Hey👋,
I'm Giacomo

10 metres can turn a colleague into a stranger.

In the 1970s, MIT professor Thomas Allen studied how engineers shared technical information and communicated at work.

The finding sounded obvious:
The farther apart people sat, the less they talked.

But the curve was not gentle.
It collapsed.

Within a few metres, communication was very common.
Around 10 metres, it dropped sharply.

By 25–30 metres, it was close to zero.
After that, extra distance barely mattered.

Someone on the other side of the floor behaved, communication-wise, like someone sitting in another building or even in another city or state.

Fast forward to 2026,
a study in the Quarterly Journal of Economics followed 1'055 software engineers at a Fortune 500 company.

Some teams sat in the same building.
Some were split across two buildings, only a 10-minute walk apart.
Others were remotely distributed.

Same-building teams received 24% more code-review comments.
When offices closed during the pandemic, that advantage largely disappeared.
The estimated loss of proximity reduced online feedback by 18%.

Online feedback.
Despite Slack, Zoom, GitHub etc.

I see it every week in my work.
The people I learn from most are the people sitting right next to me.

Spontaneous comments, random questions.
Not to mention sneaking a peek at colleague's screens :)

I learnt more from those conversations than from most courses.

Remote work can improve productivity and efficiency.
This is fact.

But measuring only short-term productivity is dangerous.
An employee can look more productive today because nobody interrupts them. But they will grow much slower, ultimately delivering lower value in the long term.

And this is not just bad for the employee, it's bad for business too.

This doesn't mean 5 days in the office is the answer.
It means physical proximity has real value,
especially when equipped with enough unstructured space for spontanous conversations to happen.

Being forced to come in the office all the same day has no value if it feels like a chore. People will come disengaged and try to sit as far apart as possible not to be bothered.

It has to feel spontaneous.
Spontaneous conversations by the desk fuelled my career more than anything else.

Or do you feel like remote productivity trumps spontaneous learning?

Spotify is truly the king of audio.

I admit. I’m a very slow reader.
It takes me ages to finish a book.

That’s why audiobooks work so well for me.
I loved when Spotify added audiobooks to Premium last year.

But not every book works well in audio.

Some books require more focus than just listening.
Sometimes it’s easier to go back to a paragraph, or a highlighted section in a physical copy. Other audiobooks are poorly narrated.

So I've found a system.
I listen to the audiobook and buy the physical book as well.
Most of the time, I only read a few sections on paper, but I complete the book through audio.

I've always wanted a way to start listening exactly from where I left off in the physical book. Sci-fi, I thought.

Well, Spotify just made it!

Now they're asking me to pay extra for Audiobook+. I think I'm gonna do it, they've earned it!

Jenni Romaniuk glad your book is included :)

ChatGPT Ads are a different beast.

Even compared to other AI chatbot ads, like Google AI Mode.

Ads in Google AI Mode are mostly traditional Google Search ads with an AI flavour.

In AI Mode, someone types a rich-intent query and Google serves an ad immediately. Pretty much like on the traditional SERP.
Very often, the conversation ends there.

Because it was not really a conversation to begin with.

AI Mode often appears when users click “show more” under AI Overviews.
They wanted to read more, not to start a whole conversation.

But ChatGPT is different.
People come to ChatGPT to talk.

This means ads are more embedded in the user journey.
They appear inside a conversation the user is already invested in.

That's why ads don't disrupt conversations, but rather enhance them, making them very relevant.
On average, users complete 4 more turns after seeing an ad on ChatGPT.
In AI Mode, the "conversation" generally ends right after the ad.

On ChatGPT, conversations often don't start with clear commercial intent. Sometimes with no clear intent at all.

Intent emerges slowly. Question after question.

And this changes everything.

According to Similarweb, 46% of users who opened a conversation with zero commercial intent developed buying signals by the time an ad appeared.

So, AI conversations in ChatGPT don’t just capture intent.
They create it!

Which makes ChatGPT Ads a genuinely new form of advertising.

