11:12Greg Isenberg · Essay
SaaS Is Dead, Long Live SaaS
An 11-minute whiteboard walk-through of why per-seat software is ending, and the free-tool-plus-paid-intelligence model replacing it.
October 29thThe argument in one line.
When AI agents become the real users of software, products, pricing, team size and funding all change: build headless tools for agents, price them flat and low, keep the team tiny, and bootstrap.
Read if. Skip if.
READ IF YOU ARE…
- A SaaS founder at 1 to 3 million ARR deciding whether to keep building dashboards or rebuild around agents and APIs.
- A solo builder pricing a new software product who is torn between a $497 plan and a $27 to $97 plan.
- A founder weighing a venture round against staying bootstrapped and profitable with a very small team.
- Someone selling data or outreach tooling who wants to see an agent-driven list-building run end to end.
SKIP IF…
- You want a step-by-step build tutorial for a SaaS product. This is a strategy talk with one short demo.
- You run a business that needs a physical or network-effect moat and has no software-as-dashboard exposure.
The full version, fast.
The speaker, who runs software companies doing over $30 million a year, argues that AI agents are rewriting the SaaS playbook. His five rules: build for agents instead of humans, expect headless software accessed through API and MCP, replace seat and credit pricing with flat predictable plans, aim for one or two people at 10 million ARR, and bootstrap instead of raising venture money. He demos an agent building a verified 450-person email list from one sentence inside Claude, explains why he cut his price from $497 to $97 to $27, and admits his bundled course-and-software funnel failed because the buyers were too early. He closes by naming who is safe: vertical software, agent tooling, network effects and data companies.
Free for members
Chat with this breakdown — free.
Sign in and you get 23 free chat messages on us — ask for the hook, quote a framework, find the exact transcript moment, generate a markdown action plan. Bring your own key when you want unlimited.
Create a free account →Where the time goes.
00:00 – 00:47
01 · Cold open: the price cuts
Price cut from $497 to $97 to $27, a YC founder's objection, and the promise of five new rules.
00:47 – 01:53
02 · Rule 1: Build for agents, not humans
The fastest growing startups will build for the agents that do the work, illustrated by MoltSets living inside Claude and ChatGPT.
01:53 – 02:51
03 · Rule 2: Headless software takes off
UI stops being a moat. Clay shipping a public API is the example.
02:51 – 03:20
04 · Rule 3: Seat and credit pricing collapses
Agents burn credits in one run, so they need an API key and humans need a predictable price.
03:20 – 05:14
05 · Demo: an agent builds a lead list
Claude with a ready-to-send-list skill pulls 1,000 contacts, grades them, keeps 450 of 997, and writes three short emails.
05:14 – 06:42
06 · Pricing for a bigger market
One API key for 75 clients, why $97 beats $497, and the $27 entry plan.
06:42 – 08:10
07 · Rule 4: One or two people to 10 million ARR
Simple products run by AI. RB2B at 10 million ARR with engineers idle for six months.
08:10 – 09:50
08 · Rule 5: Bootstrap for cash flow and freedom
VC concentrates in a few AI companies, so build a business that pays you from customers.
09:50 – 10:49
09 · The rule that didn't pay off
A $100,000 course, community and software funnel for RB2B failed because buyers were too early.
10:49 – 12:24
10 · Who should be scared, who wins
Dashboard SaaS is at risk. Vertical software, agent tooling, network effects and data companies win. Two options for early-stage founders.
12:24 – 12:35
11 · Closing question
If your next customer is an agent, what does your company need to look like?
Lines worth screenshotting.
- The next wave of startups will build for the agent that runs the workflow, not the human who logs in.
- In 2025 the pitch was building with AI. From 2026 it is building for AI.
- A public API used to be seen as a stickiness killer. Clay shipped one anyway because the market demanded it.
- Credit pricing breaks when an agent can burn a 4,000-credit month in a single run.
- An agency with 75 clients needs one API key, not 75, if the product is truly built for agents.
