I've been circling this conversation for months now.

In client sessions, in newsletters, on stage: I keep coming back to the question of what happens to traditional industries when AI stops being a productivity tool and starts being the interface. I've talked about it in the context of travel, of software, of org design. It's been a drumbeat.

But something shifted for me last week when I heard someone from Mercury speak, and I want to tell you what did it, because I think it'll shift something for you too.

I sat down with my own company financials. Not a quarterly check-in with my bookkeeper. I pulled everything into Claude and interrogated it. Where am I bleeding money? Which revenue streams have the best margin? If I cut this category by 20%, what happens to runway? Given the mix of retainer vs. workshop revenue, how should I think about pricing the next engagement.

Twenty years ago, I would have needed a CFO, a week, and a spreadsheet built by a junior analyst to get to those answers. Ten years ago, I would have needed QuickBooks plus a consultant. Five years ago, I would have needed a dashboard tool and still wouldn't have trusted the output. Last week, I had a conversation. I've done this with my personal finances and also with taxes for my company.

And somewhere in the middle of that conversation, the bank branch analogy I've been using for months finally landed on me, not just on my audience.

Here's the visual:

• Bank branches (1950s): you went to the money
• ATMs (1980s): the money came to you, on the bank's terms
• Websites (2000s): the bank came to your desk
• Mobile apps (2010s): the bank lived in your pocket
• MCP and CLI (2025+): the bank becomes a service your agent calls on your behalf

Each shift did two things at once. It changed where the interaction happened, and it changed who captured the margin. But the deeper pattern (the one I missed until I was staring at my own P&L) is this: each era redistributed who did the work, who held the customer relationship, and where the margin pooled. The agent era is violent because the customer relationship itself gets intermediated by something that isn't loyal to anyone. In finance, Mercury has nailed this in 2025. How long will take for others to follow?

Three places I'd press on this if I were underwriting a deal right now or choosing what company to work at.

B2B SaaS: the seat-based business model breaks. Salesforce, Workday and ServiceNow are still down 30-40% YTD because of this dynamic. The entire SaaS playbook (land a champion, expand seats, charge per user) assumes humans log in. When an agent does the work, how many seats does a 500-person company need? Maybe 50. Maybe 5. Salesforce, Workday, ServiceNow all have revenue models predicated on human-in-the-loop usage. MCP turns them into backend databases that agents query. The value doesn't disappear, but the pricing power does. Question for you: Which portfolio companies price on seats vs. outcomes? The seat-based ones are holding branch-era real estate.

Consumer brands: the shelf disappears. Amazon was the website era. Shopify and D2C were the app era. The MCP era kills the storefront. When my agent buys dish soap, it doesn't browse. Instead, it picks based on my stated preferences, household inventory, and price. Branding, packaging, shelf placement, influencer marketing, SEO (the entire surface area of CPG competition) becomes irrelevant to the transaction. What matters is whether your product sits in the agent's consideration set and can transact frictionlessly. Procter & Gamble has spent a century building branch networks made of Super Bowl ads. Question for you: Which CPG brands have product data clean enough for an agent to evaluate them? Most don't.

Professional services: the billable hour was always the ATM. Law, accounting, consulting. Each tech wave squeezed the lower work, but the billable hour survived because judgment lived at the top. MCP breaks this differently than people think. It's not that AI replaces lawyers (because lawyers will continue to exist). Instead, it's that the client's agent does the first pass on the lawyer's work. My agent reviews the contract before I see it, flags the terms, drafts the redlines, and presents me three options. The lawyer's output becomes an input to my agent, not a deliverable to me. Question for you: Which firms are building agent-consumable outputs vs. defending PDF deliverables?

Here's what changed for me, sitting with my own financials.

I've spent months telling clients what was coming. I believed it. But watching an agent do in forty minutes what used to require a team and a week (in my own business, with my own money) made the analogy visceral in a way it hadn't been. The branch isn't going away because someone built a better branch. It's going away because the work that used to happen inside it now happens somewhere else entirely, in a conversation the customer has with a system that knows them.

The meaningful takeaway: I need to book my next vacation before Claude tells me that my current YTD spend is outside my budget and decide not to serve up luxury hotel options for me.

Alex

P.S. Don't know what an MCP is and why you should care, reply and tell me. The other shoe may drop for you.