I think one of the interesting things here is that AI doesn't need to be able build B2B SaaS to kill it. So much of the overhead of B2B SaaS companies is thinking about multitenancy, intergrating with many auth providers and mapping those concepts to the program's user system, juggling 100 features when any given customer only needs 10 of them, creating PLG upsell flows to optimize conversions, instrumenting A/B tests etc...
A given company or enterprise does not have to vibe code all this, they just need to make the 10 features with the SLA they actually care about, directly driven off the systems they care about integrating with. And that new, tight, piece of software ends up being much more fit for purpose with full control of new features given to company deploying it. While this was always the case (buy vs build), AI changes the CapEx/OpEX for the build case.
And in many cases, it's 12 features, with 2 of the features not even existing in the big SaaS.
I'm pretty sure every developer who has dealt with janky workflows in products like Jira has planned out their own version that fits like a glove, "if only I had more time".
Pretty much. My employer was looking to cut costs and they were spending ~500k a year on a product that does little more than map entra roles/groups to datasets and integrated with a federated query engine through a plugin. Took a couple days to build a replacement. The product had only a few features we needed.
Maybe it's mostly from AI, maybe it's mostly general economic cutbacks. I also feel like these "wrapper" style SaaS products are the first ones companies are dropping when they are looking to cut costs, and I think a lot of companies are looking to cut costs. I do agree with the overall conclusion either way, that System of Record products/companies are the most likely to survive. There are a lot of SaaS companies with questionable long-term businesses who are getting hit, but that was bound to happen.
The stocks of a lot of these SaaS companies were priced on the expectation that they could become the next IBM: become entrenched with the customer and then hike prices until their eyes bleed.
A lot of companies have been too smart for that, and a lot of SaaS offerings are too small to be truly entrenched. Arguably the investment horizon is too short (IBM took decades getting to that point).
The only real vendors who managed to become the next IBM are the cloud providers.
Analytical systems. I see a lot of add-on services that will add intelligence/analytics/etc and companies try them out to solve some issue they have and bounce off them frequently due to growing costs. I can only assume as mentioned that over time these are also easier for companies to in-house vibe-code as well, I just haven't seen a ton of that yet, but people are definitely trying which still shrinks the available pie.
> How to keep asking customers for renewal, when every customer feels they can get something better built with vibe-coded AI products?
Wrong take. You don't need to build something better, you only need something good enough that matches what you actually need. Whether you build it or not and ditch the SaaS is more of an economic calculus.
Also, this isn't much about ditching the likes of Jira not even mentioning open source jira clones exists from decades.
This is more of ditching the kind of extremely-expensive-license that traps your own company and raises the price 5/10% every year. Like industrial ERP or CRM products that also require dedicated developers anyway and you spend hundreds of thousands if not millions for them. Very common, e.g. for inventory or warehouse management.
For this kind of software, and more, it makes sense to consider in-housing, especially when building prototypes with a handful of capable developers with AI can let you experiment.
I think that in the next decade the SaaS that will survive will be the evergreen office suite/teams, because you just won't get people out of powerpoint/excel/outlook, and it's cheap enough and products for which the moat is mostly tied to bureaucratic/legal issues (e.g. payrolls) and you just can't keep up with it.
Having participated in the build of an inventory system / system of record for a large national retail company, I can't see vibe coding helping anything more than the prototyping in the discovery / requirements gathering parts of the process.
The sheer volume of data, the need for real time consistency in store locations, yada yada means that bad early decisions bite hard down the road.
Lots of drudge work can be assisted by AI, especially if you need to do things like in ingest excel sheets or spit out reports, but I would run far away from anything vibe coded as hard as possible.
One of my clients spends 500k+ on XXX licensing per year (for a 200M revenue company that's not peanuts), and on top of that has to employ 12 full time XXX developers (that command high figures just for their expertise on that software while providing very little productivity) and every single feature takes months to develop anyway. Talking about stuff like adding few fields to a csv output.
So the total cost of XXX is in the 2M/year range, and it keeps ballooning.
My (4 men) team already takes care of the entire warehouse management process except inventory, the only thing that XXX provides, we literally handle everything: picking, manufacturing, packaging, shipping phase and many others.
In any case, nobody has mentioned vibe coding.
I stated that a handful of good engineers with the aid of AI in a couple of months can provide a working prototype to evaluate. In our case it's about extending our software that already does everything, except inventory management.
When you spend 2M/year on a software (1% of your revenue), growing every year by 100/150k it makes sense to experiment building a solution in house.