- Pricing at $97 instead of $497 can multiply the number of buyers by 100 or more.
- If you price below what it costs competitors to acquire a customer, their funnel cannot work.
- A $27 entry plan lets individual reps put the tool on a personal card.
- In the demo run, 450 of 997 pulled contacts survived grading, and throwing out the rest is the point.
- A dead simple product can be run almost entirely by AI, while a complex one needs humans to sell, onboard and support it.
- RB2B reached 10 million ARR without adding features, with engineers not touching the code for six months.
- One person spending under two hours a day handles all escalated customer interaction at RB2B.
- Venture money is concentrating in foundation-model and data-center companies, so the little guy in VC gets almost nothing.
- At 10 million ARR bootstrapped, two co-founders can each take over a million dollars a year in salary and dividends.
- Spending around $100,000 on four courses and weekly webinars failed because community members had no website traffic to use the software.
- The price cut came from customer discovery with agencies, not from cleverness: the market knows what it will pay.
- If your SaaS is mainly a dashboard that stores data and reports on it, a headless data asset with an AI on top replaces it.
- Safe categories are verticalized software with a real-world component, agent tooling, network-effect products and data companies.
Five shifts when agents are your customers
WHAT TO LEARN
When the user is an agent, the product, price, team size and funding model all shift toward headless, flat-priced, tiny and bootstrapped.
02Rule 1: Build for agents, not humans
- Agents are becoming the real users of data tools, so improving the workflows agents run and the accuracy of the data they get is where the value moves.
03Rule 2: Headless software takes off
- Headless access through API and MCP is replacing dashboards as the main interface, and people will ask an AI for reports in plain English.
04Rule 3: Seat and credit pricing collapses
- Seat and credit pricing break under agents that process thousands of records per run, so agents want an API key and humans want a fixed, predictable bill.
06Pricing for a bigger market
- A single API key should cover an agency's whole client list, because charging per key contradicts building for agents.
- Dropping from $497 to $97 to $27 widened the market, and a sub-$30 plan lets individual reps pay on a personal card.
- Pricing below competitors' customer acquisition cost makes their paid-ads and sales funnels hard to sustain.
07Rule 4: One or two people to 10 million ARR
- A deliberately simple product can reach 10 million ARR with one or two people because AI handles building, marketing, onboarding and support.
08Rule 5: Bootstrap for cash flow and freedom
- Venture capital is concentrating in a tiny set of AI companies, so bootstrapping for cash flow is the practical path for most software founders.
09The rule that didn't pay off
- A failed bundle of courses, community and software shows that you still have to sell to people who already have the problem and the traffic to use the tool.
10Who should be scared, who wins
- Dashboard-only and easily copied software is most at risk, while vertical software, agent tooling, network effects and data companies look safest.
- A useful test for any SaaS: if your next customer is an agent, what does your company need to look like?
Terms worth knowing.
- Headless software
- Software used mostly through an API or MCP connection by agents, with little or no dashboard. A person who wants a report asks an AI in plain English.
- MCP
- Model Context Protocol, a standard way for AI assistants like Claude to connect to outside tools and data so they can act on them directly.
- ARR
- Annual recurring revenue. The yearly value of active subscriptions, used as the main size measure for SaaS companies.
- Credit pricing
- A model where users buy a pool of credits and each lookup or action spends some. It works for humans pulling a few records but drains fast under agents.
- TAM
- Total addressable market. The full set of buyers a product could reasonably sell to at a given price.
- Verticalized software
- Software built for one specific industry, such as accounting tools for pool cleaners, often tied to real-world work the software helps coordinate.
Things they pointed at.
Lines you could clip.
02:40
“If you're still building UI for people, you're building for the past.”
05:33
“More than half of this list gets thrown out and that's the point. Sending to bad emails is how you burn your domains.”
10:30
“Even the price cut didn't come from me being clever. It came from customer discovery.”
Word for word.
Read-along
Don't just watch it. Burn it in.
