In partnership with
Hi everyone,
On August 14 I switched off almost every AI agent running my company.
Every one of them was working fine. I still could not tell you what any of them was responsible for.
Last week I started turning them back on. One at a time, with a written brief for each. The restart taught me more than the original build did.
The job that reported healthy for 20 days
The first agent back on is called Lens. Its job is to grade our marketing every Friday and post the result.
Before I let it run, I checked what it would actually read. The database behind it had stopped being updated on August 12. Twenty days of nothing.
Lens would have run fine. It would have pulled those old rows and written a clean summary. Then posted it Friday as "this week." Nobody would have caught it. The data was there. It was just old.
Then I found the worse one. The system that runs these jobs has a skip switch. A skipped job still reports success. So a dead job looked healthy on every dashboard I had.
Then the opposite problem. Nine agents were still switched off on purpose. The alarm system did not know that. Its first run would have fired nine alerts about agents that were fine.
That is how an alarm dies. It cries wolf on day one and everyone learns to ignore it.
Four problems in all, every one found before a single agent posted anything.
Tasks have a finish line. Roles have responsibility.
Here is what I got wrong the first time.
I gave those agents memory, tools and a schedule. I thought that was automation.
It was not. Those three things without a defined job do not automate work. They let confusion run on a schedule.
The difference is task versus role.
A task needs instructions. A role needs an operating system.
The Future of AI in Marketing. Your Shortcut to Smarter, Faster Marketing.
Unlock a focused set of AI strategies built to streamline your work and maximize impact. This guide delivers the practical tactics and tools marketers need to start seeing results right away:
7 high-impact AI strategies to accelerate your marketing performance
Practical use cases for content creation, lead gen, and personalization
Expert insights into how top marketers are using AI today
A framework to evaluate and implement AI tools efficiently
Stay ahead of the curve with these top strategies AI helped develop for marketers, built for real-world results.
The six things to write down first
I now score every job before I build anything for it. One point each:
Can I name the deliverable, who reads it, how often, and what decision it drives?
Do I know which systems are the real source, and can the agent reach them?
Has a human done this the same way at least three times?
Can I list what it may do and what it must never do?
Could a reviewer judge the output in five minutes?
Do I know what "stop and ask" looks like here?
Save this one. It is the whole decision in six lines, adapted from Sunil Ramlochan's work-design framework.
Six out of six, build it. Four or five, build it and watch the gaps. Three or below, do not build an agent. Write the process down instead. That step is usually where the real value is anyway.
Run Lens through that list and it falls over on number two. I knew which database it read. I never asked whether that database was still being fed.
What a brief looks like
One page per agent, written before anything gets built. Five parts:
Result. What it owns. "Post a source-linked risk report every Monday for the CS lead" creates accountability. "Help with customer success" creates activity.
Sources. Which systems count as the real source, in order, and which one wins when they disagree. One rule matters most here. Preferences live in memory. Facts live in source systems. An agent that remembers last month and treats it as today. That is the quietest way this fails.
Limits. A yes list and a no list. "Be careful" is not a boundary. Our newsletter agent has a no list. On it: "send a newsletter" and "use a statistic that does not trace to a source."
Evidence. How you judge the work. A fluent answer can still be wrong or three weeks stale. So every role ships with a standard. Sources and dates for research. An action log for operations. Claims mapped to sources for content.
Exceptions. What makes it stop, who hears about it, and what it hands over when it does. "I found nothing" is a real answer. Padding it is a failure.
Ours live in the same repo as the code they run, so the brief and the agent never drift apart.
That is it. One page. The page outlives whatever model or vendor runs it, which is the part people miss.
One thing to try this week
Pick the one job you keep asking AI to redo. Not the exciting one. The repetitive one.
Then try to write those five parts for it. Most people cannot finish the sources section, and that is the finding. The hidden knowledge in that job lives in somebody's head. Which numbers they trust. Which exceptions they catch. When they stop and ask a human.
Writing it down is most of the work. The agent is the easy part.
Where they are now
Those nine agents went live on trial this week. On Monday I go through them one at a time, against the standard written into each brief. Sign it off, or pull its schedule and send it back a step.
I will judge one thing above the rest. Can I check the output in five minutes. That is the fifth of the six questions, and it is the one people skip.
Next week I will tell you the split. How many stayed on, how many went back, and why.
The framework behind all this came from Sunil Ramlochan's piece on work design. Worth your time if this landed.
What is the job in your business that everyone assumes someone else owns? Hit reply and tell me. I read every reply, and our conversations help shape the newsletter.
Thanks for reading.
Tim