No, B2B SaaS is not dead. It commoditizes wherever a product's core value is just a model call away from replication, and it grows wherever the product owns proprietary data, encoded domain logic, or a workflow lock a general AI model cannot copy. Which bucket your product lands in comes down to three tests.
Last updated June 2026.
| SaaS type | At risk from AI? | What makes it defensible (the moat) |
|---|---|---|
| Thin AI wrappers / single-feature tools | High | None, unless they add proprietary data fast |
| Horizontal point solutions (illustrative) | Medium-high | Switching cost, integrations, workflow depth |
| Vertical / domain SaaS (illustrative) | Low | Encoded domain knowledge, proprietary data |
| System-of-record / data-platform SaaS (illustrative) | Low | Workflow lock, accumulated historical data |
| Commodity utilities behind a login (illustrative) | High | Distribution alone, easy to route around |
Total software spend is still growing 15% this year, the fastest pace in a decade, with Gartner projecting the market rising from $1.2T to $1.4T, as reported by SaaStr in July 2026. That number sits next to an uncomfortable one. SaaStr's own framing of the same data is that spend is up even as “half of SaaS is still dying.” Both are true, describing different halves of the same market. The three tests below tell you which half your product is in.
Why “Is B2B SaaS Dead?” Is the Wrong Question
SaaS does not live or die as a category. Individual products pass or fail an AI-resilience threshold, and the category-level question just averages two very different outcomes into noise.
The live search results for this query read like a shouting match. On one side, the doom read: Jason Lemkin (@jasonlk) posted on March 29, 2026 that “much of traditional SaaS is dying, in likely terminal decay,” a thesis RnD Ventures and a stream of Reddit threads echo. On the other side, the cope read: IDC calls SaaS “metamorphosing, not dead,” and a wave of vendor explainers land on some version of “it evolves.” Both sides answer an identity question about the category. Neither answers the question a founder actually needs answered: a product question.
The same week Lemkin posted his terminal-decay line, two other operators pushed back, in public, on the same platform. Alex Lieberman (@businessbarista) wrote on February 15, 2026: “SaaS isn't dead. Bad SaaS is dead, and it always has been.” SaaStr's own account (@saastr) posted on February 14, 2026: “Stop listening to the SaaS is dead crowd.” Same month, same debate, opposite conclusions, from people who all watch this market for a living.
That contradiction isn't a sign the debate is unresolvable. It's a sign the debate is asking the wrong scope. Lemkin, Lieberman, and SaaStr aren't disagreeing about SaaS as a category, they're looking at different products and generalizing from what they see. The useful question isn't “is SaaS dead,” it's “is my SaaS resilient on the three axes AI actually attacks.” The rest of this piece is a diagnostic for answering that, and it slots into the broader stage-by-stage AI growth playbook for solo founders as the first gate to clear before picking what to ship next.
The 3 AI-Resilience Tests for SaaS
AI-resilience is the degree to which a product's core value survives once a capable general AI model is cheaply available to the customer. A product that fails this test isn't “bad SaaS.” Its value was never in the software, it was in the fact that a model this cheap didn't exist yet.
Test 1: The Replacement Test
Does a cheap general model reproduce your product's core value directly from a prompt? If a customer's existing AI assistant can do the thing your product does with no meaningful loss of quality, you're commoditizing regardless of today's growth rate. A meeting-notes summarizer fails this test the day the customer's AI assistant ships native summarization, because summarization was never the moat, it was a stopgap. A vertical compliance platform encoding a decade of regulatory edge cases passes, because the moat is the encoded knowledge, not the act of writing a paragraph.
Test 2: The Proprietary-Data Test
Do you own data the customer can't get elsewhere, that a general model was never trained on? Own-data is the durable moat, the axis separating a wrapper from a platform, echoing the durable-moat argument a16z's enterprise investing team makes about AI-native software more broadly. A product that only reformats a customer's own inputs isn't accumulating anything a competitor's chatbot couldn't replicate from scratch. A product capturing structured outcomes across thousands of interactions is building something a fresh model call can't shortcut.
