Blog
Insights on AI, security, software architecture, and building what's next for ambitious businesses.
26 articles in How-to Guide · Page 1 of 2
How to Measure an AI Operations Pilot Across Multiple Sites
A practical, site-by-site method for measuring AI answer accuracy, booking completion, corrections and staff time before deciding whether to expand.
Astra vs Sonnet 5.5 vs Opus 5.5: How to Run Your Own Comparison
A reproducible protocol for comparing Astra, Sonnet 5.5 and Opus 5.5 using your own tasks, measured results and a versioned decision record.
Measuring AI-Search Referrals Without Mixing Product and Services Funnels
A practical method for separating SearchFIT product acquisition from PADISO services leads, with a worked measurement model, query, and accuracy tests.
Booking Deposits and Refunds: Keep Payment Authority Explicit
A practical workflow for booking deposits and refunds that keeps payment authority explicit, ties approval to exact amounts, and verifies outcomes.
Benchmarking Browser Agents: Check the End State, Not the Transcript
Benchmarking Browser Agents: Check the End State, Not the Transcript. Practical examples, tradeoffs and implementation guidance for technology leaders.
Browser Agents and SSO: Designing Sessions Without Sharing Passwords
A practical guide to using SSO with browser agents: scope sessions, delegate access without sharing passwords, and verify that each task ends safely.
HQ Rules and Local Overrides: Governance for Franchise AI
A practical workflow for setting franchise-wide AI rules, reviewing local exceptions to policy or pricing, and verifying each approved change.
Tenant Isolation for Enterprise Agents on Azure
A practical Azure guide to separating tenant data, agent state and execution, with an architecture, failure paths and a printable isolation worksheet.
Model Routing for Enterprise Agents: Cheap First, Escalate on Evidence
Build a cheap-first agent router that escalates on observable failure, with a versioned decision table, migration test, and safeguards against blind retries.
Canary Releases for Agent Changes: Prompts Need Rollback Too
Canary agent changes safely by versioning prompts, tools, model settings and harness together—and routing, measuring and rolling back the whole release.
Automating Customer Follow-Ups Without Losing Context
A practical workflow for consent-aware reminders, win-back messages and human handoffs that preserves customer context and makes failures measurable.
Scaling Agent Workloads on AWS: Concurrency Is Only One Constraint
A practical AWS guide to queue design, quota-aware concurrency, downstream capacity, graceful degradation and recovery for agent workloads.
Scaling Foundry Agents: Plan for Quotas, Queues and Backpressure
A practical method for sizing Foundry agent throughput, setting concurrency limits, and using queues and backpressure to protect users when demand spikes.
An AI Business Case That Includes Integration and Ongoing Operations
An AI Business Case That Includes Integration and Ongoing Operations. Practical examples, tradeoffs and implementation guidance for technology leaders.
AI for Sick-Day Roster Gaps: Where Managers Stay in Control
A practical workflow for using AI to suggest sick-day roster cover while managers retain control of eligibility, approval, and the final roster change.
Build a Private Agent Evaluation Set: Sampling and Holdout Design
A practical protocol for sampling real agent work, separating development from holdout cases, and redacting sensitive data without erasing task difficulty.
A Shared MCP Tool Layer in Foundry: Ownership, Versions and Access
A practical guide to governing a shared Microsoft Foundry MCP tool layer: define ownership, version contracts, control access and manage failures.
One Inbox, Many Locations: Keeping Customer Conversations Consistent
A practical guide to routing multi-location conversations with clear ownership, escalation deadlines, exception handling and measurable acceptance criteria.
Identity for AWS Agents: Delegation Without Shared Credentials
A practical AWS design for separating an agent’s workload identity from the user’s authority, with steps for scoped access, delegation and failure handling.
Entra Identity for AI Agents: Who Is Allowed to Do What?
Separate the person asking from the agent acting. A practical Entra identity design ties each downstream permission to the right principal and task.