In a year when everyone is focused on AI, the bigger story may be what AI enables: a massive explosion of software creation—and software failures. AI collapses build costs and timelines, which means more software ships, which means fiercer competition and faster commoditization, which means more failures.
If you build or buy B2B software, here’s what 2026 looks like.
A Message From the Field
Here’s what that shift looks like inside a real product org.
Last week, Erik Ornitz sent me a message that put words to something I’d been sensing for months. Erik was my product partner when I ran the Innovations team at TripAdvisor. Now he heads product at Topline Pro:
“With Claude Code we’re seeing 2-3x the output we’ve ever seen before by our top engineers. Literally skyrocketing the past several weeks since Opus 4.5 came out. Just feels like a fundamentally different world all the sudden.
My PMs/Designers just cannot keep up.
Does this mean a radically different shape of technology organizations? Old ratios of 1 PM / Designer to 5-8 Engineers feel out the window.
Are you experiencing the same? Does that mean a major change in the % of budgets towards defining what to build vs. actually building it?”
— Erik Ornitz, Head of Product, Topline Pro
The answer to Erik’s question is yes. To all of it.
The Data
GitHub’s 2024 Octoverse report tells the story:
100+ million new repositories created in 2024
~25% year-over-year growth in total repos (now over 500 million)
~98% growth in generative AI projects alone
1.4 million new open source contributors
These represent software being created and iterated on. Many repos are experiments, forks, or prototypes—but the volume signals a fundamental shift in creation velocity.
The productivity studies back up what Erik is seeing—at least directionally. GitHub’s controlled study showed 56% faster task completion on specific coding tasks. Stack Overflow’s survey of 90,000 developers found 70% are using or planning to use AI tools.
Erik’s 2-3x number sounds aggressive against these studies. The gap comes from what you measure. Studies measure isolated tasks with junior-to-mid engineers. Erik is measuring total output from senior engineers who’ve built multi-agent workflows—Claude Code orchestrating research, implementation, testing, and documentation in parallel. Different baselines, different measurements, different results.
I’m Living Proof
Before 2025, I had never published a single open source project. Not one. In twenty-five years of building software professionally, I never had the bandwidth to maintain a side project while doing my actual job.
In the past twelve months, I’ve published seventeen:
claude-mpm — Multi-agent orchestration for Claude Code
mcp-vector-search — Semantic code search via Model Context Protocol
kuzu-memory — Graph-based memory system for AI agents
These aren’t toys. They’re production tools I use daily. What changed: I went from spending 80% of my time on scaffolding and boilerplate to spending 80% of my time on the interesting problems. The grunt work—test generation, documentation, refactoring—happens in minutes instead of hours.
The same engineer. The same available hours. Radically different output. This would have been impossible before Claude Code and Opus 4.5 shipped in late 2025.
The Economics Have Flipped
Erik’s real question: if engineers can produce 2-3x the output, what happens to the rest of the organization?
The old model assumed building software was expensive and slow. The entire SaaS industry is built on this assumption. Why build a CRM when Salesforce exists? Why build analytics when Amplitude exists? Why build anything when you can pay per seat per month for someone else’s solution?
The math made sense when custom development meant six-figure budgets and twelve-month timelines.
The math is changing.
Based on my own projects and conversations with engineering leaders, I estimate AI tools have reduced the cost of building new greenfield software by roughly an order of magnitude for certain categories of work—internal tools, CRUD applications, API integrations, developer utilities. Not every category. Not enterprise systems with complex compliance requirements. But for the kinds of software that used to be “not worth building,” the math has changed.
A feature that would have taken a team of three engineers two months can now be built by one engineer in two weeks. That’s not a study—that’s what I’m seeing in practice.
When building gets that cheap, the calculus of build versus buy changes completely.
SaaS Vendors Should Be Worried
If I were playing the stock market right now, I would pay very close attention to SaaS renewal rates.
Think about what happens when thousands of companies simultaneously realize they can build what they need for less than their annual software licenses cost. The $50K/year internal analytics dashboard? Illustratively: one engineer, one month. The $200K/year customer data integration? Two engineers, one quarter—if it’s a narrow, well-defined workflow.
Not every category is equally vulnerable. Most at risk: horizontal admin tools, internal workflow automation, simple analytics, and single-purpose integrations. More defensible: compliance-heavy systems of record, platforms with strong network effects, multi-tenant marketplaces, and products built on proprietary data moats.
Salesforce isn’t going anywhere in the near term—their moat is complexity, switching costs, and ecosystem lock-in. That moat is real.
But the bar is rising. The threshold where it makes sense to buy instead of build is moving up dramatically.
