Everyone on X keeps repeating that SaaS is dead. AI killed it, anyone can clone your product in a weekend, margins are cooked.
Greg Isenberg looked at the same facts and came to the opposite conclusion: this is the best time in history to build software, and SaaS is minting millionaires again.
In March he dropped a 25 minute video laying out his full AI SaaS playbook: 30 steps, from picking a niche to becoming the default software layer for it. No fluff episode, no guest, just the plan. It's sitting at 101,000+ views.
Greg has earned the right to an opinion here. He advised TikTok and Reddit, built and sold three venture-backed startups, and now runs Late Checkout plus the Startup Ideas podcast.
Here's the video, then my breakdown of all 30 steps with the numbers that matter.
Why This AI SaaS Playbook Exists
Greg opens with the setup: building software has never been cheaper, building an audience has never been cheaper, and the buyers on the other side hold billions of dollars of purchasing power.
Meanwhile public SaaS companies are down 30 to 50% from their all-time highs. Investors are scared that anyone can vibe code a Salesforce competitor, and that the per-seat pricing model is losing its grip.
Greg's read: that fear is exactly the opportunity. SaaS isn't dying, it's evolving into something an individual founder can win at. Not by fighting Salesforce, but by owning one tiny workflow the giants will never bother with.
That's the whole thesis. The 30 steps are how you execute it. I've grouped them into six phases so you can actually use this.
Steps 1-5: Pick a Sub-Niche and Find Where Money Changes Hands
Step 1 is the classic Greg move: start with a niche inside a large market. Finance is the market, FIRE for Gen Z is the niche. He points to tools like ideabrowser.com for hunting these down.
Don't build for the huge market. That's where the venture guys play. You're building a money-making startup: $100K a month first, maybe more later.
Steps 2 through 5 are where most people get lazy, and it's the part I'd tattoo on every founder's arm:
- Map the workflow start to finish. His example is a local roofing company: check new leads, respond and qualify, schedule the site visit, drive out, take photos, estimate materials, write the quote, negotiate. Every step, in order.
- Highlight where money changes hands. Agreeing on the price, collecting the deposit, ordering materials. That's where software gets to take a piece of the pie.
- Circle the repetitive mechanical steps. Checking leads across the website form, Google ads and Facebook messages might eat 10 minutes a day.
- Quantify what those steps cost. If the owner makes $250K to $1M a year, their time is worth something like $400 an hour. Save them 50 to 150 hours a year and you can put a dollar figure on your product before you write a line of code.
How do you learn the workflow? Three ways: call owners and ask, be the owner (or work for one), or throw AI at it. Greg literally suggests asking Manus, Claude Code or ChatGPT to map it for you.
Steps 6-10: Build the Media Company Before the Product
This is the part most technical founders skip, and it's a third of the playbook.
Step 6: create scroll-stopping content around the workflow you just mapped. Pick ONE channel (Instagram, TikTok or X) and go all in. Use AI as your marketing director: ask it for non-obvious content ideas in your niche, make it build you a calendar, push it past the generic first answers.
Step 7: study what gets saved, replied to, and DM'd. Greg has been posting on Instagram for 6 to 8 months and pulls up his own analytics in the video: one recent post had 2,700 likes and 5,000 bookmarks. Bookmarks beating likes is the signal. That's utility.
Steps 8 and 9: double down on the organic angles that convert, then run paid ads on the proven winners. If a piece worked organically, there's a good chance it works as an ad with decent ROI. And "viral" is relative to your niche. 60 likes, 80 likes, 120 likes can be viral for local roofing software.
Step 10: collect emails from day one. When sales dip, you can't force the algorithm to show your content. You can always send an email.
→ The point of all this: by the time your product exists, you already have an audience of exactly the people it's for.
Steps 11-13: Do the Work Manually (Yes, Really)
Step 11 is the least sexy advice in the video and probably the most important: execute the workflow manually.
Greg is blunt about it. Most of these future SaaS businesses actually start as service businesses. You are the software at the beginning. You check the leads, you write the quotes, you chase the deposit.
I know that's not what you signed up for. You wanted the 24/7 product with the high multiple and the exit. But you can't automate a job you've never done. (I learned this one the expensive way with my first SaaS.)
Step 12: document every single step precisely while you do it.
Step 13: separate judgment calls from mechanical tasks. AI is incredible at mechanical work and hit-or-miss on judgment. Your automation roadmap is literally just the mechanical column of that document.
Steps 14-19: Turn the Mechanical Work Into Agent Workflows
The product doesn't show up until step 14 of 30. (That ordering is not an accident.)
The build phase, compressed:
- Step 14-15: turn the mechanical tasks into agent workflows and build agents that complete them end to end.
- Step 16: connect agents to real tools through MCP. Email, Slack, the CRM, Stripe. An agent without tools is a chatbot.