See every word as it's spoken — crank it to 2× and still catch all of it. The same dual-channel trick behind Amazon's Kindle + Audible.
A few months ago, I cut the price of my newest SaaS company from $4 .97 a month to $97 a month. Then I cut it again to $27 a month. A Y Combinator founder told me I was nuts and that if I doubled the price from $4 .97, nobody in his cohort would even push back.
And you know what? He's probably right, but he's playing by the old rules of SaaS and I think those rules are about to break. You see, my software companies make over $30 million a year and the biggest shift that I'm seeing is that AI agents are changing how you build software, how you price software, how many people you hire and whether or not you should raise money at all.
In the next few minutes, I'll break down the five new rules of SaaS and tell you exactly what I would change if I wanted to scale a SaaS today to 10 million ARR and beyond. Rule one. Build for agents, not humans.
The old rule was simple. You build software for people. Somebody logs in, clicks around and gets value out of it.
The new rule is that the fastest growing startups for 2027 won't be building for humans anymore. They'll be building for agents. This means that making the workflows agents run.
Better, faster, and smarter in giving them more accurate data are the software companies that are going to win. Humans might still be in control, but the agent is the one actually doing the work. So the agent is who you're building for.
The way I put it is that in 2025, it was all about building with AI. In 2026 and beyond, it's about building. for ai and that's exactly why i built my newest sass mult sets the way i did every contact data tool out there was built for a human clicking through a dashboard but more and more the people pulling contact data aren't clicking anything they have claude doing it for them so we built the first sass that lives entirely inside of cloud and chat gpt there's no user interface at all it's a tool built for agents not humans and once your real user is an agent the product itself starts to look completely different rule number two is headless software takes off.
For the last 20 years, UI was one of the moats. The thinking was, if people's workflows live inside your product, they can't leave. And that made software sticky.
The new rule is that headless software is going to absolutely take off. There might still be some UI that humans can use, but a bigger and bigger share of your usage will come from agents connecting through API or MCP and acting on the software directly. And when a person does want to see something, they won't click through the dashboard.
They'll ask an AI. in plain English to pull a report from you. The best example of this I've seen is Clay.
Everybody smart I knew in the world told me that Clay would never release a public API because it would kill their stickiness. The workflows lived in the UI and that was the moat. But a few months ago, they released one anyway.
On paper, the move kills their stickiness. They did it anyway because that's what the market is demanding now. So my take's pretty simple.
If you're still building UI for people, you're building for the past. And if agents are the ones using your software, you charge for it has to change too rule number three is seat and credit pricing collapses the old way was to charge per seat and per credit and that made sense when a person was logging on and pulling 20 leads at a time but an agent doesn't pull 20 leads at a time it'll go through thousands of records in one run and your credits will be gone before it's even finished the new rule is that seat based and credit based pricing will collapse agents just want an api key they want to connect to your apis and do what they need to do without anyone sitting there counting credit and the humans running those agents want a fixed, predictable price so that they don't end up with a runaway bill.
Let me show you what this actually looks like when an agent does the work. First, I'm connecting multsets to Claude so that it has somewhere to pull the data from. This only takes a couple of minutes, and you only do it once.
I'll leave a link to multsets in the description so you can sign up and follow along. Then, I'll tell Claude to use a ready -to -send list skill and give it one sentence. I sell RB2B, which shows you which company visit your website.
More than 5 ,000 B2B teams use it. Find me growth and marketing leaders at US software companies with 11 to 500 employees and get me verified work emails that I can send to today. First, it shows me who it's going to search for.
Marketing and growth leaders from US software companies with 11 to 500 people. Over 10 ,000 people match and it pulls five real ones as a preview so I can check that it understood me. Then I tell it to pull 1 ,000.
