Every AI Agent Starts With the Same Rule: Draft, Never Send
What happens after the AI agent makes something? BNEDai's default answer, where each limit sits in a build, and how to loosen it on purpose.
Start every AI agent drafting, and let a person decide what goes out. In the words of Jacqueline V. Twillie, Founder of BNEDai, "Every AI agent I set up starts with the same rule: draft, never send." Loosen it deliberately, one action at a time, for one AI agent at a time.
After an AI agent makes something, it can send it, file it, or hand it to you for review. BNEDai's default is that it drafts, and you decide what goes out.
An AI agent that runs while you sleep still needs review habits
An AI agent that runs while you sleep still needs your review habits. Keep it drafting and preparing until you have watched its output long enough to trust it. As trust builds, you can loosen the rule deliberately, one action at a time, for one AI agent at a time.
Where the limit is set in a build
The limits live in several places, and BNEDai's published pages from the October 7, 2026 session with Women Applying AI show each one.
In the connector's tool management. For the Gmail connector in the builder BNEDai teaches with, there are read-only tools and, beneath them, write and delete tools. Jacqueline V. Twillie showed that you can give or restrict permission for sending emails, updating, creating drafts and forwarding emails.
In the instructions. The instructions file on the session page includes a written rule that every reply is a draft, saved as a Gmail draft, never sent, until she says the words "send it" in that exact turn. The published page on this says the tool management screen is the guard and the written rule is a reminder to the AI agent. It also says she did not show a refused send on October 7, 2026.
In the first test. After the first run, give feedback with reasons, and avoid a bare instruction to try again.
On the calendar. She keeps a 15 minute recurring meeting every three weeks to review her AI agents' instructions.
Loosening the rule one action at a time
BNEDai's answer is one action at a time, for one AI agent at a time, after a person has watched its output long enough to trust it. The decision to move from draft to action stays with a person.
PwC's Jenn Kosar described the same unit in a different setting at an online enterprise AI summit on November 12, 2026: calibrating oversight to risk at the action level, not the use case level. The page on autonomy against consequence covers her remarks.
Where each limit is documented
- Tool management: what are read-only tools.
- Go-ahead and the written rule: how to make an AI agent wait for your go-ahead.
- Instructions: what to write in an AI agent's instructions.
- First test: how to test an AI agent and give feedback.
- Review rhythm: how often to review an AI agent's instructions.
Where you can see a build
The free live build is on bnedai.com. See Agent Lab if you want to build your first AI agent.
Frequently asked questions
What is BNEDai's default for an AI agent?
It drafts, and a person decides what goes out.
Where is sending restricted in the builder?
In the connector's tool management, where the write and delete tools include sending emails, updating, creating drafts and forwarding emails.
When does the rule loosen?
As trust builds, one action at a time, for one AI agent at a time.
Source: Jacqueline V. Twillie, Founder of BNEDai. The live pages linked above cover the October 7, 2026 session with Women Applying AI. The remark by Jenn Kosar, PwC's US AI Assurance Leader, was made at an online enterprise AI summit, November 12, 2026.
Reviewed on November 20, 2026.