And an extremely powerful one!

Source: Similarweb "Ads in AI" report.

I thought OpenAI’s ecosystem couldn’t get any worse.

I was wrong.

Codex is now the ChatGPT app.

But wait…
The “normal” ChatGPT desktop app still exists.
It's been renamed to ChatGPT “Classic”.

But wait…
Inside the new ChatGPT app, I can choose between:
• ChatGPT Work
• Codex
What's the difference??

But wait…
Inside both ChatGPT Work and Codex, I can access my existing ChatGPT web conversations and continue them in a smaller side panel.

But wait…
ChatGPT Work is also available on the web.
There, I can choose between Work and “normal” ChatGPT.
What's the difference?

But wait…
When I start a new chat from the desktop app side panel, is it a GPT Work chat or a "normal" chat?
Then, I can even integrate a web conversation into an "offline" Codex chat.

As if this wasn't confusing enough,
GPT 5.6 Sol has a Max effort that is not the actual max effort. Ultra is more max than Max 😅.
I'm a Pro subscriber, but the Pro model is only available in "Chat" mode, not in ChatGPT Work.
I guess in "Work" mode Ultra replaces Pro?

Honestly,
what a mess!! 🤣

I can see only one possible rationale.
OpenAI may be trying to separate casual consumer use from more advanced, work-oriented use.

"Classic" stays free or inexpensive.
Work eventually becomes usage-based, more like Codex.

Strategically, it might make sense.
But the current product experience is a total disaster.

Same account, same conversations,
but different interfaces, different names, different models.

Same features? not clear.

Same billing, but different limits logic.

OpenAI, I've been a loyal paying subscriber since early 2023, now send a product manager into the comments to explain this!

This is why AI chatbots completely change the game in marketing.

Even a conversation that started with zero commercial intent can turn into a buying occasion.

And ChatGPT ads are right there to capture the buying intent when it surfaces.

Or rather, shall we say they create the buying intent instead of just capturing it?

Source: Similarweb "Ads in AI" report.

OpenAI product ecosystem is becoming a joke.

Today I opened ChatGPT and discovered the new “Work” toggle.

I played around with it. It’s great.

It can:
• create and use Skills
• connect to GitHub
• select projects directly from the chat
• enable/disable Plugins from the chat
• pursue a specific goal
• and more

But wait.
Isn’t that exactly what Codex does??

Then it gets even more confusing.

Click “Open desktop app” inside a ChatGPT Work conversation and it opens Codex, not ChatGPT Desktop. Misleading at best.

I’m a Pro subscriber, but the “Pro” model has disappeared. Now I get Sol and a selector for how much effort I want it to use.
I assume “Ultra” effort replaces the Pro model?
Clearly they haven't updated the billing plans yet.

Codex doesn’t have Sol or GPT-5.6, yet at least.

Meanwhile, the actual ChatGPT Desktop app doesn’t have the Work toggle and can’t select Sol or reasoning efforts. 😅

Then there are Skills.
I can now create them in ChatGPT Web. But to me they look extremely similar, if not essentially identical, to Custom GPTs.
People tell me Skills are different because they can be invoked mid-conversation.
But Custom GPTs can too! Just type @ and select one.

Finally, Projects.
They’re kind of like Custom GPTs, but not really.
Like folders in Codex, but on ChatGPT.
They can store files and documents, but not Skills.

OpenAI, I love you and I've been a loyal paying customer since early 2023.
But please help me understand your products!

ChatGPT Ads in numbers.

Not search, not display, not social.

What is this??

If you’re European and spend time on LinkedIn or Instagram, chances are your feed was flooded with posts about the Bending Spoons IPO.

And rightly so.
Bending Spoons is a massive European success story. 🇪🇺

But the IPO itself was not that special.
It was not even the biggest in Europe.

Special for Italy, sure. A proud moment.
But some Italian LinkedIn commentators went as far as saying the IPO should have been aired live on every mainstream TV channel…
with the national anthem as the soundtrack.

