Is Software Worthless Now? Even Custom Work Commoditizes
Source: In the AI era, software is already worthless (V2EX)
Core takeaway
The OP says pure software can barely build an advantage anymore. The community usually adds a sharper cut: what devalues is the layer below the model kill-line—tool shells anyone can vibe-code, settings junk, thin content sites.
We push the endgame further:
When agents are strong enough, software really is cheap—and so is most customization.
Default model capabilities swallow features. Project-priced “change the UI / change the workflow” work collapses into conversational edits. What can still hold value: outcomes, liability, distribution trust, proprietary data and workflow memory, plus hardware, licenses, and supply-chain friction software alone cannot touch.
What the OP claimed
One line: pure software—for individuals and companies—struggles to build moats.
The mood underneath is clear: when shipping gets easy, willingness to pay drops. Some replies agree: if everyone can build it, people subconsciously refuse to pay.
Community consensus: stratify first
Few people buy “all software is worthless,” but the high-signal split is consistent:
| More likely to cheapen | Harder to cheapen soon |
|---|---|
| Tiny utilities, cleaner/settings shells, niche info sites | Vertical industry software, simulation stacks, long-lived pro tools |
| Local, low-dependency personal utilities | Large cloud services + regulatory barriers |
| ”I can ship a hundred of these in a day” shells | Experience and edge cases built over years |
Portable line: model capability is the kill-line—software everyone can write loses price power; complex, reliable, situated software still sells.
Others remind: value always lived in product and service, not in a zip of code. Hardware + software bundles, production-hardened B2B polish, and willingness to maintain all push the moat outside “can AI write this.”
Why “everyone can cook, yet restaurants thrive” fails for software
Skeptics love that analogy. The counter cuts the key difference: eating is a hard need; most software is not. Restaurants also sell not cooking, better taste, and social ritual. Software is often replaceable by “vibe one up for now.”
The only durable parallel is paying for stability, taste, peace of mind, and trust—and that is buying outcomes, not code.
Where moats move
Nearly every strong reply points the same way:
- Business and operations beat code
- Data + workflow + trust + domain depth + continuous learning replace feature stacking
- Long-term maintenance gets scarcer once building is easy
- Possible upside markets: software discovery, reviews, forks, API composition—scale still matters; not everyone must vibe from zero
Earnings-report skeptics note pro software and AI software stocks still rise. Do not confuse commodity tools getting cheap with the whole industry going to zero.
Addendum: flip to the user side — build it yourself or buy?
Source: AI brings not convenience, but more people getting lost (V2EX, 45 replies)
Everything above asks, from the seller’s side, whether software still has value. Three weeks later another V2EX thread flipped it: from the user’s side, is it worth building your own with AI?
The OP, an indie developer, noticed many people using AI to write their own budgeting and to-do apps, only to find the time and tokens cost more than just buying a mature app that is more polished and steadily maintained. His conclusion: maintaining a whole system to solve one problem is over-preparation; tokens are the small part, time and attention are the cost sink.
Most replies pushed back:
| Thread claim | What it says |
|---|---|
| Cheaper attempts mean more attempts; other people’s experiments just look like being lost to you | Experimenting ≠ being lost |
| The mature app you want people to buy started as someone’s own handy tool | Good products often start as self-use |
| Like spending on a game you’ll eventually quit — you still had the experience; “AI is TikTok for programmers” | Emotional value counts |
| Buy mature software for mainstream needs; build for niche needs (the OP agreed) | Split by how common the need is |
| Fork an open-source project and bend it to your habits; for closed apps, have AI clone just the missing feature | Modify beats starting from zero |
| Subscriptions are too expensive; AI at least brings cheap alternatives | Sellers lose pricing power |
| I can’t change the code AI wrote for me | Maintenance is the real cost |
Put back into this essay’s frame, the two threads are two sides of one thing:
- Below the kill-line, both sellers and buyers are redoing the math. Anyone can vibe a budgeting or to-do app, so sellers lose pricing power — the feature layer cheapening, as above.
- Building got easy; maintaining did not. Users who build their own save the subscription and take on long-term upkeep. That is the flip side of “long-term maintenance gets scarcer,” and it’s why mature products still get paid: users buy stability, peace of mind, and ongoing iteration, not code.
- The user’s version of “cook vs. eat out”: you can vibe up a meal, but you still wash the dishes.
When is it worth building your own? I’d ask five questions:
- Is the need mainstream or niche? Mainstream needs already have mature products — buy.
- Can you modify an open-source project instead of writing from scratch?
- Is the goal finishing a task or enjoying the process? For the former, add up the full cost; for the latter, the enjoyment is the return.
- Will you still want to fix it in six months? If you can’t change the code AI wrote, you don’t really own it yet.
- Might it become a product? Once others use it, you take on their needs and experience — a different ledger.
Our endgame: customization gets cheap too
Most of the thread argues about “is it cheap now.” The more useful frame is end-state layers:
- Feature layer: notes, translation, CRUD, generic sites—swallowed by defaults.
- Customization layer: project fees for requirement changes compress into chat with an agent; margins collapse.
- Still scarce: operated outcomes and who holds liability; distribution that picks safely and stays compliant; proprietary data and workflow memory you keep when you switch agents; physical world, licenses, supply chains.
Builder implication: stop selling “another app / another custom job.” Sell agent-ready capabilities, orchestration, SLAs, and services-shaped software—see SaaS as agent-usable capabilities and software engineering splits three ways.
Decision checklist
- Is this below the kill-line? If yes, do not bet a standalone paid product on it.
- If you sell customization: assume the customer next year edits it with an agent—does your pricing still work?
- Prefer business loops, data flywheels, workflow memory, hardware/licenses, and provable maintenance liability.
- Shift the pitch from “we have features” to “we deliver outcomes that get cheaper as models improve.”
Cheaper code does not erase value—value simply leaves implementation and bespoke change requests.
Comments