Waiting for you

The list on the Agents page of what your coding agents need from a person: split tests to start, KPIs to create, and verdicts nobody acted on.

Waiting for you is a short list on the Agents page. It shows what your coding agents need from a person before they can go on.

Think of waiters and a manager

Waiters take orders all evening. Most go straight to the kitchen. But some need the manager: a big table, a special request. Those wait on the manager's desk until someone looks.

Your agents are the waiters. Most impact checks they create start measuring on their own. The ones that need a person wait in Waiting for you. You are the manager.

What can be in the list

• Split tests to start. “… asks to start …”. The agent's split tests don't start on their own: its setting is Always ask me, or it's set to start on its own only with a pull request and this change came without one.
• Releases waiting for their KPI. “… measures … once the KPI … is published”. The agent wants a before vs after or adoption impact check on a KPI Needlr has never counted.
• Proposed KPIs. “… proposes the KPI …”, with what it counts and sometimes why.
• Verdicts nobody decided on. “2 verdicts have waited more than a week for an agent's decision.”

When nothing waits, the list is hidden. Only Owners and Editors see it.

Start a split test an agent asked for

1. 1Open Agents in the menu on the left.
2. 2Under Waiting for you, press Review and start next to the test. Its check page opens, with Requested by and the agent's name.
3. 3Read what version B does and which KPI it should move. Want a different KPI? Press Edit KPI before it starts.
4. 4Press Start, then Start now. Needlr asks for a code from your authenticator app if you haven't entered one in the last 20 minutes. Your Tag Manager must be connected.
5. 5Needlr publishes anything not yet live, then starts measuring. The agent sees progress and a date, never which way it leans.

Say no to a test

There's no decline button in the list. Press Review and start to open the test, then press Delete. It hasn't started, so no data is lost, and it's gone from the agent's list too.

A release waiting for its KPI

The agent shipped something and wants to know if it helped. But the KPI it picked has never been counted on your site. Usually it's a new KPI whose tag isn't published yet.

Press Review to open it, then Start. Needlr publishes the KPI's tag to your Tag Manager first, then starts measuring. Nothing starts it for you: until someone presses Start, it keeps waiting.

Proposed KPIs

An agent can't create a KPI. When none of yours fits its change, it proposes one: a name, and what it counts, like “A visitor clicks a button or link whose text is Start trial”.

• Create KPI adds it to the project. If the project already has a KPI that counts exactly the same thing, Needlr uses that one instead of making a copy.
• Like any new KPI, its tag then waits to be published. A bar on the project page says so: press Publish now (more).
• Dismiss turns it down. Nothing is added.

Verdicts nobody decided on

After a verdict, your agent should record what it did next: keep, roll back, iterate, give up, or fix the measuring. When one of its impact checks has had a verdict for more than 7 days with no decision, it's counted here.

There's no button for this one: the agent records its decisions. Ask it to look at its pending impact checks and decide on each. Agents set up from the Agents page are told to do this at the start of every session.

Common mistakes

• Leaving tests waiting. A test that hasn't started measures nothing, and the agent can't learn from it. Look at the list once a week.
• Creating every KPI an agent proposes. Ask yourself: does the business care about this result? If not, press Dismiss.
• Starting without reading version B. You're the review step. Read what the change does before you press Start.