePoint oversharing controls, agent governance settings, Copilot Studio environment and connector policies, audit logging, data residency and retention, mapped to the controls your regulator or auditor asks about.
You get: a findings report, a prioritized remediation list with the admin center location of every change, and a rollout sequence for the first 30 days.
Deployment in phases
The highest-return, lowest-risk workflow first. Agent architecture with identities, permissions, audit and approval flows; integration with Exchange Online, SharePoint, Teams, Intune, Azure and Entra ID; cost controls; monitoring.
You get: working agents in production, documentation and the runbook to operate them.
Platform evaluation
| Platform | Best for | Considerations |
|---|---|---|
| Copilot Studio | Microsoft 365-centric teams, maker-built agents | Licensing and credits, limited customization |
| Microsoft Foundry Agent Service | Developer-built agents on Azure, managed runtime | Consumption pricing, Azure dependency |
| OpenClaw | Multi-channel and browser automation, self-hosted control | Open source, you run the infrastructure |
| Custom agents | Unique workflows, maximum control | Highest development cost, full flexibility |
Comparing platforms in detail? The free AI Agent Platform Comparison Matrix puts the four side by side on cost, governance, integration and scale.
What we deploy
- Workflow automation: email triage, document processing, approval routing and report generation, the repetitive work that does not need human judgment.
- IT operations: compliance monitoring, incident triage, configuration drift detection and patch reporting, so the team can work on projects.
- Developer productivity: coding agents with governance controls, automated review, documentation and test scaffolding.
- Decision support: data synthesis, competitive monitoring and research automation that prepares the briefing so people make the decision.
Every deployment includes audit trails, human-in-the-loop approval for sensitive operations, cost controls, data residency controls and Entra ID integration with role-based permissions.
Proof, without naming names
We do not publish customer names or logos. We run AI agents in our own operations: OpenClaw, Claude Code, a vector memory system and custom MCP integrations handle real work every day, so we know what breaks and what costs more than expected. We have deployed AI agents that handle real workloads in regulated environments, connected to Microsoft 365 through proper Graph permissions, Conditional Access and managed identities rather than service account workarounds.
Kevin Kaminski has been a Microsoft MVP every year since 2008. He writes the Copilot Weekly and Microsoft AI Weekly series on this site, so the advice tracks the platforms as they change.
AI automation consulting FAQ
Which AI agent platform do you recommend?
The one that fits who builds the agent, who runs it and how much control you need. We evaluate Copilot Studio, Foundry Agent Service, OpenClaw and custom agents against your requirements and say which, with the reasoning.
Can you review the security of Copilot agents before we roll them out?
Yes. Identity and access, Purview labels and DLP, SharePoint oversharing controls, agent governance, audit logging and data residency, mapped to the controls regulated US and Canadian teams ask about.
How do you handle AI governance and compliance?
Audit trails, human-in-the-loop approval, data residency, model access controls and cost guardrails are in every design.
What does an engagement look like and cost?
A one to two week assessment first, then phased deployment. Fixed-fee, invoiced in USD for US clients and CAD for Canadian clients.
Can AI agents integrate with our Microsoft 365 environment?
Yes, through Graph permissions, Conditional Access and managed or agent identities rather than service account workarounds.
How much does it cost to run AI agents in production?
It depends on model, token volume and compute. We model it before deployment and set budgets and alerts so the bill is a decision.
Do you work with organizations in the United States?
Yes. Remote from Calgary, Alberta across Pacific, Mountain, Central and Eastern time, with on-site days by arrangement.
Related services
Azure consulting — the Azure AI Foundry, identity and network foundation under your agents
Microsoft Intune consulting — AI app governance and protected apps on managed devices
OpenClaw enterprise deployment — hardened, self-hosted agents on Windows 365 Cloud PCs