Dear Dysruptors,
Fernando Santa Cruz here in the forty-seventh edition of Weekly Synapsis — where a brokerage handed real-money portfolios to AI agents for 27 million people, one agent now writes 89% of its own code, and the Vatican published its first encyclical on AI, comparing it to nuclear weapons.
Writing from Toronto, between sessions with clients in construction and real estate, while in Mexico we continue advancing, together with partners, AI training for leaders and the next phase of the ecosystem in Yucatán.
This week, AI crossed a line.
It stopped giving us answers. It started executing.
For two years, we asked it to do things. This week, it started managing money, writing its own code, and moving boxes in real warehouses.
And in that same week, the Pope called for governing it with the urgency of nuclear disarmament.
That’s today’s tension: the agent economy accelerating on one side, and the ethical question racing to catch up on the other.
This newsletter expands on the WhatsApp summaries (week of May 25–30) to understand what changed when AI moved from answering to executing, why that creates a concrete opportunity for SMBs, and which very human decision now belongs to you before handing over control.
Anthropic Surpasses OpenAI at $965B: Revenue Beat the Promise
Anthropic closed a Series H round worth $65 billion and reached a valuation of $965 billion.
For the first time, it surpassed OpenAI, which remained at $852 billion.
But the number that matters isn’t the valuation.
It’s the $47 billion in annualized revenue, driven by companies paying for Claude Code every single day.
For years, AI investment funded a promise: AGI, general intelligence, the future.
This week, the market changed the question.
It stopped asking “What do you promise?” and started asking “What do you invoice today?”
It’s like moving from financing the blueprint of a house to financing the house already occupied, with rent coming in every month.
For an SMB, this signal is more useful than it seems. The most sophisticated capital in the world is rewarding automation that delivers measurable returns now — not automation that promises a revolution someday.
Your criteria for adopting AI can be exactly the same:
Does this tool save me time or money this quarter — yes or no?
Promise no longer pays.
Revenue does.
Question for your strategy:
Is your AI spending buying a promise for someday, or savings you can measure on your next invoice?
When AI Gets a Wallet: Robinhood Hands Real Money to an Agent
Robinhood enabled agentic trading for its 27 million users.
Now you can connect an agent — Claude, ChatGPT — to a sub-account with real capital.
The agent analyzes your portfolio, measures risk, and executes trades autonomously.
This is the first mass-market brokerage to place real money in the hands of an AI model.
Think about that for a second.
We went from asking AI for text to giving it a wallet.
But they didn’t hand over the vault keys.
They gave it a spending card with limits.
The agent operates inside an isolated sub-account — a financial sandbox — using the MCP protocol as rails: it only accesses pre-approved data and actions, while traditional systems block anything reckless before it reaches the market.
Like giving a corporate card to a new employee: trust exists, but with a spending cap and no access to the safe.
For an SMB, the story isn’t that you’ll let a bot invest your savings.
The story is governance.
This is how you delegate any consequential task to AI:
limited budget, minimum permissions, and a clear boundary where the machine stops and asks.
Trust no longer means surrendering control.
It means putting rails around control.
Question for your finances:
What limited business decision could you delegate to an agent this week — and what exact spending cap would you set before letting go?
Devin Goes from 13% to 89%: The Tool That Sharpens Itself
Cognition raised $1 billion and reached a valuation of $26 billion.
Its agent, Devin, went from writing 13% of its own code in December to 89% today.
Six months.
From apprentice to self-builder.
With its new model, Devin reduced a Mercedes-Benz software project from 8 months to 8 days.
From eight months to eight days.
The point few people are connecting is this:
we are seeing the first commercial examples of AI improving itself.
The tool building a better version of the tool.
It’s like a carpenter who stopped buying chisels and started forging them himself — each generation sharper than the last.
Honesty about the other side:
this directly pressures entry-level and mid-level technical jobs.
That’s not a minor nuance.
But for SMBs without large development teams, the story is access.
Technical projects that were previously impossible due to cost or time — integrations, internal tools, migrations — return to the table.
What once required a team and a semester now fits into a week.
And this reveals the deeper question of future work.
When the tool starts building itself, value stops living in knowing how to do.
It moves to knowing what is worth doing.
The person who directs the agent well doesn’t compete against one hundred programmers.
They coordinate them.
Question for your team:
What technical project have you postponed because it felt too expensive or because you don’t have enough people — and how does the equation change if one well-directed person can execute it in days?
The Best AI Isn’t the One That Knows the Most: Opus 4.8 Learns to Ask for Help
Anthropic released Claude Opus 4.8 with one million tokens of context and the ability to orchestrate hundreds of sub-agents working simultaneously.
The proof:
it migrated 750,000 lines of code into another language in 11 days, with a 99.8% success rate.
And it introduced a Fast Mode that runs 2.5 times faster.
But the most important advance isn’t raw power.
