“SaaS is dead” is Greg Isenberg's provocation, not his literal conclusion. In his solo episode, the old business sells a cloud tool per seat; the new one gives the tool away, then charges when agents or people complete the work. His stronger point is that software creation is getting cheaper, while solving a buyer's recurring problem is still valuable. Whether a free tool is the right entry point remains a market and unit-economics question.
Greg's “smart SaaS” is a plausible model for a narrow, repeatable job: free software helps users start, an AI agent handles paid work, and a human service covers judgment or exceptions. Do not assume every tool should be free or that per-seat pricing has stopped working. Test whether buyers want the result, what it costs to deliver, and whether you can earn their trust without quietly turning their data into the product.
Watch Greg's SaaS 2.0 Argument
Source note, 30 October 2026: The supplied transcript and video are the source for Greg's predictions and examples. I checked Sequoia's services thesis, Stripe's pricing guide, Salesforce's current plans and FTC data-practice guidance for context. The proposed market shift is an opinion; none of these sources proves that all per-seat SaaS is ending.
What Changes When the Tool Is Easier to Build?
Greg describes SaaS 1.0 as hiring a team to make a useful tool, then charging each person who needs access. His SaaS 2.0 gives the tool away and captures value in the work performed around it. The distinction is helpful, but the build is only one cost. Reliability, distribution, integrations, support, privacy and sales do not vanish when coding becomes faster.
| Layer | Greg's older model | His proposed model | What still needs proof |
|---|---|---|---|
| Entry product | Pay per user to access the tool. | Give the core workflow away. | Does the free tool solve a real job often enough to earn repeat use? |
| Value captured | Software access and features. | Agent-delivered outcomes or human service. | Can the result be defined, measured and completed reliably? |
| Advantage | Software quality and switching costs. | Data, workflow integration, distribution and network effects. | Are these advantages real, permissioned and hard to copy? |
| Revenue | Predictable seats or subscriptions. | Usage, outcomes, service retainers or a mix. | Do gross margins survive model calls, exceptions and support? |
Greg's calorie-tracking example is hypothetical: the tracker is free, while an agent that produces personalized plans or a weekly human coach costs money. A useful question is not “can AI build the tracker?” but “will people return, trust the recommendations, and pay for the next layer?” Health-related recommendations also need appropriate professional review, clear limits and careful handling of sensitive data.
Three Ways to Charge for the Work
Greg names tokens, outcomes and recurring human service fees. Stripe's pricing guide adds a practical nuance: a seat subscription or hybrid plan can still be a good fit when buyers need predictable spend. A token is a cost meter, not automatically a unit of customer value.
| Price basis | When it fits | Main failure mode |
|---|---|---|
| Usage or tokens | Costs vary with volume and buyers can forecast or cap consumption. | Customers pay for retries, not necessarily useful work. |
| Verified outcome | Completion is unambiguous, measurable and attributable. | Disputes over what counts as success or who caused it. |
| Human service | The work needs judgment, exceptions or reassurance. | Labor cost can grow faster than revenue. |
| Hybrid | A predictable base covers access and support; variable work is metered. | Too many meters make the offer hard to understand. |
Sequoia argues for selling completed work rather than only a tool, especially where a company already outsources a defined task. That supports the outcome opportunity Greg points to. It does not establish that the tool must be free or that monetizing user data is necessary.
Three Places the Headline Breaks Down
- Seats still sell. Salesforce's current Sales Cloud page still lists per-user plans, including tiers with AI. Buyers pay for more than code: administration, integrations, service levels and accountable vendors. Greg himself expects enterprises to keep paying per seat.
- Free is not free to operate. Every free user can add hosting, model, support and acquisition cost. A product with many non-paying users and an expensive agent upsell can have worse economics than a modest paid tool. Measure contribution margin on the whole customer journey, not only the paid task.
- Data is not a hidden payment method. Greg suggests free users can contribute proprietary data and network effects, even mentioning eventual sale of data. That is a business hypothesis, not permission to repurpose customer information. The FTC warns against quietly broadening data-use terms. Design for data minimization, clear choice and trust from the start.
“One or two people can build what took 20” is Greg's directional claim, not a universal productivity ratio. His broader forecast that SMB and prosumer tools trend free is testable, but not a reason to discard a paying, retained customer base today. The durable insight is to price around a job worth doing, then choose the packaging that gives both sides a workable deal.
A Better First Test Than Giving Away a CRM
Pick one buyer and one task they already do weekly. Interview five people about the current workaround. Build only the free entry step needed to make the task legible; use sample data until you have explicit permission to handle real records. Offer a human-reviewed paid result to at most three testers. If nobody uses the free step twice, or paid work cannot be completed at a margin, stop before building a broad platform.
A useful test outcome might be “prepare and verify five follow-up drafts for a local business each week,” not “install a free CRM.” The first can be checked against a time budget and acceptance criteria. The second is a category with integrations, security and switching costs you have not yet earned the right to tackle.
Copy a Smart SaaS Pilot Brief
This original prompt tests the business model without assuming that free acquisition, data use or agent performance will work.
Free entry, paid result
One buyer, one job, two weeks.
Video Chapters
The model
Monetization
| Time | Topic |
|---|---|
| 05:12 | Charge for intelligence |
| 06:18 | Services as the next layer |
| 07:22 | SaaS is dead, long live SaaS |
Who pays
Sources and Perspective
Checked 30 October 2026. This is an analysis of Greg's proposed business model, not a claim that every SaaS company should change pricing. The linked primary sources document a services thesis, available billing models, existing per-seat products and data-use expectations. None is a controlled test of Greg's free-tool strategy.
| Question | Direct sources | Why it matters |
|---|---|---|
| What did Greg argue? | Full video and chapters | His predictions, free-tool example and enterprise exception. |
| Why sell the work? | Sequoia: Services, the New Software | A related investor thesis; not an instruction to make every tool free. |
| How could it bill? | Stripe pricing frameworks, Salesforce plans | Usage, outcomes, hybrid and still-active seat pricing. |
| What about user data? | FTC data-practice warning | A counterweight to treating data as an invisible price. |