DEV Community

Cristian Tala
Cristian Tala

Posted on

The Software Meltdown: Why the Per-Seat Pricing Model Is on Its Last Legs

The Software Meltdown: Why the Per-Seat Pricing Model Is on Its Last Legs

Atlassian -63%. HubSpot -48%. Workday -47%. Figma -49%. Snowflake -40%. This isn't a market correction. It's the market processing a structural truth that the SaaS sector has been ignoring for two years.

I'm Cristian Tala — I founded and sold a Chilean fintech (Pago Fácil) for $23M. Now I invest in startups and build with AI agents.

When I saw the Software Meltdown chart this week, my first reaction wasn't panic. It was recognition.

I've spent 15 years in the tech and investment ecosystem. I founded a fintech, sold it, and invested in over 30 startups. I've seen cycles. And this one doesn't resemble any of the previous ones.

The Numbers from April 9, 2026

The chart compiled by @speculator_io shows the sector's state as of April 9:

Company YTD Drop Drop from 52-week High
Atlassian (TEAM) -63.76% -75.72%
Asana (ASAN) -58.39% -69.84%
monday.com (MNDY) -57.81% -80.24%
Figma (FIG) -49.96% -86.66%
HubSpot (HUBS) -48.84% -69.90%
Workday (WDAY) -47.48% -59.05%
Intuit (INTU) -45.75% -55.96%
Snowflake (SNOW) -40.02% -52.98%
Salesforce (CRM) -35.68% -42.48%
Adobe (ADBE) -34.58% -45.76%

The average drop is 40.3% year-to-date. The IGV software ETF fell more than 24% in the first quarter alone.

The best performers in this index: Cloudflare (-3.64%) and Zoom (-2.67%). That data point isn't random. I'll explain why later.

Why This Started in February

The specific catalyst was in February 2026, when Anthropic launched Claude Cowork — a demonstration of how AI agents can automate knowledge work that previously required multiple people: legal drafting, financial analysis, project management, lead qualification.

The market didn't take long to draw the obvious conclusion: if an agent can do the work of 10 people, the number of software seats you need collapses.

Thomson Reuters fell 15.83% in a single day. LegalZoom 19.68%. Short sellers reached levels not seen since 2016.

Atlassian is already experiencing this in its numbers. The company announced layoffs of 10% of its workforce (1,600 people) in March 2026, redirecting resources toward AI. CEO Mike Cannon-Brookes acknowledged that AI "changes the skill mix required" and reduces roles in some areas. The stock hit new 52-week lows between $67-69, with a 57% YTD decline.

Note what Atlassian didn't say: that revenues are declining today. Cloud revenue grew 26% year-over-year in Q2 FY2026. The problem isn't the present — it's the anticipation of the future.

The Model That's Breaking

Classic SaaS was built on a perfect equation:

More work = More people = More seats = More recurring revenue

This equation worked for 20 years because the only way to scale human work was to hire more humans. Every new employee was a guaranteed new seat.

AI broke the equation.

According to Gartner data, the per-seat pricing model dropped from 21% to 15% of enterprise adoption in 2025. The "outcome-based" model (payment for results) went from 15% to 40% of enterprise contracts in the same period. By 2030, Gartner projects that at least 40% of software spending will be usage-based, agent-based, or outcome-based — not per seat.

Goldman Sachs published its "AI Impact Framework" in February, identifying SaaS companies with the highest displacement risk according to six factors: orchestration risk, monetization exposure, system-of-record ownership, data integration moat, AI execution capability, and budget alignment.

The bank compared the risk of the most vulnerable companies to that of newspapers in the digital age: businesses with solid models that became obsolete not because the product was bad, but because the monetization mechanism stopped making sense.

Why This Correction Is Different

The market has experienced three major tech corrections since modern SaaS exists: the dot-com crash of 2001, the 2008 financial crisis, and the post-COVID crash of 2022. I studied the first two in retrospect — my formal work life began in 2010 as a teacher and in companies from 2011. I experienced the 2022 crash front-row as an active founder and investor.

Each time, the best businesses bounced back because the problem was the price, not the model.

The 2022-2023 decline was a valuation correction. Companies kept growing revenues — just at lower multiples. The market paid 50x P/E during the zero-rate boom and adjusted to 20x when rates rose.

