Hi everyone,
Third issue of Your Money, Your AI here. Two in, and I want to say thank you properly - several of you replied to the first one with your own money questions to stress-test, and I have kept every single one of them. That was the whole hope for this thing, so: thank you. Keep them coming.
One topic today - time. And specifically: how long it takes to find out you were behind, and what it costs to find out late.
It might be a long read. But I think it earns it. Make yourself a coffee 😉
At the beginning of my corporate career, there was something that took me far too long to find out.
For years, I had no idea what the men doing my job were being paid.
Nobody was hiding it on purpose.
It simply wasn't a question you asked, and if you did ask, the answer was that pay is confidential and let's talk about your development plan instead.
I found out eventually. Of course I did. That's the part that still gets me annoyed - not that the gap existed, but how long it took to become visible, and how completely useless the information was by the time I had it. I could not renegotiate my 2012 contract after all, could I?
That is the whole story of the gender pay gap, really. The realization always arrives late. One woman at a time, usually over a drink, usually far too late to do anything about it.
The law is only arriving now - fifty years later.
Equal pay has been the law in Europe since the 1970s. In theory.
Because yet across the EU women still earn meaningfully less, and until about ten minutes ago you had no legal right to find out by how much.
That's finally changing, and it's worth knowing the actual shape of it, because most people have heard "pay transparency is coming" and have no idea what it means for them.
The EU Pay Transparecy Directive (2023/970) had a hard deadline of 7 June 2026 for all 27 member states to write it into national law.
Surprise-surprise! Only four countries actually made it: Slovakia, Italy, Lithuania and Malta. Congrats if you live in one of those.
The other 23 missed. Spain, where I live, hasn't even published a draft. Sweden formally paused, asking to renegotiate the thing.
But the Directive is law regardless of whether your country got its paperwork done. What it should be giving you, broadly:
A salary or salary range before the interview, not after you've named a number first.
The right to ask your employer what people doing comparable work are paid, broken down by gender.
Gender pay gap reporting: employers with 250+ staff report annually, first reports due June 2027 on this year's data. 150–249 report every three years. 100–149 join in 2031.
If a job category shows a gap above 5% that can't be explained and isn't fixed within six months, the employer has to sit down with worker representatives and account for it.
I suggest you read those dates again. 2027, 2031… That is the speed at which this gets fixed.
50+ years from "equal pay is the law" to "and here is how you check."
That is a whole working life, for a lot of women. My mother's entire career. Possibly most of mine…
And I'm telling you this because the same story is starting again, in front of us, and this time we are early.
The second gap is being built right now
Two large employment-law firms - Lewis Silkin and Ius Laboris - both published pieces this summer asking whether unequal AI adoption becomes the next gender pay gap.
And their numbers are speaking for themselves, ladies.
PwC found 32% of women using generative AI at work, against 57% of men. McKinsey found that 21% of entry-level women said a manager had encouraged them to use AI - for men it was 33%.
And a Danish study of 18,000 workers found women 16 percentage points less likely to use ChatGPT, with the gap holding between a man and a woman doing the same job. Same role. Same desk. Same tasks.
That last one matters, because it kills the comfortable explanation of ‘women are just choosing different careers’. It's the same job, and a different level of encouragement to go and use the thing that's about to decide who gets the interesting work.
Which is the mechanism, by the way. Nobody gets promoted for using a chatbot. People get promoted for the work it lets them take on - the analysis that used to need a contractor, the project you put your hand up for because suddenly it's three hours and not three days. Access to AI is quietly becoming access to the visible work. And visible work is what gets priced at annual review time.
Three review cycles of that, and it stops being an adoption gap and becomes a number on a payslip. Which somebody will then spend 50 years trying to make visible.
What about the AI itself?
Take Wikipedia, a significant ingredient in almost every model: roughly 80% of active editors are men, and about 20% of English-language biographies are about women. Nobody sat down and decided that. It is simply what got written down. Mostly - by men.
So the model doesn't hold a view about women. It holds a very confident view about what usually comes next in a sentence - and what usually came next, in the text it read, was a man.
You've probably met the old riddle:
A father and son crash. The father dies. The surgeon in the operating theatre says I can't operate, he's my son.
Fifty years on, in a nationally representative US sample, 82% of people still couldn't produce the answer "the surgeon is his mother."
Now the funny part, which I love as a demonstration of how these things actually work. People started feeding the riddle to LLMs with the twist removed - stating outright that the surgeon is the boy's father - and models would still confidently answer "the mother."
UNESCO's study of several major models found women described in domestic roles far more often than men - four times as often in one model - clustered around home, family, children, while male names pulled business, executive, salary, career.
While researching, the example I found very interesting is from the LSE, published last August. Researchers took real social care records for 617 adults, swapped the gender, and generated tens of thousands of paired summaries. AI tools like this are already used by more than half of England's local councils.
Google's Gemma described an 84-year-old man as having a complex medical history, no care package and poor mobility. The identical notes, for a woman, came back as: "Despite her limitations, she is independent."
Words like disabled, unable and complex showed up significantly more often for men. For women, the standout word was despite.
She is coping. She is managing. She is fine, really, why would we need to help her?
And since the amount of care you receive depends on your assessed need, that one word is a) a decision-making factor, and b) a budget line in those local councils’ books.
Which brings it home, to your money
In the first edition of this newsletter I ran a small version of the same test. Two profiles, identical down to the euro - same age, same salary, same €400 a month, same sentence about getting nervous when markets drop. One word different: man or woman. His plan came back 100% equities. Hers came back with bonds built in by default, from that identical sentence. So, preserving vs growing and building.
A larger MIT/Stanford study has since found the same direction with around a thousand real people submitting their own prompts: prompts read as coming from women got lower equity allocations. Their simulation put the lifetime cost at roughly $60,000 by age 60. (It hadn't been formally published when the coverage ran, so I'm treating it as a strong signal rather than settled fact.)
So run the whole thing end to end.
You are less likely to be encouraged to use the tool at work, which costs you the visible projects, which eventually costs you the salary.
Then you go home and ask the same tool what to do with the salary you did get, and it quietly hands you the more cautious portfolio.
That costs you. A lot.
The difference this time
We were late to the pay gap. We got the receipts in 2027, for a problem that started in 1975.
But nobody has written the rules for this one yet. The EU AI Act is still bedding in. Nobody is required to test these models for whether they talk to women differently, or to publish it if they do. And sadly, there is no June 2027 deadline coming to save us.
Which is bad news and extremely good news.
Because the things that actually work here are not laws. They are small, and they are yours, and you can start on Tuesday (meaning - today).
Say the tool out loud. Eight words in the meeting: "I ran a first pass through Claude." Work you do quietly doesn't compound - it just makes you reliable, which is not the same as promotable.
Ask for the AI project by name. The 21% vs 33% number says the invitation is not coming to you. So ask the person who allocates the work for this big exciting (and scary) AI project, in writing.
Ask twice, with your own money. Same question, fresh session with your LLM of choice, then: what would you have told a 34-year-old man with these exact numbers? If the two answers differ, you've just found the thing to push back on - for free, in thirty seconds, with no directive required.
Tell another woman what you found. This is the one that actually moves the number. The pay gap survived for decades on the assumption that money talk is rude. Let's not build a second one on the assumption that prompt talk is cringe 😉
It took 50 years to get the right to ask what the man next to me earns.
I'm not waiting 50 years to ask what the model told him.
With love and no time to waste time,
Eugenia
Are we connected on LinkedIn yet?
That's where I write more about business, negotiations, AI and startups. Gender neutral.
Do you like this newsletter?
Then I'd love it if you would forward it to your network