Test 3: The Distribution / Workflow-Lock Test
Are you embedded in a workflow or distribution channel that survives a feature-level threat? Switching cost is the third moat, and it buys time even when a competitor ships your exact feature. PagerDuty is the illustration Lemkin's own reporting supplies: per Lemkin/SaaStr, the company has held roughly 15,000 customers flat for four years while still running at approximately $500M ARR (unverified against PagerDuty's own filings as of this writing, flagged for primary-source confirmation). That's workflow lock. Customers stay because leaving means re-wiring an incident-response process, not because the product keeps winning new logos.
Pass zero of these three tests and you're the SaaS that dies, slowly, as margin gets arbitraged away by a $0.01 API call. Pass one or more and you have a specific, defensible reason to grow. That's the bridge to what the evidence actually shows.
What Actually Declined vs Grew in the AI Era (2026 Evidence)
The pattern in the 2026 data isn't “AI killed SaaS.” Products failing the Replacement Test are under pressure, and products passing the Proprietary-Data or Workflow-Lock tests are absorbing AI as a tailwind.
Total software spend grew 15% this year, the fastest pace in a decade and up from 12.8% the year before. A dying category does not post its fastest growth rate in ten years.
SaaStr's own framing is that the growth and “half of SaaS is still dying” aren't contradictory. Spend is consolidating into fewer, more defensible products, not evaporating.
On the decliner side, Forrester's February 2026 analysis (label: projection, not audited fact) splits the market into an at-risk group, horizontal point-solutions with low switching costs, and a survive group, vertical and domain vendors and proprietary-data holders such as Epic and IQVIA, mapping directly onto the Replacement and Proprietary-Data tests. Separately, per Lemkin/SaaStr, Salesforce's headcount fell from roughly 9,000 to roughly 5,000 in go-to-market functions (unverified against Salesforce's own earnings statements as of this writing, flagged for the editor). Worth keeping two stories apart: cutting headcount because AI made a team efficient isn't the same as losing revenue because a product died. Salesforce is running leaner, not disappearing, and conflating the two is what makes the doom takes read sloppy.
On the grower side, Forrester projects total SaaS spend rising from $318B in 2025 to $576B by 2029, with the vertical-SaaS slice alone moving from roughly $133.5B to $194.0B (label: projection). The mechanism behind why thin wrappers commoditize fastest is a pricing one: general-model token costs have fallen sharply, on the order of 100x or more for comparable capability over the past two years. When the underlying capability gets that much cheaper, a product whose entire value is “I call the model for you” loses pricing power on the same timeline. A product built on accumulated data or embedded workflow doesn't. The model call was never the product. The same AI-era shift is reshaping how buyers even find a product in the first place, which is why we run our own AI search visibility audit rather than assume classic SEO still covers it.
So Is It Dead? The Founder Debate, Resolved
Classic SaaS is not dead. Products that are just a thin feature wrapped around someone else's model are dying, and that's been true since before generative AI existed. AI just moved the timeline up.
Run the three named voices back through the framework and something clicks. Lemkin's “terminal decay” read is correct about the specific companies he's describing: flat growth, rising acquisition cost, no data or workflow moat to fall back on once a feature commoditizes. Lieberman's “bad SaaS is dead, and it always has been” is correct about the mechanism. SaaS has always died when it stopped solving a real problem better than the alternative, AI just made that alternative cheaper. SaaStr's “stop listening to the SaaS is dead crowd” is correct about the aggregate: total spend is up 15% this year, and treating that growth as fake because some products inside it are dying is a category error.
None of the three is wrong. Each is looking at a different slice of the same distribution, sorted by the same three tests. A product that fails the Replacement Test, has no proprietary data, and has no workflow lock is exactly what Lemkin watches decay. A product that passes even one test is exactly what Lieberman and SaaStr mean when they call the “SaaS is dead” framing too broad. The resolution isn't a fourth opinion, it's a filter that sorts the first three into where they're each right, and it's why the binary question fails as a planning tool: it tells a founder nothing about their own product. The three tests do.
