Software is not dead. But your Software Company might be.
Software isn't dead — but the companies that can't adapt to AI are. Mike Cannon-Brooks explains why operators need to think about input vs output constraints.
The idea that software as a category is dead? Ludicrous.
That’s not me being optimistic, I run a software company too. That’s Mike Cannon-Brooks — co-founder of Atlassian, a company doing north of $5 billion in cloud revenue and accelerating responding to the loudest narrative in tech right now. And honestly, he’s right. Software isn’t going anywhere, well I hope so. But a lot of software companies are about to have a very rough few years.
This is what is actually happening, and what operators need to understand about which side of the line they’re on.
The “Death of SaaS” panic is a Category Error
Every few weeks, someone on X declares that SaaS is dead, AI ate it, and we should all pivot to agent wrappers. Then the same week, a company like Harvey raises $200 million at an $11 billion valuation. Harvey is… software. Growing 300% year-over-year.
So which is it?
On a recent episode of 20VC, Cannon-Brooks broke it down simply: will every SaaS company survive the next five to ten years? Absolutely not. Will a lot of them continue to grow and prosper? Absolutely. Is that any different from the last decade? No.
He pulled up Atlassian’s old competitor lists from 2005, 2010, 2015. Most of those companies don’t exist anymore — merged, acquired, or gone. And Atlassian has a completely new set of competitors today. That’s just how the technology industry works.
The difference now is the rate of change. In a normal cycle, maybe 5-10% of companies fail each year. During an architectural shift — cloud, mobile, and now AI — that death rate compresses into a two-year window where it could hit 50%. Not every company dies, but the ones that aren’t rebuilding fast enough absolutely will.
Software customers are going to buy more in the future. The question is whether they’ll buy it from you.
The input-output framework that changes everything
Here’s the most useful mental model from the conversation, and it comes directly from how Cannon-Brooks thinks about AI’s impact across different business functions.
Some domains are input-constrained. Legal, customer support, HR — there’s a fixed number of problems coming in. Your lawyers can’t create more legal problems just because they got faster. Your customers ask 100 questions a day, and if you have twice as many customers, that becomes 200. The input is based on some external ratio.
Other domains are output-constrained. Engineering, product development, creative work — the roadmap is never finished. You can always build more. The constraint is how much you can produce, not how much demand exists.
Here’s why this matters for operators:
In input-constrained domains, AI mostly cuts costs. You answer the same questions with fewer people. That’s a real, measurable ROI — easy to sell, easy to prove. But in year two and three, those savings are baked in. You need to keep delivering new value or you’re just a line item that shrinks.
In output-constrained domains, AI expands what’s possible. Engineering teams aren’t getting smaller — they’re shipping more. According to Cannon-Brooks, Atlassian’s 10,000 R&D people are building things faster and better than ever. Some of their AI features now run at a thousandth of the cost compared to when they launched. The features still work. The margin just went up.
If you’re building a company, ask yourself: am I selling into an input-constrained or output-constrained domain? The answer changes your entire strategy.
The Danger: Being third in a subsegment
The customer support space tells this story perfectly. Sierra just crossed $150 million in ARR. Fourteen new companies in the space have raised over $100 million in the last two years. And they’re competing against ServiceNow, Salesforce, Atlassian, Zendesk, and Intercom.
The market is real. The money is flowing. But as the 20VC crew pointed out, the worst position to be in is third place in a subsegment. You’re too small to dominate, too big to pivot, and the market leader can decide to annex your space anytime they want.
This is happening everywhere. VCs are piling into consensus winners because it’s the venture capital equivalent of “you can’t get fired for buying IBM.” Of newly minted unicorns in Q1 of last year, 40% already had one or more up rounds by Q4. The money says: once you’re winning, we’ll double and triple down.
For operators, the lesson is uncomfortable but clear. If you’re not the obvious leader in your space, you need a different playbook. Adjacent territory. A wedge nobody else owns. Or the honest conversation about whether you’re building something defensible at all.
🚀 Tools of the week
🚀 Cloudera: Your weekly playbook for navigating agent chaos and turning AI disruption into a scalable advantage. Listen to The AI Forecast podcast.
✂️ EZTrimmer: Trim, split, and export videos in seconds with AI.
✅ Notion AI: Use AI directly inside Notion to boost your productivity.
📽️ Kubrix: Create cinema-quality stunning AI-generated videos in seconds.
🤖 Lorka: Chat with the top AI models in one place.
🧪 YC Startup Spotlight
Abundant (YC W25)
• Provides on-demand human operators who step in when AI agents fail in production, ensuring 100% reliability while generating training data for improvement.
• Use when: You’re deploying AI agents in production and need a safety net for edge cases, especially in regulated industries like healthcare, legal, or finance. abundant.ai
Until next time, keep building, keep questioning, and don’t let the noise drown out what actually matters.
- Rayn