It’s honesty.
Opus 4.8 was trained to recognize its own limits: when a sub-agent becomes uncertain, it stops, raises its hand, and escalates the problem to a human or another agent instead of inventing an answer and pushing forward.
Imagine the kitchen of a massive restaurant.
The executive chef doesn’t chop every onion — they delegate to hundreds of cooks.
The system doesn’t collapse because someone makes a mistake.
It collapses because someone burns the sauce and hides it.
The best AI of the week isn’t the one that knows the most.
It’s the one that knows when to say:
“I need help.”
For an SMB, there are two lessons.
The practical one: Fast Mode lowers the cost of everyday tasks — writing, spreadsheet analysis, research — to a fraction.
The deeper one: the trait that makes AI trustworthy is the same trait that makes teams trustworthy.
Not never making mistakes.
But signaling early when something is going wrong.
Question for your operation:
Do your processes create space where someone — person or machine — can stop and say “I don’t know, this feels risky” — or do mistakes stay hidden until they explode?
AI Moves Into the Office and Your Phone: The Canvas Now Thinks With You
This week, AI left the separate tab and moved into where we actually work.
Perplexity integrated natively into Word, Excel, PowerPoint, and Outlook.
It researches, writes, and analyzes data directly inside the file, reading from sources like SharePoint and Snowflake.
No copying.
No pasting.
No changing screens.
The tab-juggling is over.
The brain used to be separated from the canvas.
Now the canvas thinks with you.
And on the phone, something even bigger happened.
Google is bringing AI Studio to iOS on July 1 across 76 languages: speak an idea into your phone and the app builds itself.
At the same time, Apple registered the domain genai.apple.com, preparing a rebuilt Siri powered by language models for iOS 27.
Human language — your English, your Spanish — is becoming the universal programming language.
For an SMB, the barrier to creating technology is almost disappearing.
A business owner without a single programmer will be able to dictate the internal tool they need while waiting for the bus.
Not next Monday.
This summer.
We stopped learning machine syntax.
The machine learned ours.
Question for your productivity:
How many cycles of copying, pasting, and changing tabs does your team still perform every day — and how much of that could be absorbed by the document itself?
Text Has No Weight. A Box Does: Figure and One Humanoid Robot Per Hour
Figure Robotics signed an agreement to deploy autonomous humanoid robots into real warehouses in Reno, Nevada.
It now manufactures one robot per hour.
AI left the screen and started colliding with physical reality.
And to get here, it had to solve something text never demanded.
Text has no weight.
A box does.
For a robot not to crush a box or trip over a person, predicting the next word isn’t enough.
It must understand gravity, weight, and space — build an internal model of the physical world.
It’s the difference between describing a staircase and walking down it without falling.
For SMBs, humanoids in your warehouse are still far away and expensive.
That’s not today’s news.
Today’s news is this:
physical automation stopped being a trade-show promise and entered the industrial calendar.
Manual bottlenecks — logistics, inventory, heavy transport — will be among the first to feel cost pressure across supply chains.
Worth watching, even if it’s not time to buy.
And it leaves us with an old question wearing new clothes.
For centuries we believed the last thing machines couldn’t touch was the body:
craft, hands, strength.
That frontier is moving too.
What remains human won’t be what we do with our hands.
It will be what we decide with judgment.
AI moved from mapping language.
To mapping reality.
Question for your plan:
What physical bottleneck in your operation still depends entirely on human hands — and are you watching it, or waiting for it to surprise you?
$150 Billion to Taiwan: NVIDIA Bets on Hardware Nobody Can Relocate
Jensen Huang committed $150 billion annually to strengthen NVIDIA’s manufacturing partnership with TSMC in Taiwan.
The move directly challenges U.S. pressure to bring chip production home.
And it reveals an uncomfortable truth about this entire race.
Software changes in days.
Hardware remains anchored to infrastructure that takes years to build.
TSMC’s advanced packaging cannot be replicated in a quarter.
There is no short-term replacement.
It’s like a deep-water port.
You can reflag ships overnight.
But the port took a century to build — and it cannot move.
For an SMB, the lesson isn’t geopolitical.
It’s strategic.
Your business also has a software layer — easy to copy, easy to move.
And a port layer — slow, difficult, nearly impossible to replicate.
Your customer relationships.
Your local reputation.
Your knowledge of the terrain.
That’s your Taiwan.
Cheap advantages get copied over a weekend.
Physical advantages don’t.
Question for your strategy:
What is the deep-water port in your business — the thing a competitor with more money still couldn’t replicate over a weekend — and are you protecting it accordingly?
The Pope Calls for AI Disarmament: A Founder of Anthropic Makes a Public Confession
Pope Leo XIV published Magnifica Humanitas, the first comprehensive Church document on artificial intelligence.
It compares AI to nuclear weapons:
a technology that must be governed with the same urgency used to dismantle a bomb.
