ARTICLES
Technology and business,
from hands-on practice.
Articles on AI agents, business systems, financial AI, and e-commerce.
We write from the perspective of designing, building, and operating technology in real business contexts.
CATEGORIES
LATEST ARTICLES
1Password for Agents: Password Protection and Approvals After Sign-In
Approving an agent's sign-in is not the same as approving its subsequent actions or use of data. This article uses 1Password for Agents to explore that distinction.
Long-Running Notion Agents: Designing Mid-Task Checks, Stops, and Handoffs
Even when an agent can work on a task for many hours, how you check its work depends not only on run time but on the information it handles, the actions it is allowed to take, and the impact of mistakes. We look at what to decide before the task starts.
Handling Requests from Customers' AI Agents: Identity, Authority, and Conditions for Action
Distinguishing between people and AI agents is not enough. Using refund and order-change requests as examples, this article examines identity checks, delegated authority, and the requirements for carrying out an action.
What Notion 3.8 Means for Keeping Agent Context and Procedures Current
The easier it becomes to carry business knowledge across agents, the more it matters which content is authoritative and which version is actually in use.
The FSB's Final Report on 12 Sound Practices: A Materiality and Risk Perspective
Materiality and risk assessments help institutions tailor controls to each AI use case. We examine how Sound Practice 5 connects those assessments to documentation, validation and human oversight.
AgentCore payments Spend Limits: Separating Who Sets a Budget from Who Spends It
Being able to pay within a budget is not the same as the spending being necessary for the business. We look at whether limits should apply per payment, per session, per task, or over a period, who sets them, and how to keep the process that makes payments from expanding them.
Gemini Agent Audit Trails: Ownership and Authority to Resume Work
Recording the agent as the actor makes its actions easier to trace. We look at how to connect those records to request and approval records and to an operational owner, and how to decide what to check before resuming, based on why the agent stopped.
Clef and the Decisions API: Choosing Thresholds for Auto-Routing
If you use probabilities to separate automated processing from human review, first define the errors you can accept and your business requirements, then use your own data to select thresholds that meet those requirements. An example routing table is included.
OpenAI dots' Four Action Rules: When to Ask and Who Acts
The four custom-rule options in OpenAI dots are a starting point for defining approval requirements and who performs each action, aligning them with connected-app permissions, and reviewing them after setup.