There's been a lot of SaaS-bashing lately. Klarna cancelling its Salesforce and Workday contracts. Gavin Baker on All-In Podcast saying that enterprise apps will be “worst-performing asset of 2025.“ Chamath railing against “the software industrial complex... enterprise software companies that have convinced organizations to spend a tremendous amount of money wrapping business rules around a CRUD database.“ Satya Nadella predicts “business apps will all collapse in the agent era [since SaaS apps] are just CRUD databases with business logic, logic is going to agents, and agents will be multi-repo CRUD.“ Ben Thompson of Stratechery likening SaaS companies to pre-Internet ad agencies and predicting their slow demise.
But is SaaS really dead? Despite long LLM context windows, structured data still matters for accuracy, access control, and latency. Deterministic workflows are needed for repeatability and control. I actually think SaaS systems of record and workflows may be more important than ever, needed for agent RAG and tools. Also, many people these days are pooh-poohing seat licensing, but seats are working quite well for OpenAI– 85% of its estimated $4B ARR (250% y/y) is from paid ChatGPT seats.
Of course, not all SaaS companies are created equal. Those with source-of-truth data and high product utilization and renewal rates will fare best. Overall, AI will exert both upward and downward pressure on software business models:
‣ As Dave Kellogg has articulated, AI may deliver a lot of value, but it may not be possible to monetize because value is the upper bound on price whereas the cost of alternatives is the lower bound.
‣ Price per seat for software could increase with addition of AI features, unless everyone else also offers those same features (eg, draft a prospecting email) in which case it becomes table stakes.
‣ Seat price on previously premium features could be pressured if those features are easily replicated with an AI prompt (eg, summarize call).
‣ # seats sold depend on factors like current utilization and TAM penetration. If utilization is low, renewals may be flat or down even after swapping in licenses for new AI features. In the current environment, CIOs are less willing to buy or renew unused seats or functionality compared to previous cycles.
‣ Margins may be pressured due to model inference costs.
‣ Replacing your CRUD database and workflow platform has high switching costs; LLMs could bring down the switching cost.
Will the Rule of 40 and high-teens or greater double-digit revenue growth that has driven SaaS valuations over the last decade be sustainable in the agent era? Or will some need to face the grim reality described by Gokul Rajaram as “the no man's land of low growth coupled with low FCF margins“, where it will be a challenge to sustain investor interest? Personally, I'm rooting for SaaS and either way, 2025 is shaping up to be a watershed moment for the industry.
But is SaaS really dead? Despite long LLM context windows, structured data still matters for accuracy, access control, and latency. Deterministic workflows are needed for repeatability and control. I actually think SaaS systems of record and workflows may be more important than ever, needed for agent RAG and tools. Also, many people these days are pooh-poohing seat licensing, but seats are working quite well for OpenAI– 85% of its estimated $4B ARR (250% y/y) is from paid ChatGPT seats.
Of course, not all SaaS companies are created equal. Those with source-of-truth data and high product utilization and renewal rates will fare best. Overall, AI will exert both upward and downward pressure on software business models:
‣ As Dave Kellogg has articulated, AI may deliver a lot of value, but it may not be possible to monetize because value is the upper bound on price whereas the cost of alternatives is the lower bound.
‣ Price per seat for software could increase with addition of AI features, unless everyone else also offers those same features (eg, draft a prospecting email) in which case it becomes table stakes.
‣ Seat price on previously premium features could be pressured if those features are easily replicated with an AI prompt (eg, summarize call).
‣ # seats sold depend on factors like current utilization and TAM penetration. If utilization is low, renewals may be flat or down even after swapping in licenses for new AI features. In the current environment, CIOs are less willing to buy or renew unused seats or functionality compared to previous cycles.
‣ Margins may be pressured due to model inference costs.
‣ Replacing your CRUD database and workflow platform has high switching costs; LLMs could bring down the switching cost.
Will the Rule of 40 and high-teens or greater double-digit revenue growth that has driven SaaS valuations over the last decade be sustainable in the agent era? Or will some need to face the grim reality described by Gokul Rajaram as “the no man's land of low growth coupled with low FCF margins“, where it will be a challenge to sustain investor interest? Personally, I'm rooting for SaaS and either way, 2025 is shaping up to be a watershed moment for the industry.
