I have been building an Excel file for weeks that runs our P&L. It tracks how Gadoci Consulting is actually performing. Recently I started extending it into a partnership model for the firm, and the file started straining against what a spreadsheet can hold. So tonight I opened Claude Code and started building the app version of it. The kind of build I have done dozens of times over the last twenty years.
The app was going beautifully. Three hours in, the old reflex fired.
This is good. I could make it multi-tenant. Add a signup flow, billing, a dashboard. Other operators would pay for this. I should productize it.
I have had that exact thought reliably since 2005. Tonight, for the first time, I dismissed it on the spot.
The reasoning that came up was new. Whoever needs this app is going to build it themselves with Claude in a weekend. There are already ten of these out there, I am sure, and a hundred more people about to ship another. The thing that is actually scarce right now is not a financial-modeling app. It is the ability to take somebody through the process I just went through and not productize the result. It is teaching them how to think through their problems, use AI to build the tools that solve those problems, and walk away owning the work themselves, instead of buying somebody else's SaaS to do the solutioning for them. That is what I want to be paid to do.
What I felt was the founder reflex of an entire generation, the one that came up between 2005 and 2025, click off in real time. It happened on a Wednesday night, in my office, alone. Then I looked at what had been crossing my screen for the last six weeks and realized the rest of the market is arriving at the same conclusion at the same time.
The Reflex We All Learned
Between 2005 and 2025, the default move for an entrepreneur was clear. See something you needed, build it, and the moment it worked, ask whether it could become a public product. That move built Salesforce, Shopify, Stripe, Figma, Notion, and roughly half the unicorns of the last twenty years. If you saw a pain point and you had any kind of building instinct, you reached for code, and then you reached for a landing page.
The product was the unit of entrepreneurial value. The reflex was so deeply trained that it ran on its own. You did not even decide to have it.
My Network Shows Me This Every Day
In the last twenty-four hours, three different people in my network have shown me a new product they built with Claude Code. All three are good. None of them needed to raise a round or hire an engineer. They just shipped.
I have been seeing this pattern for months. Friends, clients, operators in adjacent firms, all quietly producing software that two years ago would have been a serious engineering project. The reflex they are running is the same reflex I caught running in my own head tonight. The supply of perfectly good, AI-built products is now growing faster than the demand for any single one of them.
The Convergence Showing Up Everywhere
Inside the last six weeks, five different voices have made roughly the same argument from five different vantage points.
Y Combinator's request for AI-Native Service Companies says it most directly. The total spend on services is many times larger than the total spend on software, and most of those services are already outsourced. YC is not excited about companies that sell tools to help you do the work. They are excited about companies that just do the work.
Sequoia is making the same case from the venture seat. Julien Bek's "Services: The New Software" thesis lays out the arithmetic. For every dollar spent on software, six go to services. Service companies trade at one to two times revenue. Software trades at eight to twelve. Run a service business with software-margin economics and you have built one of the more lopsided arbitrages in tech.
Mark Cuban and Kevin O'Leary, neither of them venture investors, are making the same point to a different audience. I wrote about that thread here. Cuban: "Regular software is over. Everything will be made just for how your business actually works. Who's going to set that up for them?"
Anthropic and OpenAI are now standing up PE-backed consulting arms to deploy AI inside customer organizations. When the model builders themselves decide they need humans on the ground to make AI useful, that is the strongest possible signal that the gap between the model and the work is bigger than the gap between the model and the next model.
And Raphaël Dabadie at Foaster (YC P26) is making the sharpest version of the case from the consulting side. AI adoption is tools layered on the existing company. AI transformation is redesigning the company around what AI now makes possible. Their pitch is software-plus-service. Agents map the operational graph at scale. Humans handle the judgment.
Why the Reflex Stopped Working
Three things changed at once.
The cost of building a product fell to almost zero for anyone fluent with Claude Code or Codex. I made this point from a different angle in I Almost Built This in n8n. What used to take a small team a quarter now takes one fluent operator a long evening. Every product idea you have, fifty other people are also having, and most of them will ship faster than you will.
The willingness to pay for tools is structurally weakening at the same time. When the buyer can spin up their own version of your product over a weekend, the moat is no longer the code. It is the distribution, the credibility, the hand-holding, and the operator who actually shows up. When the interface compresses to a single agent, customers start asking what they are paying the SaaS company for, exactly.
And customers are now saying out loud that they do not want more tools. They want the work done. Companies have already bought the licenses. They have ChatGPT, Claude, Copilot, Gemini. What they do not have is anyone who can sit with them and turn those tools into operational reality. The 95 percent pilot-failure stat that keeps showing up in industry research is a services gap, not a tools gap.
The New Reflex
The version of the entrepreneurial muscle that used to fire said: I built something for myself. I should make it available to everybody. The version that should fire instead says: I have expertise and execution that other people need. How do I offer that as a service, and build software to support it when the support actually makes the service better?
The product does not vanish in this version. It changes role. It stops being the thing you sell and starts being the thing that lets you deliver the service better. Vertical SaaS in regulated markets is still a real business. A solo operator running a five-thousand-dollar-per-month micro-SaaS is still a real life. The claim is not that the product category disappeared. The claim is that the default reflex has flipped.
This is the shape Gadoci Consulting settled into eight months ago. AI Embed puts one of our people inside your team for a few days, doing the actual work alongside them. AI Operations runs the long-term partnership that turns AI tools into operational reality. AI Education is a service. We deliver the sessions. We do build software. We have a product called Solutioner.ai, the AI tools catalog and operator workspace our team uses to run engagements better. We deliberately do not sell it to the public. It earns its keep by making the service sharper, not by chasing public SaaS revenue.
The sharper version of the thesis is the one I felt fire tonight. The new entrepreneurial play might not be doing the work for the customer in perpetuity. It might be teaching the customer how to do the work themselves, with bespoke software supporting that teaching where the support makes the teaching better. If the cost of building a product has fallen to almost zero, the most valuable service in the world might be teaching another company how to build the products it actually needs, on its own, fast enough that it never has to buy somebody else's.
The Quiet Part
I caught the old reflex tonight. The financial-modeling app I started building three hours ago is going to keep getting built. It is going to make my service work better. It is going to stay an internal tool. The work it supports, sitting with partners, walking them through how to think about an AI Ops engagement, running the model with them, and helping them learn to run that math themselves, is the part that is actually scarce, and the part I want to be paid to do.
Twenty years of reflex stopped firing in my office on a Wednesday night. I noticed. I think a lot of operators are about to notice the same thing.