Its funny you mention excel, I see vibe coding in the business sense right now being a gateway to replace all of the ad hoc uses of excel. We've basically leveled up the quality of the software you can build before buying a SaaS product or a hiring an in house engineer.
I don't think it is killing SaaS. I have definitely had to extend my sales cycle when a potential customer vibe-coded a quick fix for a pain point that might have triggered a sale a few weeks earlier, but eventually the benefit delivered by someone else caring about the software as their entire mission really wins out over a feature here and there.
If you are selling SaaS consider that a vibe-coding customer is validating your feature roadmap with their own time and sweat. It's actually a very positive signal because it demonstrates how badly that product is needed. If they could vibe code a "good enough" version of something to get themselves unstuck for a week, you should be able to iterate on those features and build something even better in short order, except deployed securely and professionally.
Everyone's going to talk about how cool their custom vibe-coded CRM is until they get stuck in a failed migration.
I would assume one major thing here is that many orgs only need a small subset of functionality from what most products provide. Many times, that small subset of functionality is only "good enough" in and of itself, but the org is paying the premium for the entire suite of whatever it is. This makes realizing that an LLM can get them to MVP and beyond much easier.
Charging hundreds of thousands if not millions per year for very basic functionality is what is "killing" b2b SaaS.
It's not and I really doubt it will, for true SaaS platforms. A desktop .gif recorder (frequent example I've read about) is not a SaaS, even if you charge monthly for it.
Let's put an example an exception-tracking SaaS (Sentry, Rollbar). How do the economics of paying a few hundred bucks per month compare vs. allocating engineering resources to an in-house tracker? Think development time, infra investment, tokens, iteration, uptime, etc. And the opportunity cost of focusing on your original business instead.
One would quickly find out that the domain being replaced is far more complex and data-intensive than estimated.
There are many cases where the company might only use a fraction of the features (and therefore complexity) of the SaaS and so only need to develop and maintain those features they actually need. That's when ditching the SaaS can make sense if you can easily develop/maintain what you specifically need on your own with AI assistance.
no. High interest rates and a cautionary view of future economic growth are killing B2B SaaS. Money is no longer free, and so there is a bigger push for cost-cutting rather than growing your buisness with free money.
Sure, vibe coding has impacted user's expectations. They know you can ship a new update easier and faster than before - and you actually can.
But, not sure which successful SaaS companies just stopped shipping any updates to the product, never talked to their customers and never added any new features to win over major new accounts - and still managed to survive and thrive?
And the author actually confirms this:
> AI isn’t killing B2B SaaS. It’s killing B2B SaaS that refuses to evolve.
Focus is a currency and you have a limited amount of it, if all SaaS is built internally, teams would go bankrupt. There's likely always going to be a band of experts focused on solving a problem and everyone pays them to solve it for them, because they do it better and can handle the hassle of maintaining it.
"The SaaS model was built on a simple premise: we build it once, you pay forever."
I've never seen a SaaS product that fits this description. There are always things to do. Libraries to upgrade, performance bottlenecks to diddle around with, an endless stream of nonsense feature requests from people at the customer who never actually use the product, fun experiments your developers want to try out, and so on.
The hard part in SaaS is to delete code, and that's what you should do, at least some of the time. Either through simplifications, or just outright erasing functionality that very few if any of your customers rely on.
What you should not do is let your customers grow the liability that is code in your production environment, unless your entire product set is designed to handle things like this, e.g. the business models of Salesforce and SAP.
here's the secret saas can vibe code features too on top of their paid well developed and secured api. they can get off their ass and vibe code a mcp wrapper, so user can use the ai tooling they pay for to interact with their saas. and they'd be called visionary hero of the agentic revolution.
but they don't want to. and they will be replaced, as it's good and well.
Something notable for SaaS which this article doesn't mention is that in some cases the reason to buy rather than make yourself is due to needing to handle a bunch of different regulations which LLMs don't threaten (barring businesses which would rather have lawsuits than pay for a SaaS).
I don't think AI is killing B2B SaaS so much as companies are finally reckoning with the immense costs of SaaS in a markably different environment than when SaaS exploded in popularity, and AI offers an off-ramp to some. Let's break it down camp-by-camp to show you what I mean:
1) The must-haves. These are your email and communication systems, the things you absolutely have to have up and available at all times to do business. While previously self-hosted (Exchange/Sendmail, IRC/Skype/Jabber, CallManager/UCS), the immense costs and complexities of managing systems ultimately built on archaic, monolithic, and otherwise difficult-to-scale technologies meant that SaaS made sense from a cost and a technical perspective. Let's face it, the fact nobody really hosts their own e-mail anymore in favor of Proton/Microsoft/Google/et al shows that self-hosting is the exception here, not the norm - and they're not going anywhere regardless of how bad the economy gets. These are the "housing stock" of business, and there's plenty of cheap stock always available to setup shop in without the need for technical talent.
