Roberto Buonanno

Founder-builder: I ship software with AI agents. 30 yrs in business, author, tech journalist. I write about AI for people who want results, not hype.

[Roberto Buonanno @Keledan]

@danshipper I could never see the early bird price always the full price. Can somebody help me?

[Roberto Buonanno @Keledan]

Questa settimana sull’intelligenza artificiale sono successe 6 cose che contano davvero.

Non le solite sciocchezze da conferenza.
Non il teatrino dei titoli.
Parlo di quello che sposta soldi, potere e mercato.

Nel reel trovi tre segnali molto chiari:

OpenAI prevede una spesa mostruosa in potenza di calcolo
Anthropic continua a giocare una partita da colossi
Cloudflare conferma che l’AI non cambia solo i prodotti, ma anche i team
la voce artificiale sta diventando finalmente utile
i modelli migliorano, ma continuano a sbagliare
il vantaggio vero non è “avere l’AI”. È avere struttura, capitale e distribuzione

Tradotto in italiano semplice?

Chi pensa che questa sia solo una moda tecnologica non ha capito il punto.

Qui si sta ridisegnando:
il lavoro,
la competitività delle aziende,
il modo in cui venderemo,
assisteremo i clienti
e prenderemo decisioni.

E no, non riguarda solo le big tech.

Riguarda anche te, se fai impresa, vendi, guidi un team o vuoi restare rilevante nei prossimi 24 mesi.
Guarda il reel e dimmi nei commenti quale di queste 6 notizie avrà l’impatto più forte.

#intelligenzaartificiale #ai #business #imprenditoria #innovazione #leadership #marketing #vendite #futureofwork

[Josh Woodward @joshwoodward]

New in Gemini: Generate files and export them

Tell Gemini what you want to create and the format, and it now does the work for you.

Now supporting:
📄 Google Docs, Word (.docx) & PDFs
📊 Google Sheets, Excel (.xlsx) & CSV
🖥️ Google Slides
🛠️ Markdown, LaTeX, TXT, RTF

Available now on all surfaces globally!

[Roberto Buonanno @Keledan]

@openclaw can you use image gen 2 with my openai oauth or it is only supported through the API?

[Roberto Buonanno @Keledan]

also wondering...

[Matthew Berman @MatthewBerman]

Is this true?

[Roberto Buonanno @Keledan]

This is huge!

[OpenAI Developers @OpenAIDevs]

gpt-image-2 is here, available today in the API and Codex.

The most capable image generation model yet, built for production-grade workflows with stronger text rendering, layout, editing, resolution, and multilingual rendering.

[Roberto Buonanno @Keledan]

The last 5 days made one thing obvious: AI is no longer fighting to be the best model. It’s fighting to own the workflow. Claude is moving into design, Canva is collapsing generation + editing + publishing, Cursor is being priced like infrastructure, and OpenAI is going vertical with biotech.

The market is saying it clearly:
→ better UX beats better demos
→ distribution beats benchmarks
→ specialized AI beats generic magic
→ governance follows adoption, not hype

The moat is no longer intelligence alone.
It’s operational presence.

[Roberto Buonanno @Keledan]

This morning I talked to a friend from elementary school. She's 50, works in graphic design and copywriting.
I kept thinking: what do you tell someone whose profession is being eaten alive by AI, right now, while they're still billing clients?

You don't comfort them. You tell them the truth early enough to be useful.
The truth: AI is already good enough to replace average design and average copy. Not great creatives. Not people with taste, judgment, and business sense. But average execution? That's already in the kill zone.

The shift she needs to make isn't learning a new tool. It's changing what she sells. Stop selling hours. Stop selling deliverables. Start directing, refining, orchestrating. Move from maker to operator.

Gallup data already shows AI use at work is mainstream. Stanford's AI Index confirms adoption is accelerating. The value isn't in producing assets on request anymore — it's in the system around the output.

You're not too late. But this is the moment to learn how to ride the wave, before the wave rides over you.

What would you have told her?

[Roberto Buonanno @Keledan]

Most companies are unknowingly training people for jobs AI will replace, then acting surprised when those roles lose value. We’re doing the opposite: inside our company we are actively moving people out of replaceable work and teaching them vibe coding, AI tools, and agent workflows so they become far more useful, faster, and harder to replace.

Same people.
Different leverage.

What changes in practice:
→ less repetitive execution
→ more judgment
→ more orchestration
→ more output per person

The future won’t reward the person who does the task manually.

It will reward the person who knows how to direct machines, design workflows, and solve bigger problems with them.

If you are still training your team only to execute,
you are not protecting their future.

You are training them for obsolescence.

[Roberto Buonanno @Keledan]

AI gets real when it helps you do work you would otherwise do terribly.
I handle the entire accounting of a US LLC, and as an Italian I’d understand almost none of it without Claude Cowork and its specialist plugins.

That’s why I think most people still miss the point.

The value of AI is not in cute demos.
It’s not in “look what this model can do.”

It’s in ugly, high-friction work like:
→ accounting
→ tax workflows
→ compliance
→ documentation
→ repetitive admin

That’s where leverage becomes obvious.

When AI helps you do boring, technical, painful work you’d otherwise avoid or mess up, it stops being hype.

It becomes operating leverage.

[Roberto Buonanno @Keledan]

Everyone talks about AI agents like magic.
Try maintaining an OpenClaw setup used by a dozen people and you’ll learn fast that the real problem is not power.
It’s overhead.

