Where is the Value of Software

Claude, Codex, and Gemini are writing most of the code now, and triaging incoming bugs. What used to be a high-skill craft is suddenly a commodity. People are vibe coding custom apps and prototypes over a weekend. In H2 of 2026: where is the value of a pre-packaged software anymore, and where is the value of a software engineer?

A quick anecdote from my personal life: Our building used to pay a vendor about $50 per unit for an HOA portal and services around it like dues, documents, reminders etc. We cancelled our contract and moved everything to Claude with some light data architecture around it. It seemed to work great! Well… right up until we realized Claude had quietly missed some compliance reminders.

It does look like a chunk of software is getting replaced, mostly a thin layer on top of a database. Not Google Docs, Visual Studio Code, Chase or Waymo apps. I’m not vibe coding my own word processor, and neither is anyone I know. For that software it was never about the code. Google Docs wins because everyone I know is already there, because sharing and collaboration just works. There are companies out there with more elegant code and newer frameworks than half the apps I use daily, and I don’t use their software. Internal code structure was never the differentiator, for all we know a bunch of code out there is still running on PHP. The durable moats seem to be distribution and data and that’s why we still have most of the businesses use MS Office.

And notice who did the replacing in the HOA story: an engineer, with Claude as the tool. Not a non-technical person vibe coding over a weekend. This is the pattern that I see as of today. Engineers aren’t being bypassed, they’re moving from the vendor to the buyer and they are expanding their scope. There is a less of a pushback of saying “I don’t have time to talk to the users, I got migration code to write”, the engineers are moving to wanting to connect the dots and understand the bigger picture. The value didn’t evaporate, it relocated.

The “chopping wood and carrying water” work was always boilerplate and a stepping stone to becoming more senior. CRUD web services, the tenth admin dashboard are all basically automated now. So what does the engineer bring? Going back to the HOA story again: the value wasn’t in building the portal, it was in knowing which reminders can’t be missed like ownership of the unglamorous correctness. I am learning more about when the inspections are due and spending less time creating a task tracking interface.

This is also what separates a prototype from a production system: a prototype answers “show me what it would look like if I started using it.” It’s a decision tool. A production system takes over safekeeping a part of your business processes and data. That requires trust.

Historically trust in software was pretty thin: EULA told you basically that nothing was guaranteed to work. In the world where untrustworthy software is approaching zero trust is the new moat. It’s the vendor’s brand and balance sheet standing behind the product. A company shipping 100% AI-generated code still has a bank account you can sue and a reputation it can lose. The AI agent’s loss on a bad decision is zero, but the vendor’s isn’t. A vibe coded app that drains your checking account is scary not because of the tooling, but because there’s no one to call.

That’s also why nobody I know with a real user base pushes code straight to prod without review. The gate persists even though a lot of engineers say that Opus and Argon write fantastic Go and Python. Is the review going to be AI aided.. actually it already is. But the review itself, the act of someone competent signing off, isn’t going anywhere. It’s the signature on the agreement.

Software attached to dollars, to cars, planes, trucks, national defense, to legal compliance that’s where trust matters most. The HOA portal wasn’t just software; it was attached to compliance obligations with real consequences. Claude handled the software part fine and missed the real-world part and that is the entire ballgame. The compliance reminders Claude missed are a market opportunity for whoever sells “we won’t miss the reminders.”

What about the engineers? The differentiators of code volume and framework knowledge dropped. What remains, what was always the actual job and is just visible now, is knowing what matters, owning the correctness nobody applauds, and signing your name to it. Skipping that works for a prototype. It fails the moment you ask anyone to trust you with their time and money.

What happens in the future is anyone’s guess, but until AI gets skin in the game the actual skin is required.