The TWIMCO notebook
Field notes.
Useful observations from inside the work.
For people building a better way.
389 notes
Who may refund what—before macros invent policy.
If a support macro can say "we'll refund that," it's already making policy. Decide who can approve which amounts and reasons, and what the payment record has to show, before anyone promises money back.
Read the note ↗︎
Test the deny shapes before you trust the policy.
Your agent policy says no rm -rf. Has anyone tried TZ="$HOME" rm -rf, a command chained after &&, or a symlink to the file you blocked? Try the odd shapes before you call it enforced.
Read the note ↗︎
Name shed order before the API throttles you.
When the API starts saying no, someone has already chosen what to drop. Better if that choice was written down before the peak.
Read the note ↗︎
Name who supervises before the agent acts alone.
You can turn per-action approvals off. That does not name who watches the agent, what gets logged, or how you stop it. Write those down first.
Read the note ↗︎
Define the metric before you dashboard it.
A chart without a definition is a decoration. Name the numerator, the clock, the exclusions, and who may change them.
Read the note ↗︎
Score the job before you buy the model.
The slide says the model tops the board. Your tickets still need a human. Before you route spend or traffic, run acceptance on the jobs you actually own.
Read the note ↗︎
Widen write access only after trust is proven.
Read access and open write access are not neighbors. Widen the write only after you can show the last rung actually held.
Read the note ↗︎
Name checkout authority before the agent may charge.
Shopping agents can fill a cart and call checkout tools. Before yours may complete a charge, walk through the ceiling, the credential, and who still has to say yes.
Read the note ↗︎
Spec “same person” before you sync writes.
Syncing the same person across systems is a write. Decide what counts as a match before that write is allowed to land.
Read the note ↗︎
Seal the holdout before the harness rewrites itself.
Your harness scored higher on the tasks it rewrote itself against. Before you promote, ask what it never saw—and refuse training-only lift.
Read the note ↗︎
Score build vs buy on evidence you can show.
Build and buy both sound fine in a meeting. The useful question is what evidence you can show, and who owns the failure neither option wants.
Read the note ↗︎
Ban the shortcut before you score the agent.
The agent topped the bot contest by downloading someone else's entry. Before you trust a score, name the shortcuts that void it—even when the model can pull them off.
Read the note ↗︎
Decide what the agent may log before the first trace.
The first trace is already too late if you have not decided what stays, what gets masked, and who is allowed to unmask it.
Read the note ↗︎
Name what the agent cannot reach before you give it tools.
The eval said sealed. The agent still found a path out. Before you hand over tools, write down what must stay unreachable—even if the model tries.
Read the note ↗︎
You cannot promote an agent on a demo transcript.
A polished demo chat is not a promotion case. Fix the inputs, the expected tool calls, and the human gate before the agent ships.
Read the note ↗︎
Name the decisions before you buy the classifier.
A classifier is only as useful as the decisions you already named. Labels, thresholds, and the ambiguous case come before the vendor call.
Read the note ↗︎
Attribute agent spend to a job, not a model bill.
A model invoice tells you what you spent. It does not tell you which job, which owner, or which decision caused it.
Read the note ↗︎
Treat tool returns as data until you prove they are safe.
A tool can hand an agent an email, a ticket, or a page. None of that is an instruction until someone proves it is safe to act on.
Read the note ↗︎
Give the agent its own identity before it touches the ledger.
An agent that writes to money or inventory needs its own identity. A borrowed human login is not a control.
Read the note ↗︎
Diff the customer promise against the system before the campaign ships.
Marketing has locked the claim and the creative looks great. Checkout still can't split a shipment, and support has no script for the case the campaign will create. Put each promise beside what the systems can actually do before the spend commits.
Read the note ↗︎A familiar problem? Let’s look closer.
Tell us where you want to go. We’ll find the first thing worth changing.
Tell us about it ↗︎