On a typical data tool at $99 for 4 ,000 credits, that's a fifth. your whole month gone in one run an agent won't run this at once it'll run it again tomorrow for a different segment and one the day after that that's why the unlimited contact piece is so important when using agents then it grades every email and throws out anything that's not safe to send so it checks the title and the company fit and that it's a real work email and not a gmail and then it's a grade a verified in the last seven days and an email that actually belongs to a person and on this run 450 of the 997 people it pulled made it all way through that flow more than half of this list gets thrown out and that's the point sending to bad emails is how you burn your domains so you want this filtering to happen and then it writes these emails too it gives me three different versions to test one talks about the job they're responsible for one leads with proof that works and one just asks them a simple question and they're all short under 75 words with one question at the end and in my experience that's what actually gets replies short
clear and to the point so if agents are your users like i just showed you you can't charge per seat or per credit i had an agency owner ask me if he needed 75 mold sets api keys for his 75 clients i told him no you only need one and he thought i was insane but i can't say i'm built for agents and then turn around and charge per key or per seat that's the old model that's also why i cut the price the way i did at the start of this video when i first priced mold sets a few of the biggest clay agencies told me there were probably 10 to 30 people would pay $500 a month for it.
That's not a business. So at $497 a month, people think about it. At $97, they don't.
And I think the number of people who will buy something at that price is around a hundred times bigger than at $500. Maybe it's a thousand times bigger than at $500. And on top of that, if you're priced below what it costs your competitors to acquire a customer, they can't make their funnel work.
They have to build more and they have to charge more just to pay for ads and for salespeople. Now I offered a $27 entry plan because I think in Individual sales reps will actually put it on their personal card.
And if they're working inside of ClaudeWeb or ChatGPT, they don't actually need the rate limits of the $97 plan. So I think it's opening up an entirely new TAM beyond the original TAM I was going after having a sub 30 plan. What I just did with Claude used to take me an SDR team, a data contract, and most of a week.
Here it required one sentence from me in a few minutes, which brings me to the fourth rule. Rule number four is one or two people to 10 million AR. Every SaaS playbook I learned said the same two things.
If customers are churning, build more features. And if you want to grow, hire more people. More features meant more stickiness and more people meant you could build and support all of it.
I don't think that's how the best companies get built anymore. In 2027, the best in class is going to look like one or two people at 10 million ARR and beyond. Think about what an early stage startup actually is.
It's building a product, which AI helps more than absolutely anything else. then it's the cycle of marketing and improving the product which ai can also stream incredibly ai has amplified individual contributors to an incredible degree so the one or two person 10 million era startup is going to be a commonplace by 2027. rb2b is the perfect example of this we never kept adding features rb2b is still the simplest possible version of a pixel that identifies a website visitor and then pushes them to slack our churn has stayed horrible but the business is now at 10 million ARR.
Our engineers haven't even touched the code in six months, and it takes one person under two hours a day to handle all of the escalated customer interaction. Churn isn't solved, but I don't care. It's super profitable and barely needs anything from us.
So the reason it works is because we kept it simple. A complex product needs humans to sell it, onboard people, and support it. A dead simple one can be run almost entirely by AI.
And when a business can run like that, raising venture capital makes less sense than ever rule number five is bootstrap for cash flows and freedom for as long as i've been in sas the default move once you had traction was to go raise money today i think that's one of the fastest ways to kill your company almost All of the venture money is going to the foundational model companies and the ones that need to build data centers.
It's always been a winner -take -all economy in BC, but now it's a winner -take -all the capital too. The top 0 .1 % of startups are sucking up all of the dollars and the little guy is getting nothing. That creates two problems at once.
First, it's harder than ever to raise money to fund your ideas in the first place. Second, it's harder than ever to actually get liquidity as a founder. That means someone needs to buy your company and the valuations of...
Software companies outside of OpenAI and Anthropic, of select few other AI companies, are collapsing. So being the little guy in the VC game has become impractical. You just shouldn't do it.
So instead, bootstrap and build a company that pays you from your customers. At 10 Million ARR Bootstrap, you and your co -founder, even if you have a handful of employees, can each make over a million dollars in salary and dividends every single year. At Retention .com, we distributed $5 million in cash to our founders on the way to 10 Million ARR.
was pre -AI. I've been saying this for years and going into next year, it might be more important than ever. The happiest founders will be the ones building cashflow for themselves.
