Last week I wrote about pulling my son out of the traditional school system because the familiar path didn’t automatically feel like the right one anymore.
I’ve been thinking about that same instinct in business. We tend to treat the systems we already depend on as the safe choice, even when the technology around them has changed enough that the system itself may be the thing holding us back.
That is what feels important about AI right now. Agents should make software less restrictive. They can move across tools, use data wherever it lives, and do the work without forcing a person through five different interfaces. For the first time, the system underneath the work doesn’t necessarily need to be the center of it.
But the companies that built the old world have every reason to keep themselves in the middle. Salesforce can become the plumbing and hope the data is enough. HubSpot can put agents behind the orange walls. The biggest AI companies can make the technology sound terrifying enough that everyone runs back toward the vendors they already trust.
AI is opening the door, and a lot of companies are already figuring out how to close it again.
Salesforce went down globally during Dreamforce.
You cannot make this s*&t up.
Thousands of people sat inside Moscone applauding Salesforce’s plan to run the autonomous enterprise. Meanwhile, actual Salesforce customers couldn’t log in, submit support cases, or run scheduled jobs.
Salesforce wants to run your autonomous company but yesterday, it couldn’t even run Salesforce.
I would not want to be the poor bastard explaining that one to Marc Benioff this morning.
The timing is hilarious, but the bigger issue isn’t that Salesforce had an outage. Every technology company has outages. The issue is how much we’ve allowed systems like Salesforce to become load-bearing infrastructure for the business.
You put the customer data there. Then the workflows. Then the automations. Then more teams and more tools start depending on it. Eventually the platform becomes so embedded that replacing it sounds more dangerous than living with it forever.
That is what makes the next part so interesting. Because at the same event where Salesforce was selling the autonomous enterprise, it also showed us a future where people may not need to spend nearly as much time inside Salesforce at all.
Salesforce Is Becoming Plumbing
I wish I’d been at Dreamforce today to hear Marc Benioff announce the death of Salesforce.
Not the company but the shitty interface millions of people are forced to work inside.
AIforce and ClaudeForce let somebody else’s AI become the place you work while Salesforce hides underneath holding the data.
That's total surrender. Once AI owns the interface, Salesforce becomes plumbing. And plumbing is replaceable.
We already see it at Atonom. Our AI Employees sit above Salesforce, HubSpot, and the rest of the bloated tech stack. They do the work, move the data, and eliminate the login.
Soon the AI won’t care which CRM is underneath. Benioff unveiled the future today for sure. Unfortunately for Salesforce, it won’t be part of it :)
I think that is a much bigger admission than it looks.
For years, owning the interface meant owning the workflow. Eventually companies started adapting themselves to the software instead of the other way around.
Agents can break that relationship.
If an agent can pull something from one database, send the email, move the record, and report back without a person opening any of those systems, the workflow no longer belongs to the application.
That is a pretty uncomfortable future if your advantage came from being the place everyone had to work. Salesforce seems willing to accept some version of that and fight to remain important underneath it.
HubSpot apparently looked at the same future and decided it would rather keep the agent inside.
Stay Safely Inside the Orange Walls
I can’t get over HubSpot calling its conference UNBOUND, then using the keynote to trap your AI inside HubSpot.
The big keynote announcement was Agent Hub.
- Build your agents in HubSpot
- Give them HubSpot data
- Manage them with HubSpot
- Buy HubSpot credits when they do anything.
Stay safely inside the orange walls. HubSpot watched AI begin freeing people from bloated software and immediately started building a new prison.
Your AI should work for your company, not HubSpot. And it sure as hell shouldn’t need a hall pass from HubSpot to do its damn job.
This is the part I really don’t want us to screw up.
The opportunity with agents isn’t putting a smarter feature inside every piece of software we already own. It’s finally being able to move above a software stack that has become absurdly complicated.
Yet the SaaS playbook is already creeping back in.
We already know where that road goes. Companies spent fifteen years buying software to make every function more efficient and somehow ended up needing entire teams to manage the software.
If AI Employees really work for the company, they should be able to go wherever the work requires. The moment the agent needs permission from one vendor to leave the ecosystem, we’re not building a new operating model. We’re just rebuilding SaaS with better technology.
And there is one thing that makes convincing companies to stay inside those walls a whole lot easier: fear.
Fear Is GREAT Marketing
Can we all stop pretending Anthropic is some nonprofit warning humanity about AI? They’re selling software, for hell’s sake.
Every time this dude Dario predicts mass unemployment, cyber collapse, or the end of civilization, Claude gets another billion dollars’ worth of free press.
- Scare the s*%$ out of everyone
- Sell yourself as the only adult capable of protecting them
- Push regulations that suffocate smaller competitors
- Call the entire power grab “responsible AI”
It’s fu*&%#& brilliant marketing. But that’s all it is.
Anthropic isn’t warning the world, they're just selling to it.
Anthropic isn’t warning the world, they're just selling to it.
A few weeks ago I wrote about how differently people already see AI. Some people look at it and see leverage. Others hear AI and immediately think job loss, replacement, fraud, or something they should probably be afraid of.
This is part of why.
The scarier AI sounds, the easier it becomes to convince companies that the safe move is to stay with the giant vendor, stay inside the approved ecosystem, and wait for someone else to tell them how to use it.
Then executives buy Claude for everyone, call the company “AI-first,” and wonder why nothing actually changes.
Claude Is Minecraft for Professionals
Claude is Minecraft for professionals. You spend all day building cool s*&t, showing it to your coworkers, and creating absolutely no business value.
The Software Might Stop Being the Point
Salesforce wants to become the plumbing. HubSpot wants to keep you inside its walls. Anthropic wants to hand everyone Claude and call it transformation. But AI should make all of them matter less.
We’ve spent years forcing people to work the way software works. Now the software can finally work the way people do.
It would be pretty damn stupid to use AI to build the same walls all over again.
