How do I stop random AI requests all week?

TL;DR

Most AI ideas die in the handoff, not the build. Fix that with a simple intake and triage lane: one place to drop every AI request, a weekly review that ranks them, and each survivor gets an owner, a test, and a kill switch. A kill switch is a rule that says when you shut the thing off. This turns a week of drive-by asks into one ranked list you can actually work.

• Every AI request lands in one intake spot, not your inbox, hallway, or head.
• Once a week, you rank the list. Most items get parked or killed.
• Each live item needs three things: an owner, a test, and a kill switch.
• The owner is a person, not a team. The test is one number that says pass or fail.
• No test means no start. You can't grade a request that never said what winning looks like.

Why do AI ideas fall apart between the idea and the doing?

AI ideas fall apart in the handoff because nobody owns the middle. Someone has a request. Someone else could build it. Between them sits a gap where the request loses its shape, its deadline, and its point. The idea doesn't fail on the merits. It fails because it never got assigned.

You've seen this. A manager reads about AI over the weekend. Monday they want the chatbot. By Thursday three other people want three other things.

None of it is written down. None of it has a due date. It lives in Slack messages and hallway comments, and it evaporates by Friday.

The problem is not a shortage of ideas. It's a shortage of a lane the ideas travel through. Build the lane and the ideas stop leaking.

What is an AI intake and triage lane?

An AI intake and triage lane is a single form plus a weekly review. Intake means every AI request goes to one place in the same format. Triage means once a week you sort that list, rank it, and kill most of it. The lane replaces a dozen scattered asks with one ranked list somebody actually owns.

Intake is boring on purpose. A shared form with five fields beats a clever tool nobody fills out.

Ask for the problem, not the solution. "We spend two hours a week retyping orders" is a request. "We need an AI agent" is a wish. An AI agent is software that takes a few steps on its own toward a goal you set.

Triage is where the discipline lives. You rank by two things: how much time it saves and how easy it is to test. Everything else waits.

What should the intake form actually ask?

The intake form should ask five plain questions and nothing else. Long forms don't get filled out. You want the problem in the requester's own words, the time it costs today, who feels the pain, what a win looks like, and a rough guess at how hard it is. Five fields. One screen.

Notice the form bans tool names. That's deliberate. Pick the workflow first. The tool comes after.

If a request can't name the time it costs today, it goes to the bottom. You can't rank a problem nobody measured.

How do you run the weekly triage without a meeting that runs long?

You run the weekly triage in thirty minutes with a simple rule: park or kill by default, promote only what earns it. Most requests will not survive the first read, and that's the point. A short list you finish beats a long list you admire. Do it the same time every week so it becomes a habit, not an event.

1. Pull the new intake items into one list.
2. Cut anything with no time cost and no clear win. Kill it now.
3. Score what's left: high time saved plus easy to test goes to the top.
4. Pick the top one or two. Not five. One or two.
5. Give each survivor an owner, a test, and a kill switch before anyone builds.

The owner is a named person. "Marketing" is not an owner. Dana is an owner.

The test is one number that settles the argument. If the retyping request was two hours a week, the test is simple: does it drop below thirty minutes.

The kill switch is the date or number that ends it. "If we're not under thirty minutes by week four, we stop." Write that down before you start, not after it flops.

Why does every live project need a kill switch?

Every live AI project needs a kill switch because stopping is harder than starting. A kill switch is a rule you set upfront that says when you pull the plug. Without one, a project that isn't working just keeps living on hope and sunk cost. With one, quitting is a plan, not a defeat.

Kill switches show up in serious settings for a reason. Consultant Yilmaz Ozmen has written about pairing AI stage-gates with a kill switch, so a tool that stops passing its checks gets stopped, not shipped.

You don't need their scale to borrow the idea. You need one line in your triage notes.

Say the number and the date out loud. "Two hours to thirty minutes by August 24, or we kill it." Now the test grades itself and nobody has to play villain.

How does this connect to the tools you already run?

This connects to your existing tools through the same plumbing the big players are wiring up. CCG Catalyst has noted that AI agents are arriving inside the platforms banks already use, and that an agent's reach depends on its connectors. Connectors are the wiring that lets an agent read and write in your other systems. Your triage lane decides which of those connections is worth turning on.

The lesson scales down cleanly. Agents don't matter until they're plugged into your actual data.

Your triage test tells you whether a connection earns its keep. If wiring the agent into your order system doesn't drop the two hours, the kill switch fires.

Government teams run this same discipline at scale. U.S. Customs and Border Protection publishes each AI use case with a stated purpose, from screening cargo to checking identities, so every tool has to say what it's for. Your intake form is the small-business version of that: no stated purpose, no start.

Key Takeaways

• Most AI efficiency projects fail in the handoff between the idea and the doing, not in the build.
• An AI intake and triage lane is one request form plus a weekly review that ranks and kills.
• The intake form should ask for the problem and its time cost, never a tool name.
• Every promoted project needs a named owner, one test number, and a kill switch set before work starts.
• A kill switch is a rule set upfront stating the date or number that ends the project.
• Rank requests by time saved and how easy they are to test; park or kill everything else.

FAQ

How do I stop random AI requests coming at me all week?

Send every AI request to one intake form instead of your inbox and hallway. Then review the whole list once a week, rank it, and kill most of it. The random asks don't stop coming, but they stop landing on you one at a time. They pile in one place and wait for the weekly review, where you handle them as a ranked batch.

What is a kill switch for an AI project?

A kill switch for an AI project is a rule you set before you start that says when you shut it off. It's a date or a number: "if we're not under thirty minutes by week four, we stop." The kill switch turns quitting into a plan instead of an argument. It also forces you to name what success looks like on day one.

How often should I review AI requests?

Review AI requests once a week, at the same time, for about thirty minutes. Weekly is often enough to stay responsive and rare enough to avoid a standing meeting nobody wants. A fixed slot makes it a habit. Cut anything without a clear time cost or win on sight, and promote only one or two items you can actually staff.

Who should own an AI project in a small business?

A named person should own each AI project, never a department. "Marketing owns it" means nobody owns it. Dana owns it means Dana answers for the test result on the deadline. The owner doesn't have to build the thing. They have to make sure it gets tested and either passes or gets killed on schedule.

What if a request doesn't have a clear number to test?

If a request has no clear number to test, send it back or park it at the bottom. You can't grade a project that never said what winning looks like. Ask the requester for one measurable thing: hours saved, errors dropped, replies sent. A request that can't name a number isn't ready. It's a feeling, and feelings don't survive triage.

Sources

• U.S. Customs and Border Protection AI Use Cases: https://www.dhs.gov/ai/use-case-inventory/cbp
• Yilmaz Ozmen, "AI, Kill-Switches, and Living Dashboards": https://www.linkedin.com/pulse/ai-kill-switches-living-dashboards-yilmaz-ozmen-jdqyc
• CCG Catalyst, "Sector Spotlight: AI Agents and Connectors for Banks and Credit Unions": https://www.ccgcatalyst.com/thought-leadership/research-snapshot/sector-spotlight-ai-agents-and-connectors-for-banks-and-credit-unions