Software is dead? What the AI rebuild wave means for your company
A few years ago, reverse engineering a program from its compiled binary was a job for specialists with a lot of patience. Today, AI agents can drive the same tools those specialists use, like Ghidra and IDA Pro, trace how a program works and, in some cases, rebuild substantial programs without the original source code. Not perfectly, and not for every program, but the cost of understanding someone else’s software has dropped sharply.
Games are where this became visible to everyone. In the last few weeks, people have shipped browser versions of Halo: Combat Evolved, GTA: Vice City and Call of Duty: Black Ops reconstructed with AI. Others are merging entire games: Minecraft inside Elden Ring and Skyrim, Skate 3’s skateboarding inside Modern Warfare 2. Work that once took a team months, done by one person in days.
Most of these projects use code and assets the creators don’t own, and publishers’ lawyers are already paying attention. That’s not the point here. The point is how little it now takes to rebuild software.
And it’s not only games. This month, a developer released ArtCraft, free, open-source rebuilds of seven Adobe apps, from Photoshop and Illustrator to Premiere and Acrobat, written with Claude. Another developer published OpenTerminal, a self-hosted, Bloomberg-style market terminal with charts, a screener, a sector heatmap and an AI assistant, for people who will never pay $2,000 a month for the real thing. Neither is a full replacement yet: ArtCraft is an early alpha, and OpenTerminal runs on free public data that can be delayed or wrong. But a year ago, hardly anyone would have tried either on their own.
So the headline writes itself: software is dead.
It isn’t. But the price of software is collapsing, and that changes the deal you have with every tool your company pays for.
That doesn’t mean anyone can rebuild your software, though. Stories like these make it sound as if all it takes is a chatbot: a vibe coder, your nephew, or the man who used ChatGPT to help design a cancer vaccine for his dog. Look closer and the picture changes. ArtCraft’s developer has 15 years of programming behind him, a decade of it in Rust, the language he built it in. The dog owner is a data analyst with machine learning experience, and the vaccine was sequenced, made and given by university scientists and vets. AI multiplies what skilled people can do, especially people who have been working with it for a while. Someone just starting out would take far longer, and burn through a lot of tokens, to get anywhere close.
But what does this mean for you?
If building software is getting cheap, the subscription you renew every year starts to look different. You’re no longer stuck with “pay or go without”. You now have six realistic options:
- Keep paying for what you use
- Build a replacement yourself
- Build it yourself, with an expert guiding you
- Hire a vibe coder to build and maintain your internal stack
- Switch to a newer, cheaper competitor
- Hire a boutique agency to build a replacement for you
To make this concrete, let me follow one company through each of them.
Full disclosure: I offer two of these myself, coaching people who build their own tools (Option 3) and building tailor-made software (Option 6). That’s why I’ve tried to be honest about when each option is and isn’t the right call.
Meet Sofia
Sofia is the operations director of a 35-person furniture distributor. Her sales team lives in a quoting and CRM platform that costs the company €2,400 a month. It works. Nobody loves it. Half the team still exports to Excel to build proposals.
In September the renewal email arrives: a 25% price increase, plus an “AI add-on” sold separately.
Sofia has heard that software is cheap to build now. She has six ways forward.
Option 1: Sofia keeps paying
Sofia signs the renewal. This is the baseline, and it’s not a bad choice by default.
Before she signs, though, she does the homework most companies skip:
- How much time does the tool actually save? Not in the vendor’s case study. In her team’s week. Which processes run through it, and which ones leak out to Excel and email?
- Is it still improving? A product that shipped nothing meaningful in the last year is coasting on your renewal.
- Is it ready for agents? Does it have a proper API? A CLI? An MCP server? This is the question that matters most for the next few years. Soon a good part of the work in tools like this will be done by AI agents acting on your behalf. A tool that only works through its own screens will become a wall your agents can’t get through.
Good when: the tool saves real time, keeps improving, and opens itself to automation. Bad when: you’re paying a premium for inertia, and the vendor’s answer to AI is a paid add-on bolted to the side.
Option 2: Sofia builds it herself
Sofia is curious and technical enough. She spends a few evenings with an AI coding tool and has a working quoting app in two weeks. The demo to the sales team goes great.
Three months later, the app needs a new discount rule, a fix for a PDF that breaks on long product names, and a login for the new hire. The AI tools she used have changed twice since she started. Every fix takes her a Saturday.
Meanwhile, two supplier negotiations slipped because she was debugging.
Then there’s the bill. She started on a $20 AI plan, hit its limits in the first week, and moved to a top-tier plan at around $200 a month. Much of that usage goes to the AI going in circles on problems an experienced developer would spot in minutes.
Good when: you have the time, real interest, and the discipline to keep up with how fast AI tools are moving. Bad when: it pulls you away from running the business. The build is the easy part. Maintenance is the job that never ends, and you just hired yourself for it. And without experience, you pay for it in tokens: long sessions, failed attempts and expensive subscriptions.
Option 3: Sofia builds it herself, with an expert beside her
Same Sofia, same idea, but this time she doesn’t go it alone. She books a coaching session every couple of weeks with someone who has built this kind of thing before.
