AI Systems
Agent audit trail: reconstruct the action, approval, and outcome
If the only witness is the agent, you do not have an audit trail. You have a testimony.
Tag
Agent operations
How to run and retire production agents so credentials, triggers, and spend stop when the agent should stop.
AI Systems
Agent runtime budget: keep retries inside the same ceiling
A cap that resets on every retry is a cap per attempt, not per task.
Venture Building
An agent scorecard turns output review into a permission decision
One good-looking brief can hide an unsupported claim. Grade the work against the job it was given, then decide what the role may do next.
AI Systems
Agent shadow mode: evaluate proposals before granting action rights
Let the agent be wrong on paper first, where a mistake costs a comparison instead of a customer.
AI Systems
The Agent Trust Ladder: how an AI role earns autonomy
A scorecard records how the role performed. The ladder decides what it may do next. The action still sets the ceiling.
AI Systems
AI fake citations: a 15-minute check for the claims that matter
A citation makes two claims: this document exists, and it says this. Check both in the source itself, not by asking the tool.
AI Systems
Cost per route vs the monthly AI bill: meter where the money goes
A flat monthly bill can hide a route that doubled its spend. Attribute each charge, separate volume from cost per request, and give the route an owner.
AI Systems
Governed agent memory: what may persist, and who may write it
An agent's memory is a database that the agent writes itself. Give it the rules you would give any other database.
AI Systems
How many AI agents can one person supervise?
Starting another agent is a one-time decision. Reading its work, checking its handoffs and handling its exceptions come back every week.
AI Systems
When nobody approves in time, your AI agent must not act
An expired request records silence. A rejected request records a decision. Both must block the action.
AI Systems
Model registry vs vendor list: know which models run
A contract can clear a supplier while a production model remains unknown. Compare actual calls with the register before you trust it.
AI Systems
Privacy tier vs vendor promise: which data may reach a cloud LLM?
A vendor's privacy page describes the vendor. A privacy tier describes your data and decides which model hosts may receive it.
AI Systems
Prove you can stop an agent: interrupt the run and revoke access
An instruction to stop is a request. A stop is a control that someone outside the agent can operate.
AI Systems
How to retire a production agent (stopping it is not enough)
A stopped agent is not a retired agent. The next schedule, a copied key or a monthly plan can keep it working and billing.
Venture Building
What is a roster? From the Monday rota to a team of AI agents
The word started as a Dutch gridiron. The idea still fits your staff, your agencies and your AI agents.
Venture Building
When credible sources disagree: diagnose before choosing sides
Before you back the source with the bigger name, check whether the claims actually conflict. Then name the check that could help resolve whatever disagreement remains.
AI Systems
TRACE vs RAG: one fetches the text, the other checks what it proves
An answer can quote the right document and still draw the wrong conclusion.
Venture Building
Trigger map vs cron job: a schedule is one row, not the map
A weekly brief can start at nine. A client email, an arriving draft and a repeated form submission still need routes of their own.
AI Systems
What is ROSTER? How one person manages a team of AI agents
A role is defined by what it may not do. Six written decisions turn an AI agent into a specialist you can grade, fix or retire.
AI Systems
What is TRACE? Five steps to check research before you act on it
Three links can be one source. TRACE is how you find out before the decision does.
AI Systems
When not to use an AI agent: five signs, and what to use instead
An agent earns its cost only when the next step depends on what it finds. Most tasks never get there.
AI Systems
Why your RAG answer is wrong: find the broken step before you change the model
The model is the last thing to blame, not the first.