What a Solo Founder Should Do This Week to Add an AI Moat
You can't out-distribute Salesforce, and competing on scale against a nine-figure sales budget isn't the fight worth having at $5K MRR. What you can do is pass at least one of the three tests this month, as a single-product, one-person operation, no distribution war required.
Replacement Test move.Find the part of your product that is, honestly, just a model call with a UI around it, and stop competing there. If a feature is “summarize this” or “draft this email,” assume a customer's own AI assistant will do it free within the next model generation. Move your differentiation to what a prompt can't replicate: judgment calls, domain-specific rules, accumulated exceptions. This is an afternoon of honest product audit, not a rebuild.
Proprietary-Data Test move.Start capturing a dataset only your product's usage produces, structured customer outcomes (which configurations actually worked, not just which were tried), or labeled edge cases a general model has never seen because they live inside your customers' private data. No data science team required, just a decision this sprint about which field to log.
Workflow-Lock move. Embed one layer deeper into a workflow your customer already runs daily, so switching cost rises even if a competitor ships your headline feature next quarter. Become the system of record for one object customers query, or ship one integration that makes leaving mean re-wiring three other tools, not just canceling a subscription. The deeper mechanics of a proprietary-data moat live in our vertical AI agent moat playbook.
Pick one axis, not all three, and ship the smallest version of that move this week. A founder who leaves with one chosen test and one action is in a stronger position than a founder who leaves with an opinion about whether SaaS survives as a category. Our breakdown of AI-native pricing models covers how per-token cost exposure changes the math once your product leans on a model call.
Want the next diagnostic in your inbox before you have to Google your way through it? Subscribe below.
Frequently Asked Questions
Is B2B software dead?
No. B2B software as a category isn't dying, but products offering no more value than a well-prompted general AI model are commoditizing fast. Products with proprietary data, encoded domain expertise, or deep workflow lock grow alongside AI, not despite it.
Will AI take over B2B sales?
AI is taking over mechanical work (call summaries, follow-up drafting, lead scoring), but it isn't closing hesitant enterprise deals alone. As 1up.ai frames it, AI tools can organize a sales process but won't convince a skeptical buyer to sign a contract carrying real budget risk. Trust-building and negotiation stay a human function.
What is the future of B2B SaaS?
The future splits along the present line: products passing at least one AI-resilience test keep growing, while products with no moat beyond “I call a model for you” commoditize on a shortening timeline. Forrester projects total SaaS spend growing from $318B to $576B between 2025 and 2029 even as this sorting happens.
Which SaaS is safest from AI?
Vertical and domain-specific SaaS with proprietary data, and system-of-record platforms with deep workflow lock, per Forrester's February 2026 risk split and the risk-matrix table above. Safety comes from the moat, not the label.
Is AI eating SaaS?
AI is eating the SaaS that was never more than a convenience layer over a model call, and growing the SaaS with proprietary data or a workflow lock a model cannot copy. As a category-wide claim, “eating SaaS” overstates a trend happening product by product.
What does SaaS apocalypse mean?
Shorthand, popularized by pieces like Forrester's “SaaS-pocalypse” analysis, for the idea that generative AI collapses pricing power for software built on thin, easily-replicated features. It doesn't mean every SaaS company fails, it means products with no data moat and no workflow lock face margin compression faster than in prior software cycles.
Sources: Forrester, “SaaS As We Know It Is Dead: How To Survive The SaaS-pocalypse” (Feb 2026); Jason Lemkin/SaaStr, LinkedIn (Mar 1, 2026); SaaStr, “Tired vs Wired... why software spend is up 15% yet half of SaaS is still dying” (Jul 2026, data via Gartner); Medium/Activated Thinker, “Everyone Says SaaS Is Dead” (Mar 12, 2026); IDC, “Is SaaS Dead? Rethinking the Future of Software in the Age of AI” (Dec 2025); jasonlk, businessbarista, and SaaStr on X (Feb-Mar 2026). Related reading: our AI-native SaaS growth playbook for solo founders and building a compounding growth loop.