Ephemeral Software
Programs as exhaust, generated on demand
"I just wanted a bespoke piece of software for myself specifically and then caused it to get written by the LLM."
The Pattern
Software as exhaust:
- Generated on demand
- Shaped to the moment
- Disposable (or keepable)
- The API token is the real primitive
When semantic transpilation is fast enough — and it's getting faster — the economics of software invert. Custom becomes cheaper than generic. Bespoke becomes the default. SaaS becomes a relic.
Why This Kills SaaS
| Traditional SaaS | Ephemeral Software |
|---|---|
| Generic features for generic users | Exact features for exact needs |
| Lock-in via data silos | Your data stays yours |
| Rent-seeking on commodity capabilities | Generate, use, discard (or keep) |
| You adapt to the software | The software adapts to you |
The moat is taste plus knowing what you want.
If you can describe what you want precisely, you can have it. The constraint isn't technology — it's clarity of intent. Which is why valence mining matters: it's the practice of developing that clarity.
Examples
One-Off Scripts
Need to process a CSV? Merge two datasets? Convert a file format? Generate the script, run it, delete it. The script is exhaust. You never needed a "data processing tool" — you needed that specific transformation, once.
Landing Pages
Describe the page. Generate. Deploy. Done. The page exists because you needed it, not because you bought a template and configured it. This very site was generated this way.
Client Deliverables
A client needs a proposal document. A website. A dashboard. Describe it precisely and the LLM builds it. The deliverable is ephemeral software — generated for their specific needs, not a WordPress template configured for their content.
Internal Tools
Need a timesheet viewer? A task tracker? An invoice generator? Don't buy one. Don't even build one permanently. Generate it when you need it, shaped to exactly how you work.
The Economics
Traditional software has fixed costs amortized across users. Ephemeral software has marginal costs per generation.
As generation cost approaches zero:
- Custom software becomes cheaper than generic
- The build-vs-buy calculus inverts
- SaaS moats erode
- The API token becomes the real primitive — the wand's core
We're not fully there yet. But we're approaching fast enough that betting on this direction is rational.
What Persists
If software is ephemeral, what persists?
- Data — Your data is yours. The software is just a lens.
- Intent — The specifications (pages, tables, issues) survive across generations
- Patterns — What worked before works again (the memory bank)
- Relationships — The people you serve, the problems you solve
The software is exhaust. The rest is real.
Opus-class models running self-hosted. Every person has a "little guy" that generates software on demand. Proprietary software can't compete with free generation. The question isn't "which product do I buy?" but "what do I want to exist?"