Is your demo killing deals? Get AI-powered gap analysis in 5 mins.

Analyze my demo

What Happens to Software in 5-10 years

March 13, 2026
During our Board meeting last week we were mapping out the next 5 - 10 years of Enterprise Software. I then went and talked with founders, VCs, Partners from Bain and listened to other tech leaders. It all ultimately led me to staring at a wall for a while before I wrote this.
Sam Senior
Table of Contents

During our Board meeting last week we were mapping out the next 5 - 10 years of Enterprise Software. I then went and talked with founders, VCs, Partners from Bain and listened to other tech leaders. It all ultimately led me to staring at a wall for a while before I wrote this.

This is my emerging thesis on where enterprise software will be in 5 - 10 years:

  1. Building software will be basically free and instant
  2. It will feel like the 90s again with custom, in-house software; the tech debt will be staggering
  3. Most software companies in the middle will die and be replaced by millions of hyper-specific tools akin to the mobile App Store today
  4. Only vendors with deep business logic and trust will be thriving
  5. The human browser will be a relic and (almost) everything will be built for agents
  6. We’ll be furiously working to safely integrate thousands of apps that nobody maintains

The cloud transition took 15-20 years and many enterprises still aren't done.

Might we be overblowing the SaaSpocalypse right now? Short-term yes, long-term no.

We're experiencing AI in our personal lives first, which will severely compress the enterprise adoption curve compared to the cloud.

Note: I'm writing this from San Francisco, as a founder that’s living in the heart of it all. My daily reality is not most people's daily reality… yet.

Hypothesis#1: Building Software is Entering a New Era

Building software will be basically free and instant

The AI Labs are claiming that nearly 100% of their code is AI generated while industry-wide, reports indicate roughly 41% of all code is now AI-generated or AI-assisted. Lovable alone is generating over 100,000 projects per day.

If you've used Cursor, Claude Code, Lovable or Replit, you’ll know how it feels to turn your imagination into something real. The $49-500/mo SaaS tools that were "good enough" are getting replaced by things people build themselves in hours. Personally, I’ve replaced multiple 5-figure software contracts in just 20 minutes of using Claude code and prototyped multiple of our upcoming product capabilities in a few hours. These are the tools I turn to first to get things done.

And I’m not alone - if you’ve caught the bug, you know all about it.

But what does it mean for the industry when this keeps accelerating?

Hypothesis #2: The Tech Debt

It will feel like the 80s and 90s again with custom in-house software; the tech debt will be staggering

In the late 80s and 90s, people built incredibly custom software for themselves. Basic by today's standards but people were duct taping together MS Access databases, Excel spreadsheets with macros and Lotus Notes apps to run the entire systems in their business.

Then we professionalized it. In the SaaS era, we moved to one-size-fits-all software that was polished, scalable, and had pre-defined workflows. Instead of building your own thing, you bought Salesforce, HubSpot, Asana, Monday.com, or one of ten thousand other tools. These tools were more reliable, secure, cheaper, and much more. But we traded customization for convenience (it's partly why so many implementations fail).

Now the pendulum is swinging all the way back and we're just at the start.

  • The ops team is now vibe-coding a custom workflow app in an afternoon
  • The finance team that would have built a monster spreadsheet is now building a purpose-built forecasting tool.
  • We’ve totally replaced a revenue reporting dashboard software tool.

But… nobody's thinking about what happens next.

What happens when the finance person leaves the company? That app becomes an unreadable black box to their successor and people will be left asking:

  • Does this need maintaining?
  • Is it connected to something else that'll break if we touch it?
  • Should we just build something new?

Hypothesis #3: Software Companies in the Middle will Die

Most software companies in the middle will have died and be replacedby millions of hyper-specific toolsakin to the mobile App Store today.

The “SaaSpocalypse” started as $285B evaporated from software stocks in 48 hours just a few weeks ago. The bet is that AI is going to eat these products.

What will emerge on the other side of this transition are three tiers:

  • A handful of massive platforms at the top
  • A huge explosion of micro-apps at the bottom
  • And a graveyard in the middle.

The people building these micro-apps won't be traditional developers shipping products. They'll be people building into their specific hyper niche area:

  • Not just a project management replacement, but a project management replacement for construction companies in Colorado that specifically do decking for single family homes with a team of five contractors.
  • It'll be the pastor building a church management app specific to their religion, part of the country, and weekly schedule for Sunday school.
  • The recruiter who built a candidate scoring tool just for hiring contractors into service roles that's now used by thirty other recruiting firms.
  • And so many more…

There will be small businesses and entrepreneurs publishing their AI generated software products. It will be the App Store economy coming to enterprise software. Millions of hyper-specific tools built by people who understand the problem better than any vendor ever could.

Hypothesis #4: Deep Business Logic and Trust will be Critical

Only vendors with deep business logic and trust will be thriving.

Despite this, some vendors will survive, and many will thrive.