- Step 17: add orchestration, retries and checks.
- Step 18: store user preferences and long-term memory. This becomes your moat later.
- Step 19: launch narrow, with onboarding that grabs the right data without asking a thousand questions.
On step 17, Greg pulls up a tweet from his friend Scott Belsky (previously chief product officer at Adobe): "the orchestration layer is the new interface layer." We're entering the era of conductors. The founder who masters coordination between agents owns the flow.
If you want to see what this looks like in practice, I broke down Ras Mic's system for shipping 15 features in parallel with agent workflows in my software factory piece. Same philosophy, running code.
Steps 20-23: Price Per Task, Not Per Seat
This section alone is worth the 25 minutes.
Step 20: publish measurable proof. Hours saved, revenue recovered, quotes sent. Real numbers from real users.
Steps 21 and 22 are the pricing shift: move from per-seat pricing to per-task, then to pricing by outcomes. Greg's example: "complete this workflow for me, and every run costs $200."
This is exactly why those public SaaS giants are down 30 to 50%. Their per-seat model was built for a world where humans did the work inside the software. When the software does the work, customers want to pay for the result.
And you, the small founder, can make that switch on day one. No board, no legacy revenue to protect, no sales team comped on seats. Your cost structure is tiny, so you can undercut on price AND charge more per unit of actual value.
Step 23: as trust and workflows stack up, raise prices. Value-based, not cost-plus.
Steps 24-30: Compound Until You're the Default
The last phase is about turning one working workflow into a compounding machine:
- Step 24: expand into adjacent workflows (build them yourself; Greg notes buying companies makes less sense when AI makes building this cheap).
- Step 25: orchestrate multiple agents across the whole customer lifecycle.
- Step 26: build switching costs with data and memory. Every month of usage makes leaving more painful.
- Step 27: turn power users into public case studies. Not a tweet. Send a film crew, then put ad spend behind the footage.
- Step 28: hire operators from inside the niche, funded by profit.
- Step 29: reinvest into distribution and product depth.
- Step 30: become the default agent layer for the sub-niche.
That end state is the millionaire-minting part. One founder, a handful of operators, agents doing the mechanical work, owning a workflow nobody else cares enough to win. The data backs this trajectory up too: I covered the numbers behind the rise of one-person software companies in the million dollar solo founder.
My Take After Watching It Twice
What I love: the order. Niche → workflow → media → manual service → agents → outcome pricing. Most founders run this backwards, building the product first and praying distribution shows up later.
What I'd push back on: the 30 steps read like a checklist you could finish in a quarter. You can't. Steps 6 to 10 alone (building the media engine) took Greg 6 to 8 months of daily posting to develop intuition, and he's Greg Isenberg. Budget a year of unsexy reps.
Also missing: what happens when the manual service phase starts making real money and you're tempted to just... stay a service business. It's a trap I see constantly with founders in my mastermind. The playbook only compounds if you actually make the jump at step 14.
But as a map of where SaaS is going (vertical, agent-powered, priced on outcomes), this is the clearest 25 minutes you'll watch this year.
FAQ
Is SaaS dead in 2026?
No. Public SaaS companies are down 30 to 50% from their highs because per-seat pricing is under pressure, but that's a repricing of the old model, not the death of software. Greg Isenberg's argument is that SaaS is evolving: smaller teams, vertical niches, AI agents doing the work, and pricing tied to tasks or outcomes instead of seats.
What is Greg Isenberg's AI SaaS playbook?
Pick a sub-niche inside a big market, map one owner's workflow start to finish, and find where money changes hands. Build an audience around that workflow before you build anything. Do the job manually as a service, document it, then turn the mechanical steps into AI agent workflows connected to real tools. Charge per task or per outcome instead of per seat, then expand into adjacent workflows until you're the default software for that niche.
How do I find a niche for an AI SaaS?
Start with a market that already moves serious money (finance, home services, legal, logistics) and slice down to a group with a specific repeated workflow, like FIRE-focused Gen Z investors or local roofing companies. Greg suggests tools like ideabrowser.com, talking directly to operators, or using Claude Code, Manus or ChatGPT to map niches and their daily workflows.
Do I need to know how to code to build this?
Less than ever. The playbook leans on AI coding tools (Claude Code, Manus, ChatGPT) for everything from niche research to building the agents themselves, and MCP makes connecting agents to email, Slack, CRMs and Stripe mostly configuration work. What you can't outsource is knowing the workflow cold, which is exactly why the manual service phase exists.
Steal Playbooks From Founders Who Already Did It
Greg's video is the map. The founders I interview are the proof.
Every week on the Profitable Founder Podcast I sit down with bootstrapped SaaS founders doing $100K to $10M a year and pull out exactly how they picked their niche, priced their product, and got their first customers. Real numbers, no theory.