2) The juggernauts. The, "we can do this ourselves, but the pain will be so immense that we really don't want to". This is the area where early SaaS solutions cornered and exploded in growth (O365, ServiceNow, Google Workspaces), because managing these things yourself - while feasible, even preferable - was just too cheap to pass up having someone else wrangle on your behalf with a reasonable SLA, freeing up your tech talent for all the other stuff. The problem is that once-focused products have become huge behemoths of complex features that most customers neither need nor use on a regular basis, at least after the initial pricey integration. Add in the ease of maintainability and scalability brought by containers or microservices, along with the availability and reliability of public cloud infrastructure, and suddenly there's more businesses re-evaluating their relationships with these products in the face of ever-rising prices. With AI tooling making data exfiltration and integration easier than ever from these sorts of products, I expect businesses to start consolidating into a single source of truth instead of using dozens of specific product suites - but not toppling any outright.
3) The nice-to-haves. The Figmas, the HubSpots, the myriad of niche-function-high-cost SaaS companies out there making up the bulk of the market. Those whose products lack self-hosted alternatives risk having vibe-coded alternatives be "good enough" for an Enterprise looking to slash costs without regard to long-term support or quality; those who compete with self-hosted alternatives are almost certainly cooked, to varying degrees. If AI tooling can crank out content similar in quality to Figma and the company has tech talent to refine it for long-term use, why bother paying for Figma? If AI tooling can crank out a CRUD UI for users that just executes standard REST API calls behind the scenes, then why bother paying for fancy frontends? While it's technically interesting and novel at how these startups solved issues around scaling, or databases, or tenancy, the reality is that a lot of these niche products or services could be handled in-house with a container manager, a Postgres instance, and a mid-level IT person to poke it when things go pear-shaped. The higher per-seat prices of a lot of these services make them ripe for replacement in businesses comfortable with leveraging AI for building solutions, and I expect that number to grow as the tools become more widely available and IT-friendly in terms of security.
Ultimately, the core promise of SaaS to business customers was all the functionality with none of the costs of self-hosting support. Nowadays, many of them have evolved into solutions that are more expensive than self-hosted options, and businesses that have shifted IT into public clouds or container-based systems have realized they can do the same thing for less themselves, at the cost of some UI/UX niceties in the process. Now that we (IT) can crank out integrations with local LLMs with little to no cost, we're finally able to merge datasets into singular pools or services - and I'm not talking about Snowflake or its "big data" ilk so much as just finally getting everything into Salesforce or ServiceNow without having to bring in consultants.
The must-haves and many of the juggernauts will remain - for now. It's the niche players that need to watch their moats.
No it isn’t. Writing the code was never the issue with making software, it was designing it.
You can shit out an app with AI, just like you could with Indian workers. But that doesn’t mean it will work properly or that you’ll be able to maintain it.
And most importantly, it only works for code they could steal from GitHub. It has no idea how to replicate sensitive systems which aren’t publically documented, and those are some of the most valuable contracts.
I didn't realize B2B SaaS products were in freefall like his numbers suggest. I'm not convinced customers are leaving to vibe code their own products but I do believe we're seeing a major shift in the market, pushed by the sudden relative ease of coding. There are a lot of B2B SaaS products which are outdated and I wouldn't be surprised if they're supplanted by much faster competition
Yup it's definitely not because _customers_ are coding solutions, but the trend and motivation seems to come from the fact that customers are realizing there's something else possible except being tied into expensive recurring yearly subscriptions.
I was surprised when I saw the numbers from Bloomberg myself as well!
Having worked in enterprise B2B SaaS for a long time, almost every feature I built could have been a simple spreadsheet or some emails. So I'm highly skeptical AI is going to change anything.
Enterprise sales basically works like this: A non-technical sales team aggressively promises everything to win a deal to a non-technical procurement or exec team. When the deal is won, the SaaS sales team tells engineers "go build this" regardless of how stupid it is. And the customer tells their employees "you now have to use this SaaS" regardless of whether it makes sense.