This time is different. The companies falling aren't overvalued for what they are today. They're overvalued for what they'll be tomorrow.

The 2026 decline anticipates structural revenue compression, not just multiple compression.

What's Happening in Each Category

CRM — HubSpot (-48%), Salesforce (-35%): AI agents qualify leads, send personalized emails, follow up, and update the CRM without a human in the loop. Monday.com has already replaced 100 SDR roles with AI. That's 100 fewer seats. Multiplied by thousands of companies.

Project Management — Atlassian (-63%), Asana (-58%): If agents create tickets, assign them, follow up, and generate reports automatically, how many humans do you need to manage the backlog?

HR and Finance — Workday (-47%), Intuit (-45%): Workday has already cut 8.5% of its workforce. Jefferies downgraded them to Underperform citing "quantified AI impact on future revenues." This isn't speculation — it's analysis of how many seats will disappear over the next 3 years.

Data and Analytics — Snowflake (-40%), Datadog (-20%): LLMs can do analysis that previously required entire teams of analysts. The entry barrier to data analysis has collapsed.

Design Tools — Figma (-49%): When AI agents generate functional interfaces from text, how many designers do you need with access to Figma?

Those Defending Well — and Why

The two outliers in the index are Cloudflare (-3.64%) and Zoom (-2.67%). What do they have in common?

Cloudflare: Infrastructure. It's the layer through which the internet passes. AI agents need networks as much as humans do. If there's more AI traffic, Cloudflare wins — it doesn't lose.

Zoom: Human communication. For now, meetings are still people with people. And the business is reinventing itself with AI instead of competing against it.

The pattern is clear: those offering infrastructure or integrating AI instead of competing against it survive better.

Goldman Sachs identifies their "resilient buys" in this context: MongoDB (consumption vs. seat), Rubrik (data security), Procore (vertical construction with proprietary data), Nutanix (infrastructure). The common denominator: either infrastructure, or data that AI can't easily replicate, or a usage-based pricing model.

The New Model: Pay for Outcome

If the per-seat model is dying, what replaces it?

The "outcome-based" model: you pay for the delivered result, not for access to the tool.

Real examples:

  • If AI qualifies 1,000 leads for you this month, you pay for qualified leads — not for how many users have CRM access
  • If software automates 500 hours of analyst work, you pay a fraction of that value — not for the number of analysts using it
  • If a support agent resolves 10,000 tickets, you pay per ticket resolved — not for the human agents supervising

IDC projects that by 2028, 70% of SaaS vendors will have migrated from seats to consumption or outcome. What was once a marginal trend is becoming the new standard.

The Contrarian Argument — and Why It's Only Partially Correct

JP Morgan, Wedbush, and Morgan Stanley argue that the selloff is "overdone" and that SaaS has real moats: long-term contracts, high switching costs, proprietary data, compliance.

They're right that the speed of the adjustment may be exaggerated. Salesforce has $21B in contracts that won't be canceled tomorrow. Workday has CIOs who need 18 months to migrate to another system.

But the switching cost argument only delays the inevitable repricing — it doesn't avoid it. Contracts get renewed. And when they do, the negotiation is going to be different.

Andreessen Horowitz argues that AI increases software demand because more code will be written. They're also right — but that new software will be built and operated with fewer humans, which collapses the "seats" metric.

Software volume can explode. The number of paid seats can collapse simultaneously. They're not contradictory.

My Story With This Software

I used Asana since 2011. Over a decade.

When generative AI appeared, my first reaction wasn't "I'm going to hire more people on Upwork, Workana, or Fiverr to scale." It was the opposite: "I'm going to see what I can solve with this before hiring someone."

It wasn't a philosophical decision. It was pragmatism. If AI could do something that previously required hiring a freelancer, why not try that first?

Over time, that mindset spread across the entire stack. When autonomous agents arrived, I started actively migrating tools. n8n instead of Zapier. Listmonk instead of MailerLite. NocoDB instead of Airtable.

Asana stayed in my stack by inertia for a long time. I'd open it, create some tasks, and at some point stopped opening it. I didn't cancel it overnight — it simply became irrelevant. The tracking of my projects migrated to NocoDB + an agent that manages priorities automatically.

