Field Notes
You do not believe everything you see on the internet.
So why would you believe everything that comes out of AI?
I asked that question to a room of entrepreneurial and leadership students at the University of Montevallo, my undergraduate university.
I was back on campus talking about business, taking an unconventional path, the future of work, AI, and what they can do today to win in the future.
I was not talking to a room full of people who had the whole map figured out.
I was talking to students trying to understand the world they are walking into.
That question landed because everybody already knows the internet is messy.
Some of it is true.
Some of it is wrong.
Some of it is old.
Some of it is confidently presented by people who have no idea what they are talking about.
AI pulls from that world.
So if you would not blindly trust the source material, you cannot blindly trust the output.
That is where my last post comes back in.
Last time, I wrote about lanes.
Builder. Operator. Customer-facing. Somewhere in the messy middle.
The point was not to turn AI into an identity test. It was to give people a map. Some people are going to build systems. Some people are going to run the business closer to the customer. Some people are going to translate.
But every lane has the same rule underneath it.
Trust but verify.
A builder can now ship the wrong thing faster.
An operator can now trust a polished summary too quickly.
A customer-facing person can send a confident answer that came from stale context.
The person in the middle can accidentally carry a bad assumption from the business side into the system side, then watch the system multiply it.
Speed is not the enemy.
Unverified speed is.
That was the real point underneath my last post. If AI is compressing handoffs, changing company roles, and letting more people build things that used to require technical teams, then verification cannot stay hidden inside one department.
It has to become a normal operating habit.
At Montevallo, a student asked me what one piece of advice I would give someone younger than me.
My answer was simple:
"Trust but verify."
Then I kept going.
Never take anything at face value. Even what I said in that room. Go find it for yourself. There is no reason you cannot go to the source now. The difference between you getting what you want out of life and not is directly tied to how much truth you know. The only way you know truth is by going to the source.
You can only verify what you actually understand. Or at least what you have enough knowledge to question.
That is why people become more important in a world of AI, not less.
AI can produce the draft. It can build the first version. It can summarize the meeting. It can act like a genius intern.
But the genius intern still needs guidance, training, feedback, and someone who understands the work well enough to know when it is off.
That is where the job changes instead of disappearing.
Because AI did not create the problem of people trusting bad information. It just made the problem faster, cleaner, and easier to dress up.
The summary looks good.
The dashboard looks good.
The email draft sounds good.
The agent says it completed the task.
The problem is that "looks good" is not the same as true.
Every's "After Automation" makes a related point from a different angle. As AI automates more default output, the valuable work does not disappear. It changes. Judgment, taste, context, and expert review become more important because average output is suddenly everywhere.
That tracks with what I am seeing.
When anyone can create a draft, app, tracker, report, or workflow, the value shifts from "I can make the thing" to "I know whether the thing is right."
That is the operator's edge.
Not cynicism.
Discipline.
The new world rewards people who can move quickly without losing their grip on truth.
Playbook
Here is the practical version.
Take one workflow from your lane.
If you are closest to the customer, pick one answer, follow-up, recap, or recommendation that leaves your hands.
If you are closest to the system, pick one automation, dashboard, report, or agent output.
If you sit between both, pick the handoff where business context becomes system instructions.
Then ask six questions.
First, what is the source of truth?
Not "the AI said." Not "the dashboard showed." Not "someone told me." Where should the real answer live?
The CRM. The transcript. The contract. The property management system. The bank account. The live app. The customer message. The calendar. The source file.
Name it.
Second, did I check the source or a summary of the source?
Summaries are useful. I use them every day. But summaries are a layer above the thing. If the detail matters, open the thing.
Third, what would make this wrong?
This is the question most people skip. Maybe the data is stale. Maybe the agent read the wrong file. Maybe the customer changed their mind. Maybe the calendar event moved. Maybe the tool returned success but the public page never changed.
Say the failure mode out loud before you trust the output.
Fourth, can I test one before I trust many?
One email before the full send.
One record before the batch update.
One market before the scraper run.
One customer path before the workflow goes live.
This is boring until it saves you.
Fifth, what receipt comes back?
A link. A timestamp. A sent email. A log line. A screenshot. A live page. A record ID. Something you can inspect later.
If the system cannot produce a receipt, be careful with the claim.
Sixth, where should the system stop?
This connects back to my piece on AI lying when you do not give it an out.
If the agent cannot verify the answer, it should say so.
If the source is missing, it should say what source is missing.
If the data is stale, it should mark it stale.
"I do not know" is not failure.
It is how the system protects trust.
Orientation
This is the next layer after knowing your lane.
Builders need it because building faster raises the cost of being wrong.
Operators need it because customer trust gets expensive when the machine sounds confident.
Leaders need it because an AI-first company without verification becomes a faster rumor mill with nicer formatting.
I do not think the winners in this era will be the people who distrust every tool.
That gets exhausting fast.
I also do not think the winners will be the people who trust every output because it looks professional.
That gets dangerous fast.
The useful posture is in the middle.
Use the machine.
Move faster.
Then go to the source.
Next time, I want to get into Specified Intelligence. The idea that the smartest model is not always the best model. Sometimes the best intelligence is the one shaped for the exact job in front of it.
For now, pick one workflow from your lane and ask:
What am I trusting because it looks finished?
Comment below and tell me what you are verifying this week.
I read every one.
— Brian