AI Reflections
In the history of grand pronouncements this will be a big one. There is not much money to be made in the AI space. This assertion is true for most of the vendors offering point solutions or implementation heavy options. I will explain more on these later. That said, most of what makes up the AI ecosystem today are rushed efforts by half baked AI Tech Startups resulting in a pile of rubbish when they fold. The noise surrounding the proliferation of “AI” powered services is only eclipsed by the proclamations of the arrival of Artificial General Intelligence (AGI) by 2030. Let us hope that on that fateful day we see a resurgence of taste and a preference for analog mediums instead of a sea of generative content.
Doomsday aside, I came to these observations after having been in the trenches for years at AI adjacent and AI first SaaS companies; along with discussions with friends in the same space. I am not judging the validity of AI as a valuable and necessary service within a company or for individual workflows. I am talking about the hype and investment in every AI idea out there even when the pitch does not come with a working demo. To further punch a hole in the AI bubble, Apple released a white paper titled The Illusion of Thinking: Understanding the Strengths and Limitations of Reasoning Models via the Lens of Problem Complexity. This was an excellent thought piece that will go unread by most people who want the hype to stay alive.
A look through any AI newsletter like the Voice AI Newsletter will be littered with announcement after announcement of the latest AI vendor who has raised tens or even hundreds of millions of dollars. Sometimes without even a working product. All of that noise would make you think that the everyone involved in AI is minting money without even trying. After two yeas working for a Voice AI SaaS company, I can tell you that it is not all rosy behind the scenes. My observations, customer experiences, and graveyard of failed AI offerings has led me to the belief that there are only three good models for AI solutions right now.
The Services Companies
Think OpenAI, Anthropic, Gemini, DeepSeek.
All of these are the AT&T / Verizon businesses of the AI era. They provide a self service model with individual and business pricing and ready to use interfaces. They also offer APIs that most Point Solutions are built on top of.
There is real money in this approach and the success of these services is directly correlated to user adoption.
The average private user can start leveraging these services with little or no barrier, and businesses can experiment on these without committing the farm.
The Point Solutions
The majority of these are built off of the available APIs referenced in the first model. Resulting in a messy AI Landscape picture that resembles the abomination that is the Martech landscape.
A thousand and one also run competitors grabbing for slivers of a large pie. Stalling out at $30 to $50 million in revenue. Cursor being the pleasant exception to the rule.
These AI solutions mostly get hung up on their requirement for first level access to data and root actions (via API) across your enterprises systems.
Enterprises face deploying a number of these point solutions which don’t often talk to each other and solve narrow use cases. Some of which you can build yourself with a tool like the aforementioned Cursor; I have another blog post coming with an example of how I did just that.
There will be massive contraction here when investors see the momentum slow down. Owing to the fact that a majority of potential customers have bad data design and integration gaps. Meaning a number of these AI point solutions can never really provide meaningful value.
The Value Added Features
In this approach existing services like your CRM, email platform, billing platform, and others are adding an AI feature that enriches and improves performance of existing functionality.
The companies using these SaaS solutions don’t have to deploy anything additional and usually pay fees for the AI features as part of a tiered product offering.
Many of these AI enabled features leverage the APIs that the main service providers offer but the operations happen within their own walled garden on top of workflows that don’t need additional integrations or data access like the point solutions.
Every SaaS provider has to add AI enabled features to their existing services in the next 2 years or they will be replaced.
More broadly speaking, my reflections have shown me that the best use of AI is when it enriches data, speeds up workflows, and connects the dots across a customer journey or user experience. Enterprises need to think carefully about who they are bringing into their ecosystem and the point solutions are not it. Instead, enterprises should leverage a combination of build and buy when deploying AI as a service layer across the whole business. The buy motion should be focused on evaluating the third-party services in your enterprise architecture for how they leverage AI as a value added feature. Companies should avoid point solutions that are require deep integrations across various services and root level access.
The build motion should be an approach that pulls the data in from all of the various services in your enterprise architecture into a single store and runs continuous analysis against these observations. Serving up next best actions/recommendations/workflows as a service that is surfaced to all your internal teams and systems. This method is wholly controlled by the enterprise in a secure fashion. You are giving full data access and actionable abilities to an AI Agent in your control. This will be the harder approach to achieve. You need seasoned prompt and ML engineers, data scientists, and tight coordination across the enterprise to make this happen. An AI Technical Program Manager with real agency and resources can drive this model.
My last reflection is more personal. My experience at a Voice AI SaaS company resulted in a deep refresh of the lenses through which I analyze where to work. Three key factors have been critical in the way I evaluate where I want to work. These perspectives came from the angle of what didn’t work where I was and what must be present in the DNA of a company that I want to work at. Simply put, I am looking for companies that are rooted in:
Product Lead Growth - two thirds of interest and/or revenue requires no sales interaction.
Self Service First - customers can “go-live” after providing a credit card and with no professional services engagement.
Rapid Product Innovation - the company moves fast and iteratively to deploy features on a monthly basis.
There are of course other nuances to consider such as workplace culture and the viability of a company to remain a going concern. These variables aside, I cannot stress enough that the above criteria are foundational for me now. Especially in a world where more and more apps and services will be built on tools like Cursor making markets more cut throat and prone to economic bubbles. That’s it, that is my AI talk for today. More later on this journey.
