Not really a tech guy, so I’m hoping someone familiar with enterprise AI cost structures can help me out.
I’m currently on Claude’s $200 plan, mainly for coding and research. I regularly hit the token limit and have to wait 4 hours for the next session to reset.
Let’s say an HR department is currently running 3–4 separate software tools (ERP, payroll, timesheets, etc.). How many tokens would it roughly take for AI to fully handle those same tasks?
At what point does the cost of using AI become so much higher than running standalone SaaS tools that the traditional software ends up being the cheaper (and better) choice? In other words, will AI actually be able to replace SaaS in this kind of scenario, or will the token costs make it impractical?
The cost of replacing SAAS with AI
byu/Far-East-locker instocks
Posted by Far-East-locker
11 Comments
No AI will not be replacing SaaS any time soon. It might augment it, but replacement is absurd.
Unless the saas you’re using is a web API that tells you whether a number is odd, it’s absolutely impractical.
You’re not paying for a few lines of code, you’re paying for proven solutions, for access to marketplaces with other companies, for compliance and security, for not having to maintain a product in an area where you have no domain knowledge and don’t even understand the complexity under the hood, for paying only a fraction of what your own infra might cost and for someone to yell at over the phone when shit breaks.
You can already get a lot of these software tools for free, via open source. If they didn’t use that, why do you think companies opted for a paid version instead? Apply that same logic to some AI tool instead and what do you think will happen?
They’d use APIs and not the Claude apps. As such, they have more flexibility in addressing quotas, but at a cost.
Either way, gen/agentic AI doesn’t replace SaaS, nor is anyone going and building a replacement for *any* SaaS product over a weekend.
The individuals saying they are and the analysts buying into / pumping this shit have zero business speaking.
Source: software engineering leader and practitioner of 30+ years and heavy user of all of these tools. Also, I evaluate these fuck all vendors on the daily.
So enterprise did this exact thing for a few decades – boutique homegrown business software – and it was kind of an unmitigated disaster. People retired, knowledge bases moved on, requirements changed. It was a pain in the ass to maintain on a micro level.
The other issue with Claude replacing companies like door dash or PayPal is to build the network. Find the drivers etc.
Been working with a client going through something along these lines recently.
I advised them to use Claude Code to design custom tools/web apps tailored to their exact use case (stuff they currently rely on a suite of SaaS tools for) then deploy on their own AWS instance.
Once the thing is built, it won’t be running on tokens.
Maybe Cowork is capable of this stuff? I dunno, I haven’t really dug into that side yet.
Saas not going anywhere. Enterprise level application need compliance, sox, security etc. But, it can lower down barrier of entry to have new company to create brand new application/saas for sme that don’t need all those things
Vibe coded patchwork solutions will be implemented by small companies. No medium to large enterprise will be replacing major systems with vibe coded apps. They do not have the appetite for that risk and need to focus their energies to their core businesses not maintaining large vibe coded systems. Even if the models get better this will not be the case.
The real risk to SaaS is shrinking headcount as say 25% of white collar workers are made redundant.
Your question is a bit ambiguous so I would try to answer it for both cases.
If you expect to replace HR (human) and add agents then you would have a recurring cost (similar to a clawed bot).
For actually replacing saas tools, it would require you to hire developers to build and maintain those tools (even if you use AI to build it).
All this will really depend upon what your needs are and do you have the budget and buy in to do such a thing. Ie it have to save you non trivial amount of money to do that.
Created my own “capacity planner”. Add resources, places blocks on a timeline, calculate capacity, create a timeline/roadmap. Only took a few hours.
Something like Jira’s resource planning tool or many other similar tools already out there that you would have to buy “seats” for. Mine is obviously not as robust as those, but it does what our team needs it to do. We were previously just building something in a Google spreadsheet.
I could see this being better than spending money subscribing to those tools that you only use 20% of and you can scale it exactly for your proprietary needs and connect it with your other proprietary tools.
I am not an engineer.