That's why I'm totally fine with mold sets being small. Is it a VC investable business at $27 a month? Not at all.
It's capped, but I don't need it to be a unicorn. I need it to be profitable. But breaking rules doesn't always work.
Now, I want to be honest here because not every rule I've broken has paid off. A couple of years ago, I tried to sell RB2B in a way no SaaS company really had other than Russell Brunson of ClickFunnels.
It was an influencer -led, one -to -many selling funnel. I bundled courses, community, and software together. I spent around $100 ,000 making four courses, and I was doing webinars every single week with a Hormozy -style value stack and close at the end.
I spent six months trying to perfect it, and then I walked away from the whole thing. And here was the problem. The people who were perfect for the community weren't the people who could actually buy the software.
They were too early. they had no traction but they were looking to learn but you need website traffic to get value from rb2b and almost nobody in the community had any so i tried to break the traditional way of selling software and it bit me so some things haven't changed no matter what ai does you still have to sell to people who have the actual problem you have and even the molt sets price cut didn't come from me being clever it came from customer discovery and from agencies telling me what they'd actually pay for the solution the market knows and you just have to ask it.
So who should be scared and who wins? If your SaaS is basically a dashboard that holds data and reports on it, you should be terrified people won't need that in the future they'll be using a headless data asset with some kind of ai sitting on top of it and they'll just ask what they want to see in plain english the same goes for software that anyone can copy and is really just a good experience for humans over the next five to ten years those companies are going to be under a lot of threat on the other hand there are a few kinds of companies that i think are set up to win first is highly verticalized software for businesses that have a real world deployment component picture quickbooks made for pool clean AI first, handling their whole business.
The human still has to go out and do the job, but the software is the digital layer that runs everything else. Anything like that, I believe, is safe for a long time. Then I think companies will thrive who build tools that actually help agentic workflows.
And companies with network effects. Cloud Code can't recreate the value of Uber or Instagram. And lastly, I think data companies will crush.
So, if you're early stage, maybe at 1 to 3 million AR right now, I'd do one of two things. The first option is... to look at what's happening with agents in the market you serve, and then build something that makes those agents work better.
The second option is to do what we did a year ago and set a deadline. We said that within 36 months, we would be a pure play data company. No user interface in our software.
We'll be providing data to agents, and that will be our entire value proposition. This was across all three of our business lines, retention .com, RBDB, and MaltSets, and we are working towards that reality. Either way, start with one question.
If your next customer is an agent, does your company need to look like? I hope this video is valuable and if you want to see exactly how I'd scale a SaaS to 10 million a year with Claude if I started over today, watch this video next.
The Hook
The bait, then the rug-pull.
A founder who runs $30 million a year in software companies opens by cutting his own price twice, then argues the whole SaaS rulebook is about to break. Five rules follow, one demo, and one honest failure.
Named ideas worth stealing.
00:47list
The five new rules of SaaS
- Build for agents, not humans
- Headless software takes off
- Seat and credit pricing collapses
- One or two people to 10 million ARR
- Bootstrap for cash flow and freedom
The talk's organizing frame for how agents change building, pricing, hiring and funding.
Steal forauditing a SaaS roadmap against the agent shift
10:56list
Who wins in the agent era
- Highly verticalized software with a real-world deployment component
- Tools that help agentic workflows
- Companies with network effects
- Data companies
The categories he believes are safe, versus dashboard-only or easily copied software.
Steal forchoosing what to build next
How they asked for the click.
VERBAL ASK
12:24next-video
“If you want to see exactly how I'd scale a SaaS to 10 million a year with Claude if I started over today, watch this video next.”
Soft next-video handoff after a closing question; the product link sits in the description.
MENTIONED ON CAMERA
FROM THE DESCRIPTION
PRIMARY CTAWhere the creator wants you to go next.
OTHER LINKSAlso linked in the description.
Visual structure at a glance.
