Every SaaS playbook you know assumes a human is on the other side of the screen.
Landing pages built to persuade a person. Demos, testimonials, sales calls. All of it aimed at a human brain.
Greg Isenberg thinks that assumption is about to break. In a recent episode of his Startup Ideas podcast, he laid out a prediction: over the next 10 years, AI agents will outnumber humans on the internet. Billions of them. With wallets.
And almost nobody is building for them yet.
I watched the full episode and pulled out the parts that matter if you're a bootstrapped founder trying to figure out where the next wave of customers comes from. Here's the breakdown.
The customer is no longer (always) human
Greg's core claim is simple.
For 30 years, the user of the internet was a person. You built websites and apps, and a human clicked, compared, and bought.
That's changing. Agents are starting to discover tools, evaluate them, pay for them, and renew them. Machine to machine. No human in the loop.
He tweeted the thesis a few months back: "Build startups for agents. Over the next 10 years, you're going to have a market of billions of customers, aka agents, with millions of wallets that want to use your services."
Then he asked the question that stuck with me:
→ Take every SaaS tool you use today (Notion, Slack, Stripe) and ask: what's the version of this built purely for agents?
Agent-native payments. Agent-native communication. Agent-native memory. Every category gets rebuilt.
I run a stack of AI agents in my own business every day (they book podcast guests, publish content, chase backlinks). So when Greg says agents will be buying software, I don't think it's sci-fi. I think it's a Tuesday in 2027.
The agent buying journey (yes, agents have one)
Weird sentence to write. But Greg maps it out, and it's worth internalizing.
A human customer wants persuasion. An agent customer wants structured capability, permission, and trust.
Here's what the journey looks like when the buyer is an agent:
- Finding. "Find a payroll tool for 40 contractors." The agent runs the search, not the human.
- Evaluating. It reads your docs, your pricing, your API reference, your reviews. Not your hero section.
- Trust-checking. It looks at your policies, limits, and identity requirements.
- Transacting. It pays, books, signs, subscribes.
- Using. It files tickets, changes settings, pulls reports.
- Recommending. It tells other agents what worked.
That last one is the strangest part. Agents recommending tools to other agents. Greg points to Multbook (the social network for agents that Facebook acquired) as an early glimpse.
Every step of that journey has missing infrastructure, because the old internet assumed a person was doing the work.
What agents need that humans don't
Greg's list. Six things:
- Identity. Who is this agent acting for?
- Tools. What actions can it safely invoke?
- An inbox. Where do replies, OTPs, and confirmations land?
- Memory. What does it know about its owner's preferences and rules?
- A wallet. What can it spend, and who approves it?
- Receipts. What did it see, decide, change, and buy?
His analogy: it's like onboarding an employee. You don't hand a new hire the company card on day one. Trust builds, then limits go up.
Each of those six is a product category that barely exists.
Who's already making money on this
This isn't theoretical. Greg names names.
AgentMail is building email inboxes for AI agents. Their pitch: "the email inbox API for AI agents." Gmail, but the account holder is software. YC-backed, and Greg says they're doing really well.
Stripe launched agent wallets. Imagine a purchasing agent with spend caps, approval rules, shared payment tokens, and a full audit trail.
Then the use cases stack up fast:
- A support agent that files the ticket, attaches the logs, asks for the refund, follows up, and escalates when ignored.
- A CFO agent that compares 12 vendors, reads the SOC 2 docs, negotiates terms, and recommends the one that fits policy.
- A SaaS that exposes MCP tools (search customers, create invoices, refund orders) so the agent never scrapes the UI.
- A travel agent that books dinner, calls the hotel, pays the deposit, and updates the calendar.
If agent traffic passes human traffic, all of this becomes table stakes. Companies like TwiLead are already selling done-for-you business automations to agencies and SMBs, and that market only gets bigger as the buyers themselves become agents.
What this changes for your SaaS
This is the section to reread if you're at $5K to $50K MRR and wondering what to do about any of this.
Greg's translation table, founder edition:
→ SEO becomes AEO. You're no longer optimizing for a human scrolling Google. You're optimizing to be the answer agents cite, trust, and recommend. (I wrote a whole playbook on this: how to get your SaaS recommended by ChatGPT.)
→ Forms become tool calls. The CTA on your site becomes an action endpoint an agent can invoke.
→ Support docs become executable support. The agent doesn't read your help center. It does the refund, the reschedule, the troubleshooting.
→ Landing pages become capability manifests. Agents don't care about your slogan. They care about what they can actually do with your product. Structured docs, schemas, policies, endpoints.
→ Sales calls become agent procurement. Buyers send their agents to build the shortlist before a human ever shows up on Zoom.
→ Analytics become agent analytics. Which agents visited? What did they ask? Where did they fail?
That last one hit home for Greg. Earlier in his career he obsessed over voice-of-customer data to push conversion from 1% to 4%, which was worth millions. Now the question is: what does conversion rate optimization look like when the visitor is an agent?
Nobody has a real answer yet, and that gap is the opportunity.
His most tactical tip: create a /agents page on your website. A dedicated entry point where an agent can find your schemas, policies, endpoints, and sandbox. If an agent can't understand what you do and safely act on it, you're invisible to it.
9 startup ideas hiding in this shift
Greg rapid-fires these at the end. If you're idea-hunting, steal one:
- Agent SEO (AEO) agency
- Agent identity and permissions infrastructure
- Agent receipts and audit trails
- Agent-ready docs generators
- Agent inbox security
- Agent-readable pricing pages as a service
- MCP servers for franchises
- Agent support desk
- Sandboxes for agents to test SaaS safely
The pattern behind all nine: take a boring, proven category and ask "what's the agent version?" Same move that worked when everything went mobile.
This thesis pairs well with his earlier episode arguing that AI agents are the new SaaS. First agents replaced the software. Now they're becoming the customer.
The 10-year window
Greg's big prediction: the internet bifurcates into two. A human internet and an agent internet.
The founders who understood mobile-first in 2010 got a decade of tailwind. Agent-first feels like the same setup, except the window is opening right now and the incumbents haven't moved.
You don't need to build agent infrastructure to benefit. You just need to make your existing SaaS legible to agents before your competitors do.
But if you're still hunting for your next product, billions of customers with millions of wallets is not a bad place to start looking.
FAQ
What does "selling to AI agents" actually mean?
It means building products where the buyer is an AI agent acting on behalf of a human or business. The agent discovers your tool, evaluates your docs and pricing, pays with its own wallet (with spend caps set by its owner), uses your product through APIs or MCP tools, and recommends it to other agents. The human sets the goal; the agent runs the whole purchase.
Is this real today or still hype?
Early but real. AgentMail (YC-backed) already sells email inboxes for agents. Stripe ships agent wallets with spend controls. MCP servers are becoming a standard way for SaaS products to expose tools to agents. The infrastructure is maybe 5% built, which is exactly why Greg calls it an opportunity rather than a trend to watch.
How do I make my existing SaaS agent-ready?
Start small: publish structured docs and schemas, expose your core actions as an API or MCP server, add a /agents page with policies and endpoints, and make pricing machine-readable. Then work on AEO so agents actually find and cite you when their owners ask for recommendations.
Do bootstrapped founders have a shot here, or is this a VC game?
Bootstrappers might have the edge. The categories are so new that distribution and speed matter more than capital. A solo founder can ship an agent-ready docs generator or a niche MCP server in weeks. The big players are busy protecting their human-facing products.
Steal playbooks like this every week
I break down founder stories and shifts like this one every week on the Profitable Founder Podcast, with bootstrapped SaaS founders doing $100K to $10M a year.