A given company or enterprise does not have to vibe code all this, they just need to make the 10 features with the SLA they actually care about, directly driven off the systems they care about integrating with. And that new, tight, piece of software ends up being much more fit for purpose with full control of new features given to company deploying it. While this was always the case (buy vs build), AI changes the CapEx/OpEX for the build case.
I'm pretty sure every developer who has dealt with janky workflows in products like Jira has planned out their own version that fits like a glove, "if only I had more time".
A lot of companies have been too smart for that, and a lot of SaaS offerings are too small to be truly entrenched. Arguably the investment horizon is too short (IBM took decades getting to that point).
The only real vendors who managed to become the next IBM are the cloud providers.
Wrong take. You don't need to build something better, you only need something good enough that matches what you actually need. Whether you build it or not and ditch the SaaS is more of an economic calculus.
Also, this isn't much about ditching the likes of Jira not even mentioning open source jira clones exists from decades.
This is more of ditching the kind of extremely-expensive-license that traps your own company and raises the price 5/10% every year. Like industrial ERP or CRM products that also require dedicated developers anyway and you spend hundreds of thousands if not millions for them. Very common, e.g. for inventory or warehouse management.
For this kind of software, and more, it makes sense to consider in-housing, especially when building prototypes with a handful of capable developers with AI can let you experiment.
I think that in the next decade the SaaS that will survive will be the evergreen office suite/teams, because you just won't get people out of powerpoint/excel/outlook, and it's cheap enough and products for which the moat is mostly tied to bureaucratic/legal issues (e.g. payrolls) and you just can't keep up with it.
The sheer volume of data, the need for real time consistency in store locations, yada yada means that bad early decisions bite hard down the road.
Lots of drudge work can be assisted by AI, especially if you need to do things like in ingest excel sheets or spit out reports, but I would run far away from anything vibe coded as hard as possible.
One of my clients spends 500k+ on XXX licensing per year (for a 200M revenue company that's not peanuts), and on top of that has to employ 12 full time XXX developers (that command high figures just for their expertise on that software while providing very little productivity) and every single feature takes months to develop anyway. Talking about stuff like adding few fields to a csv output.
So the total cost of XXX is in the 2M/year range, and it keeps ballooning.
My (4 men) team already takes care of the entire warehouse management process except inventory, the only thing that XXX provides, we literally handle everything: picking, manufacturing, packaging, shipping phase and many others.
In any case, nobody has mentioned vibe coding.
I stated that a handful of good engineers with the aid of AI in a couple of months can provide a working prototype to evaluate. In our case it's about extending our software that already does everything, except inventory management.
When you spend 2M/year on a software (1% of your revenue), growing every year by 100/150k it makes sense to experiment building a solution in house.
I rather use Excel. It's likely More robust and safer than the vibe coded app that could trigger data loss / incorrectness / issues any time.
If you are selling SaaS consider that a vibe-coding customer is validating your feature roadmap with their own time and sweat. It's actually a very positive signal because it demonstrates how badly that product is needed. If they could vibe code a "good enough" version of something to get themselves unstuck for a week, you should be able to iterate on those features and build something even better in short order, except deployed securely and professionally.
Everyone's going to talk about how cool their custom vibe-coded CRM is until they get stuck in a failed migration.
Charging hundreds of thousands if not millions per year for very basic functionality is what is "killing" b2b SaaS.
Let's put an example an exception-tracking SaaS (Sentry, Rollbar). How do the economics of paying a few hundred bucks per month compare vs. allocating engineering resources to an in-house tracker? Think development time, infra investment, tokens, iteration, uptime, etc. And the opportunity cost of focusing on your original business instead.
One would quickly find out that the domain being replaced is far more complex and data-intensive than estimated.
But, not sure which successful SaaS companies just stopped shipping any updates to the product, never talked to their customers and never added any new features to win over major new accounts - and still managed to survive and thrive?
And the author actually confirms this:
> AI isn’t killing B2B SaaS. It’s killing B2B SaaS that refuses to evolve.
Can you though? With major bugs? We've been getting more and more crashes, downtime, issues etc lately and a lot of it has had to do with vibe coding.
The whole point of these B2B SaaS is meant to be quality.
i.e. it's set users' expectations but in the wrong way.
I've never seen a SaaS product that fits this description. There are always things to do. Libraries to upgrade, performance bottlenecks to diddle around with, an endless stream of nonsense feature requests from people at the customer who never actually use the product, fun experiments your developers want to try out, and so on.
The hard part in SaaS is to delete code, and that's what you should do, at least some of the time. Either through simplifications, or just outright erasing functionality that very few if any of your customers rely on.