She asks better questions, the AI wastes fewer attempts, and she uses far fewer tokens. That’s cheaper, and better for the environment too.
It’s also faster. Sofia starts with a smarter setup, the right tools and a project structure the AI can work with, and she’s warned before making the kind of mistake that costs weeks of development and a pile of tokens to undo, like a data model that doesn’t match how quotes really move through the company.
She doesn’t need the most expensive plan either. I’m a heavy user myself, and I build several projects at once on two or more $20 plans from different AI providers instead of one top-tier subscription. That has three advantages:
- More options: different models are good at different things, so you can pick the right one for each task.
- A second opinion: you can have one AI agent check the work of another, which catches mistakes either would miss alone.
- Resilience: when one provider is down or you hit a limit, you switch to the other and keep working.
Good when: you want to build and own your tools, can set aside a few hours a week, and want to get there faster and cheaper than going alone. Bad when: you can’t spare those hours at all. An expert speeds you up, but you’re still the one building. If your calendar is full, look at Options 4 and 6.
Think of it as the gym with a personal trainer. You still lift the weights, at your own pace, but someone checks your form and makes sure you’re doing the right exercises. And because the next session is already on the calendar, you keep showing up. A side project with no deadline is the first thing to slip when sales gets busy, and that bit of accountability is often what turns Sofia’s promising demo into a tool her team actually uses.
This is what my Starter plan is for: regular coaching sessions where we set up your tools, plan what to build, and keep you moving without burning money on tokens.
Option 4: Sofia hires a vibe coder
Sofia doesn’t have the time, so she finds a freelancer who builds with AI tools. Fast, cheap, enthusiastic. Within a month the new quoting app is live, connected to their ERP and the email server.
A year later, the freelancer has built five internal tools. They’re the only person who knows how any of them work. The code lives in their personal accounts. The server is on their cloud billing. When Sofia asks for documentation, she gets a voice note. And the monthly invoice keeps growing, because the freelancer’s AI subscriptions and token usage are passed on to her.
Then the freelancer takes a full-time job.
Good when: you want owned, internal tools and don’t have the time to build them, and you pick someone trustworthy who works in the open. Bad when: you end up vendor-locked to one person who holds the keys to your company’s infrastructure. Be very careful who you delegate this to. At a minimum, insist that code, accounts and hosting are in the company’s name from day one, and that someone else could pick the work up.
Option 5: Sofia switches to a cheaper newcomer
Because software is cheap to build, a lot of new players are entering every market. Sofia finds a two-year-old startup with a slicker quoting tool at half the price. The founders answer her emails personally.
The migration takes a month. Some features are missing, a few bugs show up in the first weeks, but the team ships fixes fast and the sales team actually likes the product.
Good when: the product is good, the team is motivated, and the same agent-readiness questions from Option 1 have good answers: API, CLI or MCP server, data export, decent support. Bad when: you pick on price alone. It’s a new product, so expect bugs. And ask yourself what happens if they run out of money, or get acquired by the vendor you just left.
Option 6: Sofia hires a boutique agency
Sofia brings in a small team of senior developers. The first two weeks aren’t about code. They map how quotes actually move through the company, and find that the real bottleneck isn’t the CRM at all. It’s the back-and-forth with the warehouse on stock availability, which the old tool never handled.
The replacement they build fits how her company works: quoting, stock checks and approvals in one flow, with an API so their AI agents can draft quotes from customer emails. It’s live in two months. The code, infrastructure and documentation belong to the company.
Good when: you want something better than a copy of the old tool. An experienced external team will see things you haven’t thought of, and will get you there faster, with higher quality and fewer tokens than going alone or with a single vibe coder. Bad when: budget is tight. Up front, this costs more than doing it yourself. But consider what you get: software you own is intellectual property your company can leverage in a valuation, it stops the yearly price increases, and it fits your processes far better than someone else’s product ever will.
It’s a tailor-made suit, in software terms. This is the work behind my Scale and Partner plans, or a custom quote for larger projects.
So which option is yours?
There’s no single right answer, and two things shift it more than anything else: who will use the software, and how complex it is. A quick way to decide:
| If you… | Team size and complexity | Consider |
|---|---|---|
| Have a tool that saves real time and is open to automation | Any size, standard needs | Keep paying |
| Have time, curiosity and a small, contained problem | Solo or a very small team, simple tool | Build it yourself |
| Want to build it yourself without wasting time and tokens | Small team, simple to moderate complexity | Build it yourself, with an expert guiding you |
| Want owned tools fast, and can keep control of code and accounts | Small to mid-sized team, moderate complexity | A vibe coder, carefully chosen |
| Need a standard tool, cheaper, and can live with early bugs | Any size, standard needs | A newer competitor |
| Want software shaped around how your business really works | Larger team, complex or core to the business | A boutique team |
Most companies will end up with a mix: keep the tools that earn their price, replace the ones that don’t, and own the few that are core to how you compete.
Whatever you choose, apply the same filter everywhere: can AI agents work with it? That’s where we’re heading, and every software decision you make this year should keep that door open.
If you’re staring at a renewal and not sure which option fits, let’s look at it together. In one working session we map the tool against your processes and decide whether to keep, switch, or replace it.