AI agents are starting to eat SaaS

There is no yes and no. This is a fact. Even a small startup of 3-5 people will pay more in terms of salaries than the total license costs they consume. A larger enterprise will will spend 50 to 100 times more on salaries then software license fees.

The reason software licenses are easier to cut by the finance team when things are not going well is because software does not have feelings although we all know that this not making a dent. Ultimately software scales much better than people and if the software is "thinking" it will scale infinitely better.

Building it all in house will only happen for 2 reasons: 1. The problem is so specific that this is the only variable option and the quickest (fear enough). 2. Developers and management do not have real understanding of software costs.

Developers not understanding the real costs should be forgiven because most of them are never in position to make these type of decisions - i.e they are not trained. However a manager / executive not understanding this is sign of lack of experience. You really need to try to build a few medium-sized none essential software systems in-house to get an idea how bad this can get and what a waste of time and money it really is - resources you could have spent elsewhere to effect the bottom the real bottomline.

Also the lines of code that are written do not scale linearly with team sizes. The more code you produce the bigger the problem - even with AI.

Ultimately a company wants to write as few line of code as possible that extract as much value as feasibly possible.

I'm seeing the opposite. AI is actually increasing the demand for what would previously be too expensive, bespoke integrations and solutions. Those are now becoming more feasible and doable. There is also the notion that a lot of companies are actually very behind on embracing software or SAAS. Especially in manufacturing it's common to see operations that haven't materially changed anything in decades.

The fallacy here is believing we already had all the software we were going to use and that AI is now eliminating 90% of the work of creating that. The reality is inverted, we only had a fraction of the software that is now becoming possible and we'll be busy using our new AI tools to create absolutely massive amounts of it over the next years. The ambition level got raised quite a bit recently and that is starting to generate work that can only be done with the support of AI (or an absolutely massive old school development budget).

It's going to require different skills and probably involve a lot more domain experts picking up easy to use AI tools to do things themselves that they previously would have needed specialized programmers for. You get to skip that partially. But you still need to know what you are doing before you can ask for sensible things to get done. Especially when things are mission critical, you kind of want to know stuff works properly and that there's no million $ mistakes lurking anywhere.

Our typical customers would need help with all of that. The amount of times I've had to deal with a customer that had vibe coded anything by themselves remains zero. Just not a thing in the industry. Most of them are still juggling spreadsheets and ERP systems.

I bet companies are circumventing this in a way that allows them to derive almost all the benefit from your data, yet makes it very hard to build a case against them.

For example, in RL, you have a train set, and a test set, which the model never sees, but is used to validate it - why not put proprietary data in the test set?

I'm pretty sure 99% of ML engineers would say this would constitute training on your data, but this is an argument you could drag out in courts forever.

Or alternatively - it's easier to ask for forgiveness than permission.

I've recently had an apocalyptic vision, that one day we'll wake up, an find that AI companies have produced an AI copy of every piece of software in existence - AI Windows, AI Office, AI Photoshop etc.

> IMO As soon as the bubble bursts, which is already evident and imminent, these agents will raise their subscription fees to ridiculous amounts. And when that happens, entire organizations will abandon them and return to good ol' human engineering

It's worse than you state; a primary premise of the current AI expansion and investment, continuously repeated, is that computational resources are increasing for the same price-point.

So tell me, when we have these hardware gains in 5 years, why the fuck would I pay for fractionally better output on a cloud subscription when I could run on-prem GPUs for a fraction of what the actual subscription would be, giving me 24x7 agents working with no limit?

Current highly-invested AI providers are token providers - they are sitting at the bottom of the value-chain. They are trying to climb the value-chain, but the real value is in the models, and since all models have mostly converged on the same performance, running your own gives a very tiny drop in value.

Their problem is that, even at $200/m, it is not feasible to offer 24x7 access to the models - that max subscription has tiny limits: on that subscription you most definitely are not enabling even a single agent to run f/time.

You might get 3 hours per day, maybe. Best case, you get half a working day (4 hours), for five days. So double the subscription, and you're still looking at only a single simply agent that you can run f/time for $400/m.

For collaborating agents, you'll have, what, 4 - 5 diving into a large task in parallel? That's maybe $1600/m - $2000/m to solve tasks. You still have to pay the (increasingly incompetent) human operator to manage them.

That sort of workflow is what is being sold as the vision, and yet that will only be economically viable if you're self-hosting your own model.

> Domain expertise + tight feedback loop

This is the answer to a happy B2B SaaS implementation. It doesn't matter what tools you use as long as this can be achieved.

In the domain of banking front/back office LOB apps, if you aren't iterating with your customer at least once per business day, you are definitely falling behind your competition. I've got semi-retired bankers insisting that live edits in production need to be a fundamental product feature. They used to be terrified of this. Once they get a taste of proper speed it's like blood in the water. I'm getting pushback on live reloads taking more than a few seconds now.

Achieving this kind of outcome is usually more of a meat space problem than it is a technology problem. Most customers can't go as fast as you. But the customer should always be the bottleneck. We've done things like install our people as temporary employees to get jobs done faster.