A week ago I posted that Claude kept telling me the work was done when it was not. The post picked up 382 reactions and 130 comments, and nearly every comment was a person asking for the same thing: the file. That told me the frustration is common. The part people got wrong is why it happens. The model was not the problem. The problem was that I kept explaining what I wanted one message at a time, and by the next morning none of it existed anymore. What fixed it was eight rules written into a file the agent reads before it touches anything. That is the whole ai agents vs prompts question in practice, and it has almost nothing to do with which model you are paying for.
I run Growth Cab, a GTM advisory, and we put AI inside outbound, research and reporting for B2B teams selling contracts above fifty thousand dollars. So this is a working note from someone who has to live with the output, rather than a taxonomy.
in
“Claude kept saying the work was done. It wasn't. So I wrote 8 rules into the file it reads first.”
AI Agents vs Prompts: The Difference Is Where the Instructions Live
Most definitions of ai agents vs prompts talk about autonomy, tools and memory. Useful, but it misses the thing that changes your day. A prompt is an instruction you spend. You type it, you get output, and the instruction is gone. An agent is a thing that acts, and anything that acts repeatedly needs standing orders. The question is simply whether your instructions live in the message or in a file the system reads first.
Once you see it that way, a lot of frustration makes sense. If you find yourself typing the same correction for the third time this week, you do not have a model problem. You have an instruction that is living in the wrong place. Every correction you give in chat evaporates. Every correction you write into the file survives.
The Eight Rules I Actually Wrote
Here is the file, in plain language. Rule one: never say it is done without showing proof. Rule two: try to disprove your own work before submitting it. Rule three: every correction I give becomes a permanent rule. Rule four: get confirmation before anything irreversible. Rule five: do the reversible work without asking permission. Rule six: if you need something from me, say so upfront. Rule seven: turn what you learn into a reusable tool. Rule eight: work without interrupting my screen.
Read them again and you will notice what they have in common. Not one of them is about the task. They are all about how the work gets checked, when to interrupt me, and what happens to a mistake after it is made. That is the category of instruction that a prompt is bad at carrying, because you would have to repeat it every single time and you never will.
The first two rules did most of the work. An agent that has to show evidence before claiming completion stops producing the confident summary of a job it did not finish. In our stack that means the run does not report success on an outbound sequence until it has fetched the sent count back and shown it. It is boring. It is also the difference between a report and a guess.
Rule three is the one I would install first if I could only pick one. It turns a correction from an annoyance into an asset. You fix the same thing once, it goes into the file, and you never spend that minute again. Over a few months the file becomes the accumulated judgement of everyone who ever corrected the system, which is a thing a prompt can never be.
What AI Agents vs Prompts Changes on Revenue Work
Translate this to a GTM team and the stakes get concrete. Prompting an AI to write a cold email is fine. It is a single act with a human reading the output before it goes anywhere. Now let that same system send on your domain every morning and the calculation changes completely, because nobody is reading every message and the cost of a bad one is your sender reputation.
The rules we run on client work look like the eight above, translated. Never claim a campaign is live without pulling the live status back. Verify an email exists before it enters a sequence. Ask before anything that touches a real prospect for the first time. Do the research, the drafting and the list building without asking, because all of that is reversible and asking permission on reversible work is how you end up doing it yourself.
The effect is not that the AI got smarter. It is that the number of things I have to personally check dropped, because the checks moved into the system. That is the actual return, and it is a boring operational return rather than a magical one.
Where the AI Agents vs Prompts Framing Stops Being Useful
Three places, and I would rather say them than sell you a system.
The first is one-off work. If you are writing a single job description or summarising one document, a prompt is the correct tool and a rules file is overhead you will never earn back. Most of what most people do with AI is genuinely one-off, and there is nothing wrong with that.
The second is bad judgement. Rules constrain how work gets checked. They cannot rescue a task that was wrongly defined. If your outbound is aimed at the wrong accounts, an agent with eight rules will pursue the wrong accounts more reliably and with better documentation. The rules layer sits on top of strategy and never replaces it.
The third is maintenance. The file is a living thing. Ours has been rewritten dozens of times, and rules that contradict each other produce worse behaviour than having no file at all. If nobody owns it, it rots. Budget the fifteen minutes a week or do not start.
How to Write Your Own
Do not sit down and draft a policy document. It will be abstract and the agent will ignore it. Instead, keep a note open for one week and write down every correction you give. At the end of the week you will have somewhere between six and twelve lines, and they will be specific, because they came from real failures rather than from imagination.
Then paste them into the file your tool reads first. In Claude that is a CLAUDE.md. In other tools it is an AGENTS.md, a project instruction, or a system prompt you set once. The filename matters far less than the habit behind it, which is that a correction gets written down instead of repeated.
The honest summary of ai agents vs prompts is this. Prompts are for things you do once. Agents are for things you do every day, and the only way an agent stays useful over months is if the lessons it has already been taught outlive the conversation they were taught in. Eight lines in a file did more for my output than any model upgrade this year.
If you want one play like this every morning, I write The Revenue AI Brief, a short daily note on the AI and GTM systems we run inside Growth Cab. Drop your work email below and tomorrow's edition lands in your inbox. And if you think I am wrong about where the rules layer breaks, tell me on LinkedIn. I answer everything.