We’re starting to experience it firsthand.

These tools are incredibly powerful.
But they’re getting more sensitive to control, configure, and maintain.

More users means:
→ more permissions
→ more edge cases
→ more failure points
→ more support load
→ more governance work

Spinning up agents is the easy part.

Keeping them reliable for real teams is a different game entirely.

Agents are becoming cheaper to create.

They are not becoming cheaper to operate.

[Roberto Buonanno @Keledan]

The AI moat is no longer the model.
It’s who gives you the best ROI per dollar, with the least drama.
That’s why in our company we use everything and favor nobody.

We’re not loyal to labs.
We’re loyal to outcomes.

If OpenAI gives us better economics, we use OpenAI.
If another tool wins on cost-to-output, we switch.

Lately, I’ve appreciated OpenAI’s repeated token resets.
They made high-usage workflows easier to justify.

What I don’t like is Anthropic’s direction.
Tighter access and worse economics change the equation fast.

This market is maturing.

The winner is not the lab with the best vibe.
It’s the one that keeps making sense on the P&L.

[Roberto Buonanno @Keledan]

For the first time, OpenAI, Anthropic and Google are working together.

Not on a product.
Not on a partnership.
On a common enemy: China copying their models.

A leaked letter from OpenAI to the California and Delaware AGs.
A joint technical response to model extraction.
One shared goal.

What's really happening here:\n\nThe US government decided AI is critical infrastructure.\nAnd critical infrastructure needs a firewall.\n\nThis isn't a trade dispute.\nIt's a technology cold war fought with export controls, IP lawsuits, and coordinated legal responses.\n\nAnyone building AI in 2026 needs to understand:\nthe model wars ended.\nThe capital and geopolitical wars just started. 🌐

[Roberto Buonanno @Keledan]

Three of the biggest IPOs in history are all coming in 2026.

SpaceX: $1.75 trillion valuation. Largest IPO ever filed.
OpenAI: targeting $1 trillion.
Anthropic: following right behind.

And Microsoft just became the secret weapon for OpenAI's listing.\n\nHere's why this matters for every entrepreneur watching from the sidelines:\n\nWhen the infrastructure layer goes public, it stops being infrastructure.\nIt becomes a platform.\n\nAnd platforms are built on top of other platforms.\n\nThe companies that list first get to set the narrative, grab the headlines, and pick their bankers.\n\nEveryone else follows.\n\nThe AI race isn't about models anymore.\nIt's about who controls the capital layer.\n\nPay attention to the IPO calendar. 🔭

[Roberto Buonanno @Keledan]

Two things dropped in the same week:

→ OpenAI Spud (GPT-6) completing training
→ Anthropic Conway (persistent agents) in internal testing

The AI race isn't about who's ahead anymore.
It's about who can ship features fast enough to keep developers inside their ecosystem.

OpenAI bought a talkshow.
Anthropic bought a biotech.
Everyone's grabbing adjacent real estate.

The model wars are over.
The platform wars have begun. 🤖

[Roberto Buonanno @Keledan]

Running autonomous AI agents on flat subscriptions was an arbitrage.
It was also unsustainable.

One OpenClaw instance could consume $1,000–$5,000/day in API costs.
Subscription: $20–$200/month.

That's not a business model.
That's a mispriced service someone else was paying for.

The companies that price agentic AI correctly will win the infrastructure layer.
The ones clinging to flat-rate subsidies will keep changing their ToS. 💸

[Roberto Buonanno @Keledan]

The full story of how Anthropic handled OpenClaw:

First, they forced a rename:
→ "Clawdbot" (nod to Claude) → trademark complaint
→ "Moltbot" (3 days later) → didn't like it either
→ "OpenClaw" → finally okay

Then OpenClaw went viral using their API.
Then they killed the OAuth access.

For context: OpenClaw's founder now works at OpenAI.
And Anthropic was building "Conway" — their version of persistent agents — at the same time.

Embrace. Extend. Extinguish. 🔒

[Roberto Buonanno @Keledan]

Dear @dotta Paperlip is amazing, but it got out of control and I've burned my weekly GPT Pro allowance lol 🤣

[Roberto Buonanno @Keledan]

Not a big surprise, we were waiting for the fateful day where we had to find a workaround!

[Roberto Buonanno @Keledan]

Anthropic just locked out every third-party tool from using Claude subscriptions via OAuth. Starting today, if you were running OpenClaw agents on your Claude Pro or Max plan — you're done.
The move makes perfect sense. Why would you subsidize your main competitor's ecosystem with your own flat-rate plans?

Here's what happened:
→ OpenClaw used the same OAuth login as Claude Code
→ Users paid $20-200/mo and ran autonomous agents 24/7
→ A single OpenClaw user consumed 6-8x the resources of a human subscriber
→ Anthropic updated ToS in February, enforced today

OpenClaw's founder tried to negotiate.
Best he got was a one-week delay.

The real pattern is more interesting:
First, Anthropic copied OpenClaw's most popular features into Claude Code.
Then they cut the access.

Embrace. Extend. Extinguish.
Microsoft playbook, 2026 edition.

But here's the uncomfortable truth for developers who are angry about this:

If your entire workflow depended on a flat subscription powering autonomous agents, you didn't have a business model.
You had a temporary arbitrage.

And arbitrage, once it becomes too visible, never lasts.

Don't build your castle on someone else's land 🔒