When the Agents Pick: Why Marketo, Outreach, Salesloft and Atlassian Aren’t Built for AI Agents
The other day we built an Agentic API grader. The setup was simple. I had Claude, OpenAI, and Gemini take the top 120 APIs together and grade which ones they thought were the best for agents to use. Stripe came out on top with the only A+ grade. That was the predictable part.
Here’s what surprised me.
When I asked the agents what they’d use for marketing automation, sales engagement, and conversation intelligence, all three models responded the same way. They mocked Marketo. They mocked Outreach. They mocked Salesloft. They said these products have no place in an agent-driven workflow. The exact framing was that an agent will never send an email through Outreach, Salesloft, or Marketo because it will just craft and send a better email itself.
No use to agents. That was the consensus from three different foundation models.
These tools still have plenty of value to humans today. The question is what happens as the ratio of agent-driven work to human-driven work shifts, which it’s doing fast. Even if you assume agents only handle 30% of these workflows by end of 2026 (and the real number is going to be higher), that’s 30% of the customer base that has no native need for the product category.
This is the part of the agent transition that very few people have actually internalized. It’s not that agents will use the same tools we use, just faster. It’s that agents have no need for entire categories of software that humans built to compensate for the fact that we couldn’t write a thousand personalized emails ourselves.
The Categories Agents Will Bypass
Marketing automation. Marketo, HubSpot’s enterprise tier, Eloqua. The whole point of these products is to let a marketer schedule, segment, and template communications at scale because a human can’t manually send 50,000 emails. An agent can. An agent can also generate each email fresh, personalized to the recipient, with context pulled from your CRM, the prospect’s recent activity, and the current state of your product. The templating layer that defined this category is exactly what the agent replaces.
Sales engagement. Outreach, Salesloft. These tools exist so SDRs can run sequences. The sequence is the workaround for the fact that a human SDR can’t remember to follow up with 400 leads in the right cadence. Agents have no such limitation. They don’t need a sequence builder. They run the cadence natively and they generate each touch in real time based on what the prospect actually did.
Conversation intelligence. Gong, Chorus. These tools record sales calls, transcribe them, and surface insights for humans to act on. An agent doesn’t need a dashboard of insights extracted from a transcript. The agent listens to the call, ingests the transcript directly, and acts on it. The insight layer was a translation tool for humans. Agents don’t need translation.
Project management. Atlassian (Jira, Confluence), Monday, Asana. These exist so humans can coordinate work across teams. Agents don’t need a Kanban board to track what they’re doing. They don’t need a wiki to share knowledge with each other. They have memory and context windows. The whole UX of project management was a coordination layer for biological brains.
Design tools beyond agent reach. Even Canva, which just shipped a great agentic suite, isn’t going to be picked by an agent. Why would an agent open a UI to drag assets around when it can just generate the assets directly? The agent doesn’t want a canvas. It wants an output.
What This Looks Like in Practice at SaaStr
Our Salesforce bill went from $12,000 to $22,000 a year. Our seats went from 10 to 2 plus one. The token consumption is dramatically up because the agents run constantly. Salesforce captures more value per agent than per human seat. That’s bucket 3, agent-leveraged software.
Now look at our Marketo equivalent. We don’t have one. Our AI VP of Marketing, 10K, doesn’t need it. It writes the campaigns, picks the segments, sends the emails, and measures the results. There’s no Marketo seat to upgrade. There’s no Outreach contract to renew. There’s no Salesloft expansion. The category just got cut.
This is happening across our stack. Vendor consolidation is taking 30-50% of new AI dollars by Gartner’s numbers. The first things to get cut are the tools that exist purely as a productivity layer for humans. Marketo, Outreach, Salesloft are the cleanest examples because their entire value proposition is “let a small team do what a larger team would have to do.” Agents are an even smaller team that does even more.
The Shifting Ratio
Here’s the framing that matters. Every pre-AI productivity tool was built around a constraint: humans can’t do this work at scale, so the tool exists to make humans more productive. Marketo exists because a marketer can’t write 50,000 personalized emails by hand. Outreach exists because an SDR can’t track 400 cadences in their head. Jira exists because a 50-person product org can’t coordinate work in chat.
Those constraints are real. They still apply to humans. The tools still help humans do more.
But the ratio is what’s changing. As agents take over more of these workflows, the share of customers (or customer workflows) that natively need the productivity layer starts to compress. The marketer doesn’t need Marketo because the agent writes 50,000 emails fresh. The SDR team doesn’t need Outreach because there is no SDR team. Some of the work coordination is now happening through different primitives that don’t need a Kanban board.
The customer keeps paying for the human seats they still have. The question is what the curve looks like as those human seats compress and the agent workflows expand. The bear case isn’t “Marketo dies tomorrow.” The bear case is “Marketo is selling into a category where the addressable workflow shrinks 20% per year while the products growing fastest are agent-native.”
That’s enough to break a growth thesis. It’s enough to make debt service hard. It’s enough to compress multiples. It doesn’t require the product to disappear.
What Doesn’t Get Bypassed
Not everything goes. Some categories survive because agents need them:
Systems of record that hold customer data, financial data, or operational data the agent has to read or write. Salesforce, Workday, NetSuite, Stripe. These are infrastructure for the agent. Agents don’t replace them, they call them.
Communication infrastructure. Twilio, SendGrid, the actual delivery layer. Agents need to send the email through something. The thing they cut is the templating, scheduling, and segmentation layer above it.
Compute and AI infrastructure. Cloudflare, the hyperscalers, the foundation models themselves.
Code repositories and CI infrastructure. Agents write code. They check it in somewhere. They run tests through something.
Vertical software with deep domain integration. The more your product is the connective tissue to a specific industry’s workflows, regulatory requirements, or physical infrastructure, the harder it is for an agent to bypass.
What gets cut is the human productivity layer. What survives is infrastructure.
What This Means If You’re Running One of These Companies
If you’re an exec at Marketo, Outreach, Salesloft, Gong, Atlassian, or Monday, the strategic question is no longer “how do we add AI features.” Adding AI features to a product agents have no native need for is rearranging deck chairs. The strategic question is “what do we become as the share of work being done by agents instead of humans grows every quarter, and what’s our product when an agent is the buyer?”
The honest answer for some of these companies is “infrastructure for agents instead of tools for humans.” That’s a full rewrite of the product, the pricing model, and the go-to-market. It’s not a feature release.
For others, the honest answer is more uncomfortable: the human user base is real and durable but it’s a flat or declining curve, and the company’s job is to compound cash from the existing base while figuring out a second act. That’s not a fun answer but it’s the rational one for some categories.
What This Means If You’re Investing
The bare case for the entire pre-AI B2B stack is what the public markets are now starting to price in. Atlassian and Monday have been hammered specifically because the market figured out that project management has no native agent demand. Twilio and Cloudflare have outperformed because they do.
The longer-tail names like Marketo, Outreach, Salesloft don’t have public stock prices to watch but their PE owners do. And several of those positions are sitting on debt that’s getting harder to service against revenue that’s at risk of decline, not growth.
It’s worth taking the framework seriously. Sit with each name in your portfolio or each tool in your stack and ask: would an agent pick this? If the answer is no, what’s the thesis for why the company keeps growing?
In a lot of cases, there isn’t one.