Think about what actually makes Salesforce or SAP or Workday valuable. It's not the interface, it's everything underneath. Approval workflows, territory rules, pricing logic, payroll across jurisdictions, revenue recognition, years of custom automation that encodes how your company actually operates.

I’m not saying it’s these specific tools that will survive but they have a better chance than many.

Anyone can vibe-code a brilliant workflow tool in an afternoon, but AI cannot sign a BAA for HIPAA compliance, pass a SOC2 Type II audit, or be the contact point when PII is accidentally leaked.

Enterprises don't just buy software to solve a problem, they’re also buying trust and security.

So what actually survives the SaaSpocalypse? Two things.

  • Products you can't rebuild in an afternoon because it carries the support, security, compliance, and governance that enterprises actually require.
  • Products built with the kind of taste that only comes from product managers staring at a wall for weeks, figuring out an experience you didn't know you wanted yet.

Neither of those go away because of AI generated code.

Hypothesis #5: Say Goodbye to the Human Browser

The human browser will be a relic and (almost) everything will be built for agents

Today when using any software we’re clicking through dashboards, filling out forms, navigating menus, and toggling between tabs. The entire experience is designed around a human sitting at a screen.

That's already starting to feel incredibly dated.

For example, I’m no longer accessing HubSpot directly, I just ask Claude Code to give me updates and move things along for me. I can speak to Claude and just make things happen. No need to navigate the software and all its intricate buttons.

This change is going to happen pretty fast, most software interactions will happen between agents. And eventually will be entirely invisible to us. Your CRM will talk to your marketing tools while talking to your ERP. All just doing work without a human in the loop.

If we do have visual, browser based experiences they'll be throwaway screens generated for that exact moment, not something a front-end team designed and shipped as a fixed product.

And that's what makes the next problem so much harder. When a human uses software, there's a person noticing when something looks wrong that many agents won’t pick up (at least not yet anyway). When agents are talking to agents, nobody's watching. The speed and scale of what can go wrong is completely different.

6. We’ll be furiously working to safely integrate thousands of apps that nobody maintains.

Software was always built to last. What's coming is the opposite. We’re going to have throwaway tools built for a single purpose that get abandoned weeks later.

Over time the best ones will become valuable enough that other teams want access.

Someone will realize the lead scoring app would be way more useful if it could pull data from the campaign tracker, scheduling tool, and revenue forecasting model.

So now your vibe-coded tools need to talk to each other too, not just to the core systems.

Suddenly you have dozens of apps that all need to connect and nobody designed any of them to do that.

We'll need something that can look at all of these apps and understand what each one actually does, how it behaves, what data it reads and writes, and figure out how to safely connect them.

Not just plug A into B, but actually understand the logic of both sides well enough to know what will break downstream and what shouldn't be connected at all.

Without it, things will go wrong fast.

An AI agent might reclassify expenses in your ERP based on vendor descriptions. That might change how costs roll up into quarterly reporting. A quarter later, the CFO asks "why are our margins off by 3%?".

The answer will involve figuring out which tool made the change, what instructions it was following, what data it was looking at, and whether it should have been approved.

Companies will need to treat this the same way they treated shadow IT in the mid 2010s. There will be teams that sit between AI prototypes and production who decide what's safe to keep, what needs rebuilding, and what needs to be killed.

The knowledge of how all this software actually behaves, not what the docs say, but what really happens when you connect them, will be the foundation everything else depends on.

So, where does this all go?

I keep coming back to an exercise I do: what am I very confident about 20 years from now?

20 years out: nobody is clicking through a static screen and all software is ambient. Most interactions happen between agents. When you need a visual interface, it's generated on the fly.

10 years out: most software is throwaway. Built for a purpose, used and discarded.

5 years out: we're deep in the messy transition. Cutting-edge and legacy coexist in the same team. Software is everywhere but the systems to make it all work together is the bottleneck.

Right now: we're at the very start of the mess.

‍

In the 90s, someone's dad built an inventory system in Access that ran an auto parts store for a decade. It wasn't pretty. It wasn't scalable. But it worked, because he understood his business better than any software vendor ever could.

We're about to learn that lesson again, just at a completely different scale. The tools are powerful enough now that everyone builds. But the mess it’s going to create will need to be solved.

What do you agree with? What am I entirely wrong on? Does anything surprise you?

Sources & References

AI-generated code statistics:

Software market impact:

Integration & API data:

  • 28% of enterprise apps connected: MuleSoft Connectivity Benchmark Report
  • ~15% of APIs publicly available: Postman State of the API Report

Commentary & analysis:

  • Benedict Evans on taste/product thinking: Another Podcast, "AI and SaaS" episode
  • Ben Thompson on software ≠ code: Stratechery, "Microsoft and Software Survival"

‍

‍

‍

‍

‍

‍

‍

‍

‍

More revenue, less effort.

Spin up live demo environments, trials, and POCs in seconds with TestBox. Close more deals and speed up your sales cycle — all with fewer resources than ever before.

Get a demo
© 2026 TestBox. All rights reserved.