Last month I wrote about the SaaSpocalypse. The structural collapse of SaaS valuations. The distressed assets piling up. The shopping spree.
I got a lot of responses. Most of them were some version of the same question: "So is software dead?"
No. Software is not dead. The screen is just losing its audience.
There have only been three moments in modern history where an asset class got valued this aggressively and then corrected this hard.
Tulips. Circa 1636. 55x.
Dot-com stocks. Circa 2000. 45x.
SaaS. Circa 2020. 20x.
People look at this correction and blame AI. That is wrong. SaaS companies were trading at 20 times revenue. That number was never real. It was a bet that growth would last forever and margins would expand forever. Neither happened. AI accelerated the correction. But the correction was gravity.
Here is what is actually happening.
The software still works. CRMs still hold customer data. ERPs still run operations. The products do what they have always done. What is changing is where people spend their time.
People's eyeballs are moving into the LLMs. That is where they ask questions. That is where they give instructions. That is where they make decisions. Less and less time is spent inside the actual software.
And it goes further than that. People are not just spending less time in software. They are starting to send their AI agents into the software on their behalf.
Nobody wants to update a CRM. Nobody ever did. We just did not have a choice. AI is faster, more accurate, more consistent. It does not forget to log a call. It does not wait until Friday to update the pipeline. So the human does the human part of the job. The conversation. The relationship. The judgment call. And the AI handles the rest.
This is not theory. We built a product at Trilogy called CRM Magic two years ago that did exactly this. Took the data, gave AI access, let it work on behalf of the user. Nobody was ready for it. We were too early. Now we are rebuilding it because customers are finally asking for it. The market caught up. We are building it for our own business as well.
Here is what most companies are missing about how to survive this.
If your product's value depends on a human sitting in front of your screen, you need to think hard about the next two to five years. That human is not leaving your product. But they are spending less and less time looking at it. Their eyeballs are in the LLM. Their AI agent is in your product doing the work.
This means your product needs to be ready for two things. First, it needs to be easy for AI to use on a human's behalf. Second, it needs to be accessible from inside whatever LLM your customer lives in. You want your product at their fingertips without them ever opening a new tab.
I have written before about the Business Body framework. The way I think about AI architecture inside an organization. It maps perfectly to what software companies need to build right now.
It starts with the Skeleton. The data model. Data is the skeleton of AI. There is so much data out there and without structure, none of it is useful. Without clean, well-organized data, nothing else works. Most software companies already have this. It is their biggest asset. They just do not treat it that way.
Next is the Circulatory System. The context layer. This is where most companies fall apart. You need AI to understand not just what data you have, but what it means. What does "closed-won" mean in your system? What is a "deal"? What does "at risk" look like? In the technical world, people call this an ontology. Think of it as a dictionary that your product and AI agree on so they speak the same language. Without it, AI is just guessing. The context layer is the piece that touches everything and makes the whole body work together. It is a combination of those ontologies and a deep understanding of what the data actually represents.
Then you need to build the Muscles. The functional tools that let AI actually do things inside your product. The technical terms here are MCPs (Model Context Protocols) and CLIs (command line interfaces). In plain terms, an MCP is a plug that lets any AI model connect to your product and take actions. Think of it like a universal adapter. Right now your AI can only use your product if it can see the screen. An MCP lets it walk in through a side door and get work done directly. A CLI is similar. It lets someone (or something) operate your product by typing commands instead of clicking buttons. Think of both as new doors into your product. Right now, most software only has one door: the screen. You need doors that AI can walk through.
The Central Nervous System. The coordination layer that makes sure all these pieces work together across functions. Not just inside one department. Across the whole organization.
The Brain. Human judgment. The part that decides what to do with what AI surfaces. This never goes away.
And the Immune System. The safeguards. The checks. The thing that catches mistakes before they reach the customer.
Software companies that build this body will thrive. The ones that keep betting everything on the screen will watch their users quietly migrate to the LLM and send agents back to do the clicking.
Here is what excites me as someone who has spent years as a product leader.
Innovation in platforms is about to accelerate in a way we have never seen. What used to take a year to build will get done in a quarter or month for those that are really advanced. Don’t believe me, look at the pace of releases at Anthropic. The bottleneck was never the idea. It was the build. AI compresses the build cycle so much that the rate of evolution is about to look nothing like the last decade.
I do not believe AI-native startups will unseat the incumbents. Not even close replacements. Nobody has that distribution power right now. The incumbents have the data, the integrations, the enterprise contracts, the switching costs. What happens is the incumbents either adapt or slowly become invisible as AI routes around their clunky interfaces.
There is one more shift worth watching. If fewer humans are sitting inside your software, seat-based licensing starts to break. You have 500 seats but only 50 humans actually open the product. The other 450 "users" are AI agents. Nobody wants to pay per-seat for that. But the product is still delivering value. Maybe more value than before. The market is going to have to figure out what comes next. Results-based pricing. Pay for outcomes, not chairs. Every time the product delivers a result, that is the value. That shift is going to be messy and it is going to scare a lot of people. It is still being figured out. We will save that for its own newsletter.
The point is this. Software is not dying. The screen is not dying. But the screen is losing the audience it was built for. The eyeballs are moving to the LLMs. The AI agents are showing up to do the work. And the companies that build the full body, skeleton to immune system, will be the ones still standing in five years.
If you have not mapped the Body framework to your own company yet, start with the Skeleton. Get the data right. Then build the context layer. Everything else follows from there.
Kathy
THE AI BOSS