$2 trillion evaporated from SaaS in a single month.
Salesforce dropped 25%, Adobe 38%, Intuit 46% — and "SaaSpocalypse" started trending on trading desks.
Then a16z dropped the opposite take. "This isn't the end. It's the start."
Why is a16z going against the panic?
Setting the scene first: Q1 2026 saw the iShares Expanded Tech-Software ETF (IGV) drop 30%, its worst quarter since Q4 2008. Names like Salesforce, Adobe, Intuit, ServiceNow, and Veeva fell 25-46% in a matter of weeks.
The trigger was Anthropic's Claude Cowork launch on February 24, 2026. A single AI agent demoing legal, finance, support, and project management work convinced Wall Street that per-seat pricing was finished — $285B evaporated within 48 hours.
In the middle of that panic, a16z's growth partners Alex Immerman and Santiago Rodriguez published their counter on March 2: "AI Will Eat Application Software" — and that's actually good news, they argue.
Their one-line thesis: "Code was never where the value lived." If code itself were the value driver, open source would have already killed Salesforce and Adobe. Cheaper offshore developers would have eaten their lunch decades ago.
The real value SaaS has built up isn't the code — it's the seven types of moats stacked on top. Hamilton Helmer's "7 Powers" framework: Switching Costs, Network Effects, Scale, Brand, Cornered Resources, Process Power, Counter-Positioning.
So which moats actually survive?
Here's the compressed take: even as AI turns code into a commodity, certain moats get stronger. The code-era moats and AI-era moats aren't the same thing.
| Moat type | Code era (commoditizing) | AI era (where to bet) |
|---|---|---|
| Switching Costs | Hard lock-in tool | Friction drops — must hold via real value |
| Cornered Resources | Secondary asset | Proprietary data = main moat (Bloomberg, Abridge, OpenEvidence) |
| Process Power | Lives in heads and manuals | Encoded into AI, becomes non-replicable (Harvey, Hebbia) |
| Network Effects | User graph | Data flywheel accelerates it (Salesforce, Figma) |
| Brand | One trust signal among many | More important in an infinite-options world |
| Counter-Positioning | Occasional disruptor | The standard AI-native entry weapon (Decagon) |
The one to watch is Process Power. Hebbia's George Sivulka calls it "process engineering" — when Harvey deeply learns a specific firm's templates, review processes, and institutional preferences, a new entrant can't replicate that overnight even if code costs zero.
Cornered Resources gets even more dramatic. Same model, different training data — completely different output. Bloomberg owns finance data, Abridge owns clinical encounters, OpenEvidence owns medical literature, VLex owns legal corpora. You can copy code; you can't copy data.
Then there's Counter-Positioning. The strategy of entering with a business model your incumbent can't follow. Decagon entered the Zendesk-at-$19-per-seat market with "$0.99 per conversation or pay-per-resolution" — and closed a Series D at a $4.5B valuation in January 2026. Zendesk can't match because it would have to cannibalize its own revenue model.
The key insight
AI commoditizes code but inflates the value of everything else — data, process, brand, counter-positioning. The market doesn't die; it bifurcates. Winners get more durable; "code-only" companies disappear.
HarbourVest reaches the same conclusion from the PE side. Their four survival conditions — mission-critical systems of record, vertical SaaS, embedded fintech, R&D velocity — are all "non-code" assets.
Five things builders and operators should audit now
- Separate code-as-moat from data/process-as-moat
If you open-sourced just the code portion of your product tomorrow, would the business survive? "Yes" means your real moat is in data, process, and brand. "No" means you're in wrapper-trap territory. - Pilot outcome-based pricing on a small SKU
Pick one SKU at 5-10% of revenue and shift it from per-seat to per-conversation, per-resolution, or per-outcome. Measure churn, ARR, gross margin impact quarterly. Decagon's playbook works because the seat model leaks revenue when AI agents replace humans. - Asset-ize your proprietary data
Catalog the data flowing through customer workflows daily that no foundation model can ever see. That's exactly what Bloomberg, Abridge, OpenEvidence, and VLex did. Lock down training rights, data ownership clauses, and consent structures at the contract level. - Codify Process Power
Onboarding playbooks, review cycles, exception handling — anything that lives only in operator heads needs to become a company asset. SOPs, heuristics, decision trees, and eval datasets are now more valuable than the source code. - Monitor your category's "Decagon"
Competitors who match your price for better features aren't the real threat. Competitors who sell the same outcome with a fundamentally different pricing model are. Audit AI-native entrants in your category every quarter.
Caveat
This is a16z's view, and a16z is an investor in Harvey, Hebbia, and Decagon — the very AI-native challengers they argue will win the bifurcation. So the answer to "who wins?" already has chips on the board. The 7 Powers framework itself is still useful as a tool — just treat the analysis as analysis and validate the conclusion against your own market.
Going deeper
a16z — Good news: AI Will Eat Application Software The original piece by Alex Immerman and Santiago Rodriguez. The reinterpretation of 7 Powers for the AI era is the core a16z.com
HarbourVest — The Software Industry's Great Reset How AI is changing SaaS "physics," and the four survival conditions, from a private equity lens harbourvest.com
Taskade — The SaaSpocalypse: $285B Wiped, AI Agents Rising Minute-by-minute breakdown of the 48-hour Claude Cowork aftermath. Three survival scenarios compared taskade.com
TechCrunch — Decagon completes $4.5B tender offer The textbook counter-positioning case. Background on the a16z/Coatue/Index round techcrunch.com
eesel — Decagon vs Zendesk AI comparison Per-seat vs per-conversation pricing models analyzed from the buyer's perspective eesel.ai
Lenny's Newsletter — Hamilton Helmer interview on 7 Powers The framework author himself. The clearest answer to "what is a moat, really?" lennysnewsletter.com
