I want to buy a meal

Imagine a world where at noon, people go out and buy food in different shops.

One shop gives them rice, another gives them gives them chicken, another gives them veggetables. And each shop requires tipping.

In this world, fully assembled meals are not a thing. Because people want the freedom to assemble their own meals, and most shops think it's too hard. In fact, it's known to be more profitable to focus on one ingredient. Because you're more easily identifiable.

There are shops that sell fully assembled meals, but they're not as good as the ones that sell the ingredients because the ones selling ingredients individually are specialized in their craft.

In this world, people spend a lot of time picking the best shops. And when they change jobs, they have to learn again which shops are the best so they can have the best meal possible at noon.

And there are new shops opening every day, so it's really hard to keep up and now what's the combination.

Some companies even specialized in facilitating the overall process: they help you find the best, in exchange for a fee, because after all, they need to make a living.

Some shops decide to partner and offer a discount if you use both of them.

In this world, some folks eventually decided to cook their own meals and bring it at noon, but other folks don't want to bother cooking, or they don't have the time.

This world is ridiculous, right? It makes no sense. Why would one meal shops not be wildly more profitable? Why would people bother with this?

Well, we actually live in a world like this. Except that the shops are software vendors, the ingredients are vertical SaaS products, the buyers and sellers are businesses, the brokers are integration companies and the tips are the SSO tax.

In our world, CTOs and CSOs suffer from the variety of SaaS every day: their collaborators are interested in buying the latest and greatest products, but they know that each company will be one more SSO tax to pay, one more and one more potential data breach they could have.

Small startups also suffer from this, but for other reasons: wheyn they get started they want to move fast, they pick the best in class SaaS for each category.

They sign customers, everything is great. Then they have to scale and automate their business processes. And they realize they need to consolidate their data into one place before automating these processes.

They hire people to do this, and before they know it, their IT stack becomes a unique patchwerk of technologies that no one else built in the past.

They become unique. And they pay the price for it: they can't easily find people to maintain their stack.

Oh, and in the meantime, each vendor bills them monthly, and they're trying to become SOC2 compliant, but they can't because 50% of their vendors are companies that are only 2 years old and they don't specifically need to be SOC2 to sell. They just focus on delighting end users. After all, CROs, Account executives, SDRs, marketing, folks, they don't really care about being SOC2 as long as the software gets the job done. They have a quota to fill. "The CTOs and CSOs are the problems, they just slow everyone down and prevent us from buying the latest and greatest, we have quotas but you don't let us move fast".

And here is the cherry on top: AI.

AI just doesn't work with this stack. Not only do you need to consolidate your data into one place, but you also need to create what some folks these days call an action layer: a piece of software that can act on behalf of other software (think of it like a giant unified API)

I built this type of action layers for the last 6 months. And if you're doing it in house for your own company, sure, why not.

But if you try to commercialize an AI product that spans accross the whole IT stack, like I wanted to do? It's an absolute nightmare.

Nightmare in terms of complexity, security, and cost. Cost of people, but also cost of all the vendors you're going to need to buy. Oh, and some of them bill you premium just to use their APIs, or webhooks, or unlock the rate limits you need to operate.

Let's say you go through all this, like I did. Your customers don't even use vendors the same way. Some of them just don't follow the common practices: they use leads and contacts improperly on salesforce, they don't know how to double sync Salesforce/Hubspot properly, or they don't even connect their calendar to Salesforce properly.

AI in SaaS is all about results. It's not a game of giving a toy to companies and train them if they fail to use it.

But in an IT stack made of 50 different products, you just can't guarantee the success of your customers. And they churn. And you can't fix the churn, because your success is not dependent on your product, it's dependent on their IT stack.

So most folks just end up building AI features for the vertical they own, and they don't bother with the rest. Our AI use cases remain shallow because of this.

For example, a holistic AI agent that knows how a customer got acquired, uses this knowledge to streamline a customer support call, and after listening to a this customer call, decides to amplify this specific demographic and run an outbound campaign. This is impossible to have today.

You can only have the customer support one, or the outbound one. They don't communicate, unless you make them communicate. "But it's ok, I would rather have them specialize". Sure, but you're going to have to integrate them. And they might not communicate between one another.

The TLDR is that the best AI use case span accross our whole IT stack, not one vertical.

I hear so many folks building AI in B2B contexts. I recognize the ones that are really building AI agents because in general we both agree that the biggest bottleneck for AI to disrupt our work is integrations.

The most experienced "B2B AI builders" also know the fallacies of the commonly proposed solutions for this problem: unified APIs (remember the shop brokers?) and why they don't work for very specific SOPs. They're just too generic and solving an 80/20 problem. You always end up needing their passthrough offering (google "passthrough unified api") and eventually, out of frustration, you build your own unified API.

Let's come back to the food analogy: I really don't want to buy ingredients and combine them together. I want to buy a MEAL.