That might be a bit too much 😅

Bending Spoons has built an amazing business.
(more on why it's the case in my post from Monday)
But its model is complex, niche, and not easy to grasp for the average person.

Specialised media like the Financial Times and Il Sole 24 Ore covered it extensively. And that’s where this kind of news belongs.

Let's curb the hype :)

For context,
SMG Swiss Marketplace Group, my employer, IPO’d last September.
It was the largest IPO in Europe in 2025 up to that point. Tech company.

Yet, nobody played the Swiss anthem on national TV. 😅

PS.
How many companies in this list do you know?

Infographic courtesy of ChatGPT

OpenAI is becoming an advertising company.

At Cannes Lions last month, David Dugan, the new Head of Global Ads Solutions, made the direction clear.

In just 6 months, OpenAI has built a serious ad business.

• Self-serve Ads Manager
• CPC bidding
• Conversion tracking
• New measurement tools
• Distribution partnerships
• Feed-based ads and advanced audience targeting under development

AI ads are not search ads. Not social ads.
They are not classic display either.

They sit in a strange middle ground.

On search, users show intent first and ads simply respond to it.
On social, ads interrupt attention inside a feed.

In ChatGPT, the ad appears inside a conversation that may not have started with buying intent at all.

According to Similarweb, 46% of users who opened a conversation with zero commercial intent developed buying signals by the time an ad appeared.

AI conversations don’t just capture intent.
They help create it.

That makes ChatGPT Ads feel closer to brand advertising than direct response.
But the mechanics and ad format still look like performance marketing:

• 0.68% overall CTR, up to 5.4% for top brands
• Around $60 CPM
• Around $12 CPC

Too weak for search.
Too strong for display.
Possibly closer to premium display or even podcast advertising.

And OpenAI has a clear reason to push hard on ads.
Subscription revenue alone seems unlikely to cover the massive costs.

Advertising is an obvious solution.
OpenAI reportedly expects $2.5B in ad revenue in 2026.
And even $100B by 2030!

For context, Pinterest made $4.2B in 2025.

Large advertisers are already moving in.
Including Amazon, which unexpectedly started advertising product listings on ChatGPT in June.

We are witnessing the birth of a new advertising giant!

Now the questions is:
how much budget will it pull from Google and Meta?

Source: Similarweb "Ads in AI" report
Infographic courtesy of ChatGPT

Forget SpaceX, OpenAI and Anthropic.
The most interesting tech IPO of the year comes from Italy.

Bending Spoons listed on Nasdaq last Wednesday.
On day one, the stock closed almost 40% above its IPO price. Better than SpaceX.

Nice debut, but that's not the point.

The real story is what Bending Spoons says about the future of work in the age of AI.

Its business model is brutally simple:

Buy popular internet products that lost momentum,
integrate them into one lean operating machine,
layoff almost everyone,
boost revenue and profit.
Repeat.

Just in the past three years they have acquired Evernote, WeTransfer, Vimeo, AOL and many others.

Their lean machine is driven by 621 core “Spooners”,
a super team of engineers, product and monetisation specialists mostly based in Milan.

Acquired companies are cut very deeply and integrated into the Spoons machine. Evernote, for example, lost practically its entire pre-acquisition workforce.

But revenue doubled!

And that's the whole model in one sentence.

In most companies, managers ask for more people.
More headcount and direct reports to put on the CV.

At Bending Spoons, it’s the other way around.
Managers want to make their team smaller.

The end result is insane.

In Q1 2026, Bending Spoons generated around $1 million in revenue per employee.

Higher than Meta, Apple or Google.
Only NVIDIA sits above it.

And this is not just a productivity metric.

It is a warning.

You just don’t need as many people as before.

A few hundred exceptional people, supported by AI and a ruthless operating system, can now do what thousands used to do.

Bending Spoons just proved it and built an entire business model around it.

Investors loved it.

Inevitably, many companies will take note.

Source: IPO filings.
Chart courtesy of ChatGPT.