Is the business model for traditional B2B software in permanent decline because AI agents don’t need seats?
The answer right now, at least from our own data running SaaStr with 3 humans and 20+ AI agents, is more complicated than the simple “seats are dead” narrative making the rounds.
It cuts both ways. And for us … hard.
Let me give you two examples from our own stack. Same vintage, both companies that have shipped real AI features, both companies we genuinely liked. One we now pay 83% more to. The other we still pay, but stopped using months ago.
Exhibit A: Salesforce. Where We Now Pay 83% More, With 80% Fewer Humans.
A year ago we had 10+ human seats on Salesforce.
Today we have 2 human seats and 1 API seat.
You’d might think our Salesforce bill would have collapsed. It didn’t. It went way up. And it’s OK.
We now pay Salesforce ~$22,000 a year, up from $12,000. That’s an 83% increase year-over-year, with 80% fewer human users.
Why? It’s simple. It’s because our 20+ AI agents use Salesforce roughly 100x more than our humans ever did. We have an AI VP of Marketing. An AI VP of Customer Success. Four third-party AI GTM agents running on top of Salesforce, including Agentforce running win-back campaigns at 72% open rates.
They never sleep. They never stop querying. They never stop writing. They use Salesforce as the central nervous system of our entire go-to-market motion.
Human seats: down 80%. Usage: up 100x. And Salesforce captured a real share of that value through API calls, Agentforce, Data Cloud, and consumption-based pricing on AI features.
That’s the bull case for traditional B2B software when agents take over the work. Pricing models flex from seats to consumption, and the apps that are actually doing the work, the systems of record, get paid more, not less.
Exhibit B: Notion. Where We Still Pay For Now, But … We Stopped Using It Months Ago. Even Though It’s Great.
Now flip it.
We loved Notion. We were heavy users for years. And they’ve shipped a serious AI product which has re-accelerated growth. Notion is back and Notion AI is good. AI usage inside Notion has accelerated.
It’s just our AI agents don’t care. They have no use for Notion.
Our AI VP of Marketing has no reason to log into Notion. Our AI VP of Customer Success doesn’t either. The real-time dashboards our AI VP of Marketing builds (she leads our weekly standup now, which we used to run out of Notion) are better, customized to us, and integrate natively with every other agent in our stack.
Last week I realized we hadn’t actually opened Notion in months.
Not because Notion got worse. It got better. Notion AI is impressive. But the workflows that lived in Notion (meeting notes, internal wikis, project trackers, dashboards) got absorbed by agents that build their own real-time interfaces on top of Salesforce, Slack, and our own custom dashboards.
We didn’t decide to leave Notion. We just stopped needing it. The agents routed around it.
We’re still paying for the moment. Notion isn’t expensive. But when we review everything in a few months? I bet we quickly say thank you, good times, but cancel.
And many will get there over the coming 24 months,. They haven’t even made many of these churn and rationalization decisions yet. Most B2B software companies haven’t seen the revenue impact of agent disintermediation yet, because the renewals haven’t hit. The seats are still on the invoice. The usage already left. Stealth Churn. The financial impact is locked in but lagging.
The Rule Is Simple. The Application Is Where It Gets Interesting.
Why did one bill go up 83% and the other become a renewal we won’t sign?
It’s not complicated. AI agents use software they need to be successful at their jobs. They ignore software they don’t.
Our AI VP of Marketing, AI VP of Customer Success, and our four AI GTM agents need Salesforce to do their work. Salesforce is the data, the contacts, the pipeline, the activity history, the system everything else routes through. Without Salesforce, our agents don’t function.
Notion isn’t critical to any of our agents being successful. It’s a beautiful app for humans. The agents simply have no reason to open it. Notion AI being good doesn’t change that. The AI inside Notion is for humans, and we have fewer humans doing that work every quarter.
That’s the whole rule. It’s the cleanest test I’ve found:
Is this software critical to AI agents being successful at their jobs? If yes, usage and spend go up. If no, usage and spend go to zero. Eventually.
So How Does This Cut for the Rest of B2B Software?
This is the actual question, and I don’t think anyone has good answers yet. But it’s the question every public software CEO and every B2B founder should be working on right now.
Starting with the clearest beneficiaries and working down to the clearest threats:
- AI infrastructure APIs. ElevenLabs, Fal.ai, the model providers, the voice and image and video API layer broadly. Already benefiting and will keep benefiting. Agents are the customers. Every agent that talks, sees, generates images, or reasons drives spend here. This is probably the cleanest bull case in all of B2B software. Even Twilio is seeing a bit of a renaissance.
