I had an idea for a small website. A tidy little thing: pull together useful info, wrap it in a clean UI, ship. I opened a code-assist tool, generated the scaffolding, stitched the API calls, and—surprise—the whole project came together in hours. Then a colder realization landed: the same questions my site answered could be answered, more completely and interactively, by an AI assistant. If it’s this easy for me, it’s this easy for everyone. Suddenly, the project felt like packaging air.
What still makes sense to build in a world where AI can fetch the information, write the code, and even run small artifacts on demand? Are we entering a “fast-fashion” era of SaaS—rapidly produced, quickly commoditized, and just as quickly discarded? And if so, where is the durable value?
The Fast-Fashion Era of SaaS
Think of the last decade as artisanal tailoring: lots of hand-stitched UIs for narrowly scoped problems. Today, a big slice of that catalog collapses into prompts. “Answer engines” and “thin wrappers” that once needed a custom site now emerge as a quick instruction to an assistant. The cycle time “idea → prototype → clone” has compressed from months to hours.
This doesn’t mean software disappears. It means the surface area of many products shrinks. Interfaces that existed mainly to move data from A to B or to serialize a few deterministic steps get absorbed into conversation: “Do X with Y, using Z constraints.” Where we once built knobs and dropdowns, we now specify intent.
What Still Deserves to Exist
Not everything is a prompt. Durable products concentrate where AI is necessary but not sufficient. Three buckets stand out:
1. Guaranteed outcomes over generic answers.
When stakes are high (e.g., money movement, compliance, safety, SLAs…) people need assurances, not just suggestions. That means verifiable pipelines, deterministic constraints, audit logs, approvals, and graceful failure modes. AI can be the engine, but the product is the guarantee.
2. Proprietary context and distribution.
Value pools around unique data, hard-to-recreate integrations, entrenched workflows, and channels to reach users. If you own the context (customer history, domain-specific knowledge, private graphs) and the distribution (where work already happens), your product can deliver outcomes an open-ended assistant can’t reliably match.
3. Taste, trust, and time compression.
Sometimes the difference is craft. Opinionated UX that encodes best practices, prevents common errors, and gets a novice to a pro-grade result in two clicks; that’s not easily replaced. When “correct” is subjective but quality is felt (design, marketing, creative tooling), taste becomes a moat. People pay to save time and avoid doubt.
New Interfaces
Traditional UIs were built for mechanical interaction: click this, then that, then export. As systems get smarter, the primary interface becomes intent, and the secondary interface becomes verification. A useful mental model:
- Autopilot: “Do the thing.” Let the assistant run with it—generate, transform, orchestrate.
- Cockpit: Show me what you’re doing. Live status, intermediate artifacts, confidence, costs, risks.
- Manual override: I’ll take it from here. Edit, constrain, roll back, pin versions, set policies.
Products that thrive will choreograph these three modes. The UI gets thinner but more meaningful: fewer buttons, more levers that set guardrails, more receipts that build trust.
Lessons from History
We’ve been here before. Calculators didn’t kill spreadsheets; they made spreadsheets inevitable. Search didn’t end publishing; it ended low-value “content farms” and rewarded trustworthy brands. App stores didn’t eliminate the web; they forced clarity about what needed native performance and device access.
AI is compressing the low-value layer again. Commodity “answer sites” and single-purpose utilities look increasingly like features inside an assistant. But the systems that coordinate people, data, and guarantees—those tend to persist and even expand.
A Builder’s Litmus Test
Before investing months, run the idea through five questions:
1. Is this mostly an answer?
If yes, ship it as a prompt for an AI assistant, a template, or a plugin, not a standalone product.
2. What has to be guaranteed?
Name the SLAs, risk boundaries, and audits. If you can’t, you’re probably building a thin wrapper.
3. What proprietary context do you control?
Private data, integrations, workflows, or distribution channels that an assistant won’t have by default.
4. Where will trust be won or lost?
Show your receipts: citations, diffs, costs, approvals, rollback. Make “why this output?” obvious.
5. What becomes easier over time?
If your differentiation is “we call the model,” that advantage decays. If it’s “we encode expertise, guardrails, and networked context,” that advantage compounds.
What to Build Now
- Outcome layers over models. Ship “Get me from A→B under these constraints” rather than “Here’s a tool, good luck.”
- Orchestrators with receipts. Chain actions, capture evidence, log decisions, make it auditable.
- Context engines. Connect customer data, permissions, and domain knowledge so the assistant acts with your understanding, not a generic one.
- Opinionated micro-UIs. Small, high-leverage surfaces: previews that invite edits, sliders that trade speed for accuracy, toggles that set risk posture.
- Distribution-first integrations. Live where users already are: docs, chats, CRMs, IDEs, and make your value ambient.
The Near Future Is Hybrid
In the short to midterm, lots of tasks will be solved by prompting alone; a meaningful fraction will still need structured interaction. You’ll ask, “Do the quarterly roll-up,” and get a solid draft. Then you’ll refine in a small, purpose-built panel that understands your chart conventions, legal disclaimers, and accounting rules. Less clicking through menus; more steering the system with high-level intent and low-friction corrections.
Takeaway
AI didn’t make my idea worthless; it revealed which part of it was commodity. The work that matters now is moving from building interfaces that do steps to building systems that deliver outcomes with guarantees. If your product’s value is “we formatted some information,” the assistant will outpace you. If your value is “we deliver the right result, with context, trust, and speed,” the assistant becomes your cofounder, not your competitor.
- Build for intent.
- Instrument for trust.
- Compound with context.
That’s how we make things that still matter when the easy parts are free.