What you should not do is let your customers grow the liability that is code in your production environment, unless your entire product set is designed to handle things like this, e.g. the business models of Salesforce and SAP.
but they don't want to. and they will be replaced, as it's good and well.
I feel like there's an interesting story in there.
1) The must-haves. These are your email and communication systems, the things you absolutely have to have up and available at all times to do business. While previously self-hosted (Exchange/Sendmail, IRC/Skype/Jabber, CallManager/UCS), the immense costs and complexities of managing systems ultimately built on archaic, monolithic, and otherwise difficult-to-scale technologies meant that SaaS made sense from a cost and a technical perspective. Let's face it, the fact nobody really hosts their own e-mail anymore in favor of Proton/Microsoft/Google/et al shows that self-hosting is the exception here, not the norm - and they're not going anywhere regardless of how bad the economy gets. These are the "housing stock" of business, and there's plenty of cheap stock always available to setup shop in without the need for technical talent.
2) The juggernauts. The, "we can do this ourselves, but the pain will be so immense that we really don't want to". This is the area where early SaaS solutions cornered and exploded in growth (O365, ServiceNow, Google Workspaces), because managing these things yourself - while feasible, even preferable - was just too cheap to pass up having someone else wrangle on your behalf with a reasonable SLA, freeing up your tech talent for all the other stuff. The problem is that once-focused products have become huge behemoths of complex features that most customers neither need nor use on a regular basis, at least after the initial pricey integration. Add in the ease of maintainability and scalability brought by containers or microservices, along with the availability and reliability of public cloud infrastructure, and suddenly there's more businesses re-evaluating their relationships with these products in the face of ever-rising prices. With AI tooling making data exfiltration and integration easier than ever from these sorts of products, I expect businesses to start consolidating into a single source of truth instead of using dozens of specific product suites - but not toppling any outright.
3) The nice-to-haves. The Figmas, the HubSpots, the myriad of niche-function-high-cost SaaS companies out there making up the bulk of the market. Those whose products lack self-hosted alternatives risk having vibe-coded alternatives be "good enough" for an Enterprise looking to slash costs without regard to long-term support or quality; those who compete with self-hosted alternatives are almost certainly cooked, to varying degrees. If AI tooling can crank out content similar in quality to Figma and the company has tech talent to refine it for long-term use, why bother paying for Figma? If AI tooling can crank out a CRUD UI for users that just executes standard REST API calls behind the scenes, then why bother paying for fancy frontends? While it's technically interesting and novel at how these startups solved issues around scaling, or databases, or tenancy, the reality is that a lot of these niche products or services could be handled in-house with a container manager, a Postgres instance, and a mid-level IT person to poke it when things go pear-shaped. The higher per-seat prices of a lot of these services make them ripe for replacement in businesses comfortable with leveraging AI for building solutions, and I expect that number to grow as the tools become more widely available and IT-friendly in terms of security.
Ultimately, the core promise of SaaS to business customers was all the functionality with none of the costs of self-hosting support. Nowadays, many of them have evolved into solutions that are more expensive than self-hosted options, and businesses that have shifted IT into public clouds or container-based systems have realized they can do the same thing for less themselves, at the cost of some UI/UX niceties in the process. Now that we (IT) can crank out integrations with local LLMs with little to no cost, we're finally able to merge datasets into singular pools or services - and I'm not talking about Snowflake or its "big data" ilk so much as just finally getting everything into Salesforce or ServiceNow without having to bring in consultants.
The must-haves and many of the juggernauts will remain - for now. It's the niche players that need to watch their moats.
You can shit out an app with AI, just like you could with Indian workers. But that doesn’t mean it will work properly or that you’ll be able to maintain it.
And most importantly, it only works for code they could steal from GitHub. It has no idea how to replicate sensitive systems which aren’t publically documented, and those are some of the most valuable contracts.
Since when does stock price / valuation have to match actual business realities?
I was surprised when I saw the numbers from Bloomberg myself as well!
Enterprise sales basically works like this: A non-technical sales team aggressively promises everything to win a deal to a non-technical procurement or exec team. When the deal is won, the SaaS sales team tells engineers "go build this" regardless of how stupid it is. And the customer tells their employees "you now have to use this SaaS" regardless of whether it makes sense.
> And vibe coding is fun. Even Bret Taylor, OpenAI’s chair, acknowledges it’s become a legitimate development approach.
Color me shocked! Bret, who directly profits by how his product is perceived, thinks it's legitimate???? /s
??? Do you mean biased or do you mean impartial?