- Agentic-friendly data and infrastructure. Snowflake, Databricks, MongoDB. The Salesforce bull case on steroids. Agents need data. They need warehouses. They need vector stores. They query orders of magnitude more than humans do, and most of these platforms already price on consumption. If our Salesforce bill went up 83% in a year, the Snowflake and Databricks bills of agentic-first enterprises should go up vastly more. Hard to see how this category isn’t a massive beneficiary, especially in enterprise.
- CRM, Salesforce. Agents clearly need CRMs. Every GTM agent reads from and writes to the CRM constantly. Salesforce is set up well: enterprise customers, agentic-first buyers, and pricing power that flexes to consumption. Bull case is strong, and our own bill is the proof point.
- Code and engineering systems of record. GitHub, Linear, Jira. Coding agents already use GitHub more than humans do on some teams. Strong bull case.
- Communication systems of record. Slack. Agents need these to know what happened, what was promised, what to follow up on. Bull case if they’re addressable.
Now the middle bucket. Categories that genuinely could go either way:
- CRM, HubSpot. Architecturally, HubSpot should benefit the same way Salesforce does. Agents should drive consumption up dramatically. But HubSpot’s SMB focus is a real wildcard. Will SMBs tolerate the kind of pricing increases enterprise customers will swallow? Will SMBs even deploy 20+ AI agents in the first place, or stay on 2-3 humans for years? SMBs may be a tougher sell here than enterprise and agentic-first leaders. TBD.
- Design, Figma and Adobe. Genuinely unclear. Agents are starting to design. But Figma is also the system of record for design specs, components, and dev handoff. Could go either way.
- HR and payroll. Workday, Rippling, Gusto, BambooHR. Genuinely a push. AI agents don’t need payroll. Agents don’t get paid. But if human hiring keeps growing, or even stays flat, payroll providers don’t really care whether agents use the software. The customer base is still humans, and humans still need to be paid, onboarded, and managed. The category is only at risk if total human headcount actually drops in the economy. A much longer-dated and less certain outcome.
And the categories where I’d be most concerned as a founder / exec:
- Support tools. Zendesk, Intercom. Agents are taking over Tier 1. The question is whether the agents run on top of these platforms or replace them entirely. The platforms have a window to become the substrate. If they don’t, they get bypassed. They know this. All of them are racing to be that substrate. Fin is nothing like the Intercom of old, for example.
- Project management and workflow tools. Asana, Monday, ClickUp, Notion, Confluence. The category I’d be most worried about. Agents don’t need a human-readable project tracker. They can build their own dashboards from the underlying data. The “AI inside the app for humans” pitch doesn’t help if the humans aren’t doing the work anymore.
- Design, Canva and lighter creative tools. Hard to see the bull case. Agents don’t need Canva. They autogenerate images and videos inside Replit or Lovable, or call Fal.ai and similar APIs directly. The whole point of Canva was making design accessible to non-designers. When the non-designer is an AI agent, the agent goes straight to the image generation model. Canva gets routed around.
- Sales engagement and marketing automation. Outreach, Marketo, Mailchimp. Their current form is almost useless for AI Agents. The clearest replacement category. Worth asking: do agents need these, or do agents become these? An AI SDR doesn’t need Outreach. It is Outreach. The platforms aren’t being used by agents. They’re being replaced by agents. Marketo and all non-agentic first marketing automation solutions are worthless to agents if the agents themselves are generating and managing the campaigns, etc.
I don’t have 10)% confident answers on any of these. What I have is an 83% increase on one line item and a renewal we won’t sign on another, both within the same year, both at the same company, both running the same AI transformation.
And is interesting, that roughly, the public markets track these concerns: Not entirely, but overall:
This Isn’t About AI Changing Software. It’s About AI Agents Choosing Software.
This is less about if AI will disrupt SaaS and software. That’s the easy framing, and at this point it’s everywhere.
It’s about something deeper. Do AI agents, as opposed to humans, even really need your software at all?
If yes, you’re set up to capture more revenue per customer than you ever did, even as human seat counts collapse. Salesforce, Snowflake, GitHub, ElevenLabs.
If no, “AI features inside the app for humans” doesn’t save you. The humans are doing less of the work every quarter. The agents aren’t going to replace them as users of your software.
That’s the question every B2B founder and every public software CEO should be answering directly right now. Not “are we doing AI?” Almost everyone is. The real question is much harder:
Do AI agents actually need our software?
Most CEOs aren’t asking it directly enough yet. They will be soon.
How to Research Which B2B Products Are Actually Built for AI Agents
We built a tool for exactly this question. The SaaStr AI Agent Grader scores how API-ready and agent-addressable any B2B product is. Useful for evaluating the products already in your stack, and useful for benchmarking your own software against what AI agents actually need to do their jobs.
Try it here: https://www.saastr.ai/apireport:
