Last week my team demoed a dashboard to one of our enterprise customers. Their data, our platform. It took fifteen minutes to build.

The marketing ops leader on the call looked at it and asked the question every software founder prays never comes up:

I said: yeah, you could.

There were eight people on that call. Nobody spoke for a second and that is not the answer a vendor is supposed to give.

But it was the right answer. And those ten minutes taught me more about where software is headed than anything I’ve read this year. They also showed me something uncomfortable about where marketing teams are headed. And that’s why I’m writing this.

Ten years ago, build-versus-buy meant a VP saying "our engineers could build this." Everyone in the room knew it was theater. Engineering time was scarce, the roadmap was full, and the build was never going to happen.

That objection is dead. The new one is "I, personally, a marketing ops manager, could build this tonight." And it is not theater.

Retool surveyed 817 builders for its 2026 build-versus-buy report. 35% of teams have already replaced at least one SaaS tool with something they built themselves. 78% plan to build more this year. And the third most vulnerable category, at 29%? BI and analytics.

Now look at who is doing the building. Only 36% were engineers. The rest were operations, product, data, finance, marketing ops. And of the people building outside IT's oversight, 64% were senior managers or above.

Your directors are the shadow IT department now.

The plumbing caught up this summer too. In July, Model Context Protocol, the standard that lets AI tools plug directly into business systems, shipped centralized enterprise authorization with Anthropic, Microsoft, and Okta behind it. The "I could just do this myself" path stopped being an experiment on someone's laptop. It is becoming sanctioned, governed infrastructure.

So when she asked her question, she wasn't being difficult. She was being current. Which is exactly why I couldn't dodge it.

What I told her on the call comes down to two things, and I think these same two things will decide which software companies are still around in three years.

The first: go ahead, prompt Claude Code tonight. You'll get a beautiful dashboard. It will be perfectly correct at the moment it's created, and it starts dying from that second on. It doesn't refresh. Nobody else can log into it. No permissions, no audit trail, no shared version of the truth. You built an artifact, not a system. What we actually sell is everything around the dashboard: the thing that keeps it alive, current, and in front of ten people who all see the same number.

The second one matters more. Her fifteen-minute prompt only worked because of what it landed on. Underneath that dashboard is a governed data model (RevSure's) where identities are resolved and CRM, marketing automation, web, and product touches have already been stitched into one picture of the buyer.

Point the same prompt at raw Salesforce exports and you still get a dashboard in fifteen minutes. It will just be confidently wrong.

AI made the interface free. And once the interface became free, the value collapsed to the two ends of the stack: the context underneath it and the harness around it. Everything in between is a commodity now.

The moat was never the dashboard or even the AI models. It's what the AI is standing on.

A century ago a good map was a valuable product. Surveyors, cartographers, engravers. The interface to geography was expensive, so you paid for the interface.

Today anyone can produce a map in seconds. Did mapping die as a business? It became one of the most valuable businesses on earth. The value just moved. Google maps doesn't win because its map is prettier. It wins because the map knows where the traffic is right now, and because the same living map sits in two billion pockets, refreshing constantly.

A dashboard built by an AI in one session is a printed map. Exactly right at the moment of printing. Wrong by morning. And only in the hands of the person who printed it.

Your ops team can print maps now. All of them, whenever they want, free. So the question, for your vendors and for you, is no longer who can draw the map. It's who owns the traffic data, and who keeps the map alive.

Few days before that call, Gartner published a figure: $234 billion of enterprise software spending, about 20% of the market, is exposed to what they call "agentic arbitrage" by 2030. Their analyst George Brocklehurst compressed it into one sentence: agentic systems deliver outcomes directly, "bypassing traditional UX–heavy applications and making the software invisible."

The market had already voted on this once, badly. Software stocks lost a trillion dollars of value during the early-2026 AI panic, as investors feared AI agents could eat into traditional SaaS. By June, a sharp rebound suggested the market was beginning to distinguish between software that AI could displace and software that AI could make more valuable.

Read Gartner's survival advice for vendors closely, because it's the whole game in one line: capture and retain customer- specific knowledge, not just raw data.

I've been calling that the context layer for two years. So none of it comes as a surprise when we watch a customer force it out of me on a call and an analyst firm price it at $234 billion in the same month.

This is the part of the story vendors usually leave out. She didn't agree with me. She looked at the dashboard we'd just built and pointed at the labels.

Every metric used our terminology, not hers.

"Until it matches our terminology, I probably wouldn't deploy this to our stakeholders. This is going to cause a lot of confusion."

I pushed back. We're a product company, not a services company. We can't maintain every customer's internal vocabulary forever. At some point, we're just doing their analytics for them.

I still think that's true.

While the dashboard took fifteen minutes to build, getting it ready for her company was a different matter.

What does "pipeline" mean to them? Which stages count? What does "created this quarter" mean in their reporting? We didn't know.

The most important context wasn't in our data model. It was in her head. Then she said something which is the most important part of our discussion:

That question gets more important as analysis gets cheaper. When analysis took time and money, you had to be selective. You asked for the things that mattered.

Now you can get a perfectly good answer in minutes and still have no idea what to do with it.

The hard part used to be finding the insight. Now it's deciding what to do with it.

Here is why I'm telling this story to CMOs and CROs instead of founders.

Agentic arbitrage doesn't check whether the interface it's bypassing was sold by a vendor or produced by your team.

So the value of your function collapses to the same two ends of the stack that the value of your vendors did.

Your context: the metric definitions, the institutional memory, the knowledge of why the Q2 number looks weird and which board member will ask about it.

Your harness: the discipline that keeps one version of the truth current and in front of the right people, and the judgment that turns a finding into a decision with an owner and a date.

Ask her question, pointed at your own team. If an agent could rebuild every deliverable my function produces, tonight, free, what would remain?

If the answer is "our definitions, our memory, and our judgment," the machines just made you stronger, because your deliverables became free to produce. If the answer is nothing, that is not the machine's fault.

When she asked me : the insights are good, what do I do with this? This is how I answered her.

The trillion-dollar software panic and the question on my customer call were the same event at two different scales. Both were the market discovering that nobody ever paid for the interface. They paid for the truth underneath it, and for the discipline that keeps it alive.

Own those two, and every "fifteen-minute dashboard" in the world makes you stronger.