A story about spam, artificial intelligence, two email addresses, several Amy’s, a Teresa, an imaginary horse and the remarkable persistence of international electronic component sales.
Like most people who have owned an email address for more than approximately twelve minutes, I receive unsolicited commercial email.
A lot of it even!
This month alone, September 2026, I count over 50 in my mailbox.
SEO consultants promising to improve websites that no longer exist. Marketing agencies offering to increase conversions for businesses that have no customers, scammers, phishing emails… and more.
And, occasionally, electronic component distributors asking whether I have any purchasing requirements.
Usually, these messages go straight into the spam folder.
One particular email caught my attention.
It came from someone called Amy. Or possibly Teresa. The display name said Amy, the sending address said Teresa, and the Reply-To address said Amy again.
This is already getting confusing, right?
Ehhh what does her message say?, a White Horse Test? Wait.. Whaat?!
Okay, then an interesting question popped up. What would happen if I let ChatGPT write an answer?
After some iterations I sent back:
Components requested
- LMNOP-1234-LOL, automotive grade
- SN74WTF420N
- LM2940CT-5.0, LaserQuest LQX proportions, preferably 1998 specification
- NE555 timers, preferably manufactured on a Tuesday
- Wireless cable
- Left-handed RJ45 connectors
- HSDU, must support searching pack
- Replenisher, preferably small
- HQ, red/green, four-sided
This is a mix of imaginary and real components. LMNOP-1234-LOL
Yeah. That one is obvious.
Both some components actually exist. The requested specifications, however, are rather… specific.
- An LM2940 in LaserQuest LQX proportions, preferably 1998 specification?
- An NE555 that was manufactured on a Tuesday?
Good luck with that!
And then there is the perfect one!:
3 metres of wireless cable.
See the problem with that one?
Take your time!
Any apprentice engineer has probably been sent out at least once to find something equally useful: a square drill bit, a bucket of compressed air, a left-handed screwdriver, a long weight, striped paint, a glass hammer, a metric adjustable wrench, replacement bubbles for a spirit level, or a box of wireless IP packets.
It is basically an engineering rite of passage.
“Go to stores and ask them for three metres of wireless cable.”
And somewhere, an apprentice is going to spend twenty minutes looking for it before the person behind the counter starts laughing.
Insert evil laugh here.
And the rest of the specifications are IMHO even funnier:
All components had to be RoHS compliant, REACH compliant, gluten-free, compatible with Windows 3.11 for Workgroups and free from previous exposure to country music.
It is partially real and partially bait!
And I sent the mail thinking Amy will never respond ever again.
I was wrong!
Two days later I received two emails, yes two!
- amy@oneriverelec.com, RE: Fw:Re: rfq
- amy@onerivertech.com, Re:Re: rfq
Ehhh two mails? Two Amys?
And both Amys actually made a quotation.
Not just a generic “please send us the correct part numbers” response.
An actual component quotation.
- The real LM2940CT-5.0? Quoted.
- The real NE555? Quoted.
- LMNOP-1234-LOL? Unavailable. ( no shit Sherlock! )
- SN74WTF420N? Unavailable. ( D0h! )
- Three metres of wireless cable? Unavailable.
Of all the disappointing news in the quotation, this one hurt the most. The entire project had been designed around exactly three metres of wireless cable.
Someone had apparently taken my list seriously enough to check which components could actually be sourced.
This was the exact moment I realised:
Oh no. She is actually seriously making a quotation!
And things were about to get considerably worse.
Time to respond to both of them and ask ChatGPT to write a really long and eloquent answer!
The signature.
Best regards, M. On behalf of H. Okkiebokkiestein Department of Advanced Component Procurement Wireless Infrastructure & Legacy Laser Systems Division Schubbekutteveen, The Netherlands
Schubbekutteveen is a typically Dutch-sounding, entirely fictional rural place.
More English rural places would be:
Shaggington-on-Mire
Fucklebury-on-Bog
Cockswoggle Marsh
Again I did not expect Amy to respond anymore.
I was wrong, again!
She confirmed that she is the same Amy behind both email addresses.
Okay?! Why respond from two different addresses?
And again.. That White Horse test?!
So! ChatGPT! Warm up your engines!
Amy responds… Again!
Okay?! Now we have to pay a test fee for the White Horse Test?
Lets escalate this to an imaginary CEO!
Mind you, I’m still just responding from the same e-mail address!
And Amy responds!… Again!!!!
Wait.
First the White Horse Test was possible before shipment.
Then the White Horse Test required a fee.
And now…
the fee for the White Horse Test isn’t actually known?!
And she doesn’t understand all of this?!
No shit!
This is getting more and more confusing by the minute!
Again I sent a really long e-mail
And now!? Time for tumbleweed
Amy has not responded… Yet!
Use of AI
Could AI actually make cold-email spam economically useless?
Think about the economics of unsolicited sales email for a moment.
Sending email is cheap.
Really cheap.
A company can send thousands of messages such as the one that started this story. Most recipients will delete them. Some will be filtered automatically. A tiny percentage may respond.
And that’s enough.
The sender spends almost no additional effort on the thousands of people who aren’t interested.
The cost is asymmetric.
They automate sending.
We manually delete.
But generative AI introduces an interesting possibility.
What if we reverse that asymmetry?
In this case, writing a response asking about three metres of wireless cable cost me almost nothing. ChatGPT generated most of it and I had a good laugh reading the responses.
Amy, however, apparently did real work.
Someone searched for: LMNOP-1234-LOL
Someone searched for: SN74WTF420N
Someone apparently searched for: NE15555
But what if enough people did this?
Suddenly, the asymmetry might start to flip.
Basically:
Spam arrives ↓ AI classifies it ↓ Is it phishing/malware? ├── YES → Do not engage │ └── NO ↓ Is it unsolicited human-operated sales? ├── NO → Ignore │ └── YES ↓ Generate plausible but ridiculous RFQ ↓ Human salesperson responds ↓ Add another procurement requirement ↓ Human salesperson responds ↓ Create another department ↓ Repeat
Obviously, this should never blindly reply to everything. Phishing, malware and fully automated junk should be ignored. The interesting category is human-operated unsolicited sales: the kind where somebody on the other side actually starts spending time on your “lead”.
Because that is what happened here.
And the biggest question remains.
Will Amy respond?