But the surprising part wasn’t the encyclical.
It was who stood beside the Pope when it was presented.
Chris Olah, co-founder of Anthropic — a company approaching a trillion-dollar valuation — publicly admitted that commercial and competitive pressure pushes AI labs toward unsafe decisions, and called for urgent external oversight.
The fact that a leader from one of the most valuable companies on the planet went to the Vatican to ask to be monitored is an extraordinary admission of vulnerability.
And the result is paradoxical.
We are turning to moral frameworks that are thousands of years old to govern a technology that has existed for only months.
It’s like the physicists who built the bomb and later begged the world to control it.
Capability arrived first.
Wisdom to contain it came later.
And the encyclical doesn’t stop at warning.
It asks three questions that engineering alone cannot answer:
- What happens to global poverty if AI displaces work?
- What does human flourishing mean once machines begin doing what used to define us?
- And what are we actually building if algorithms begin to imitate even inner human experience?
For an SMB, this isn’t a distant Silicon Valley debate.
It’s the question you’ll answer this year — on a smaller scale.
Every business adopting AI decides what it delegates and what remains in human hands.
Ethics stopped being a luxury for philosophers.
It became an operational decision you make every week.
Question for your leadership:
Before letting AI touch a part of your business, what rule would you write for it — what should it never decide alone, no matter what?
Tools You Can Start Using on Monday
Perplexity in Microsoft 365
Now lives directly inside Word, Excel, PowerPoint, and Outlook.
Research with cited sources, draft content, and analyze data without leaving the file — reading directly from your documents and sources like SharePoint.
For an SMB, it’s like having a junior analyst inside the document instead of another browser tab.
Runway MCP
Connect Runway directly to Claude or ChatGPT and generate advertising images and videos without leaving the conversation.
Describe the campaign, upload your product image, and receive ready-to-use assets.
Visual marketing without learning another platform.
ElevenLabs Dubbing v2
Translate videos into more than 90 languages while preserving your tone, rhythm, and emotional delivery.
For SMBs creating video content, this means entering international markets without re-recording or paying a dubbing studio.
Leonardo AI – Image to 3D
Turn a flat product image — clothing, jewelry, furniture — into a 3D model customers can rotate.
Ideal for modernizing an e-commerce store or digital catalog.
MAI-Image-2.5
Microsoft’s new image model finally renders readable text inside images.
Useful for flyers, banners, promotions, pricing, and dates without the classic AI spelling errors.
Available for testing in Arena.
Microsoft 365 Copilot (Redesigned)
Loads more than twice as fast and introduces Work IQ — a layer that understands the context of your emails, chats, and files to deliver more relevant responses.
If you already pay for Microsoft 365, this upgrade comes included.
Google AI Studio
Build functional applications simply by describing your idea — no coding required.
Already available on the web for internal prototypes and arriving on iPhone July 1 for voice-based creation.
Software creation no longer needs permission from a programmer.
My Invitation This Week: Your Red Line
This week the Pope wrote rules for AI at the scale of humanity.
This week’s exercise is to write yours.
At the scale of your desk.
Take twenty minutes and a blank sheet of paper.
We’re not making a list of things to automate.
We’re starting from the opposite direction.
First, write down your three red lines.
Three business decisions that, no matter what, you will never fully delegate to a machine.
Your final price to a client.
Firing someone.
The promise that defines your brand.
Whatever yours are.
And next to each one, write why.
That reason is your most human value — the thing no efficiency should override.
Now comes the turn.
Next to every red line, there is a supporting task you can let go of.
You decide the final price — that remains sacred.
But the agent prepares comparisons and proposes three options.
You decide who to hire.
But AI screens one hundred resumes down to ten.
You define your brand voice.
But the model drafts ten versions for you to choose from.
The red line stays with you.
The surrounding work gets delegated.
And here’s the actionable part:
Choose one of those supporting tasks.
This week, actually let it go.
With clear boundaries.
The same way Robinhood limits its agent.
What matters about the exercise isn’t what you automate.
It’s realizing that almost everything you thought was untouchable had hours of surrounding work you could release.
Ethics doesn’t slow adoption.
It tells you exactly where to begin.
Start Monday.
Three red lines.
One released task.
Closing
This wasn’t the week of a new model.
It was the week AI stopped answering and started executing.
It manages money.
Writes its own code.
Moves boxes.
Speaks ninety languages with your emotion.
And because of that, in the very same week, the Pope asked for rules.
The question of the year is no longer whether AI can.
It’s what we let it do — and what we keep for ourselves.
For an SMB, that question isn’t a threat.
It’s the first act of leadership in the era that’s coming.
Because defining the line between what you delegate and what you protect doesn’t require anyone’s permission.
It starts Monday.
With three red lines.
Fernando Santa Cruz
Head of AI & Automation @ Adivor Consulting