It wasn't a conscious decision to "leave Asana." It was that Asana stopped delivering value before I noticed.

That's exactly what the market is pricing into these stocks. Not that Asana is bad. It's that the work that justified paying for it is being done by something else today.

The Investor's Perspective

I'm an LP in 7+ VC funds and have made over 30 direct investments. My position on this meltdown:

What I'm avoiding:

  • Horizontal SaaS with a pure seat model without data moats
  • Generic productivity software without deep workflow integration
  • Analytics that don't have data that LLMs can't replicate

What I find interesting:

  • AI infrastructure (computing, networks, storage) — agents need it too
  • Vertical SaaS with proprietary data that's more valuable with AI than without it
  • Companies charging by outcome with clear, auditable metrics
  • Agent orchestration — the middleware of the new world

Questions I'd ask any SaaS founder today:

  1. How many of your clients will renew contracts the same way when the renewal cycle arrives in 2027?
  2. Do you have data that LLMs can't replicate?
  3. Can you charge by outcome instead of by access?

If the answers are "I don't know," "probably not," and "it's complicated," there's work to do.

And there's a pattern I'm seeing frequently in startups seeking investment today: founders who built a SaaS with AI help that solves a problem that... AI already solves on its own. Without even needing a wrapper.

They don't have a product. They have an interface over something that already exists for free.

The problem isn't just ChatGPT. In 2026, the native AI ecosystem competing directly with startups includes:

  • Anthropic: Claude Cowork (enterprise agents with integrations to Google Drive, Gmail, Excel, DocuSign), Claude Code, Claude.ai — automating everything from clinical documentation to complete development cycles
  • Google: Gemini Enterprise with 1,000+ pre-built agents, native integration with Workspace, and no-code agent creation
  • OpenAI: ChatGPT with custom GPTs, Operator (web agents), Codex for development

There are founders who give an impressive demo of their "HR AI solution" and don't know that Anthropic launched Cowork with an HR plugin specifically for that, that Google has a native onboarding agent, and that any company with $20/month of Gemini Enterprise has it included.

A few weeks ago, I had breakfast with a founder in Latin America who showed me something different. Her solution distributes AI through proprietary hardware in industrial contexts. What was once considered a scalability problem — having physical hardware — is now seen as a real competitive advantage. Hardware is a moat that language models can't copy in 90 days. The data that hardware captures is proprietary by nature. The entry barrier isn't the AI model — it's manufacturing, the supply chain, physical deployment.

That's a business. That's defensibility.

The difference between what generates enthusiasm in me as an investor and what generates concern isn't the technology — it's whether the business value exists independently of the underlying AI model.

And here's the paradox I find most interesting: the pay-for-outcome model emerging in SaaS is exactly the same logic I tell founders when they fall in love with the technology instead of the problem. The market doesn't care how you do it — it cares what problem you solve, how much value it delivers, and whether that value is measurable. The SaaS that survives will be the one that can demonstrate this with real metrics.

The question I ask any founder who asks me for investment today:

"Can I access what you do, for free, by directly using ChatGPT, Gemini Enterprise, Claude Cowork, or any other native AI available in the market?"

If the answer is "basically yes" — it's not a business, it's an experiment with a pricing model.

Conclusion: It's Not the End of Software, It's the End of a Model

Goldman Sachs estimated that the total software market could grow 20-45% by 2030. That's not a dying industry.

It's an industry being redistributed.

The Software Meltdown isn't because we're going to use less software. It's because the work that was paying for the seats will be done by agents, not humans. The value of software doesn't disappear — it changes hands.

The winners of the next cycle will be those who captured that value where it really is: in the data, in the infrastructure, in the results — not in per-user access.

Those who arrived late to the dance will look like newspapers in 2005: with good audiences, good content, and a revenue model being drained by something they still didn't quite understand.


Do you have a SaaS startup with a per-seat model? Are you investing in the sector? I'm interested in the debate. The speed of disruption matters as much as the direction. Share your perspective in the comments or in the Cágala, Aprende, Repite community, where we're building in real-time with AI tools.


📝 Originally published in Spanish at cristiantala.com. If you read Spanish, check the original for more context and community discussion.

Top comments (0)