I expect 2026 will see an unusually high number of SaaS vendors fail—particularly in the crowded mid-market where differentiation was always thin.
The ones that survive will need to improve at a pace they’ve never attempted before. Customer expectations are rising in lockstep with capabilities. B2B software will need to approach B2C quality. Clunky enterprise UIs that customers tolerated because they had no alternative? Those alternatives now exist.
The Managed-Service Stack as Force Multiplier
None of this would be happening without the parallel explosion in managed platforms and developer services.
Ten years ago, building a new software product meant provisioning servers, managing databases, handling authentication, building deployment pipelines. The operational overhead often exceeded the development effort.
Now: Vercel deploys your frontend. Supabase handles your database and auth. Stripe processes payments. Resend sends emails. Everything connects via APIs.
The composable stack means engineers can focus on the software that differentiates their product. The undifferentiated infrastructure is someone else’s problem.
AI coding tools plus composable infrastructure equals massive leverage. One engineer can build and ship what used to require a team.
A Warning for the Ambitious
Many engineers will be tempted to take their great idea and build a business around it. The tools make it easy. The startup costs are minimal. Why not?
Because if your only defensibility is the idea and the code, you have no moat.
AI generates code nearly as easily as English—for standard patterns and well-documented APIs, anyway. Any idea you can implement, someone else can implement too—probably faster, probably with more resources, probably with better distribution.
The SaaS vendors going out of business will be replaced by a flood of new entrants. Most of those new entrants will also fail. The barrier to entry has dropped, but the barriers to sustainable success haven’t. This is why virtually everything I build is open source. It’s valuable enough for me that I’m willing to spend the time to build it. []Charging customers for it? A completely different equation.
You need something beyond code:
Distribution: An audience, channel partnerships, or existing customer relationships
Community: A user base that contributes content, plugins, or network value
Domain expertise: Deep knowledge of a niche workflow that takes years to acquire
Data advantages: Proprietary datasets that improve your product over time
Integration complexity: Deep hooks into systems-of-record that make switching painful
Something that can’t be replicated in a weekend by another engineer with Claude Code.
The golden age of software creation is also the golden age of software commoditization. Don’t confuse the ability to build with the ability to win.
The Disruption Is Already Here
Tech layoffs doubled in 2025—264,000 jobs eliminated according to Layoffs.fyi, compared to roughly 130,000 in 2024. Some of this is cyclical. Some of it is AI.
Anecdotally, I’m hearing from engineering managers that their top developers are spending dramatically less time writing code directly—they’re orchestrating AI tools instead. The code is still being written, just not by humans, or not by as many humans.
The disruption won’t be distributed evenly. Senior engineers who can orchestrate AI tools effectively will become more valuable. Junior engineers who were being paid to write boilerplate will find that work automated.
Product managers and designers face a different problem: they used to be the bottleneck’s counterweight. Now they’re the bottleneck. Erik’s question about organizational ratios is urgent because the answer affects hiring, budgets, and team structures across the industry.
The Bright Spot
This is going to be painful for many people. Layoffs are painful. Business failures are painful. Career disruption is painful.
But we’re entering a golden age of software.
More software will be created in 2026 than in any year in history. More problems will be solved by code. More ideas will ship. More experiments will run. More entrepreneurs will try.
Most of it will be crap, as I said at the start. Sturgeon’s Law—90% of everything is crap—doesn’t suspend for technological revolutions. But the 10% that isn’t crap will be extraordinary.
The tools now exist to build durable, production-quality software at a scale and speed that wasn’t possible two years ago. Those who learn to use them—really use them, not just dabble—will build things that matter.
I’ve never been more excited to be an engineer.
I’ve also never been more aware of how brutal the transition will be for those caught on the wrong side.
2026 will be the year of software. Here’s how to prepare:
Learn agent-generated coding now. Not AI-assisted (autocomplete, suggestions)—AI-generated: you describe intent, agents produce complete implementations. This means multi-agent orchestration, prompt engineering for code, and reviewing AI output instead of writing it. The paradigm shift is steep and early movers will have 6-12 months of advantage.
Tighten your product discovery loops. When engineering throughput triples, PM and design become the constraint. The organizations that figure out faster iteration on what to build will outpace those focused only on building faster.
Invest in distribution before you need it. Building is no longer the hard part. Finding users, building brand, creating switching costs—those are the new differentiators.
The explosion is coming. Make sure you’re positioned on the right side of it.
I’m writing about agentic coding workflows at hyperdev.matsuoka.com. My open source tools are at github.com/bobmatnyc.