The Promptware Window
Every week I see pitches that follow a similar pattern: “X for Y, powered by AI.” The demos look slick. The traction slides show early revenue. But peel back the layers, and it’s often the same thing: a foundation model wrapped in a clever prompt.
I call this promptware.
What is Promptware?
Promptware = Foundation model + Domain-specific prompt + UI
Take any off-the-shelf AI model that handles voice, text, or summarization. Add a prompt that says “for dentists” or “for real estate agents”. Wrap it in a clean interface with some industry lingo. Voilà—you’ve got yourself a SaaS startup!
The anatomy is always the same: constrain a general-purpose model with a domain-specific prompt. What’s new is that a well-designed prompt now replaces what used to require thousands of lines of custom code.
With tools like Claude Code and Cursor, even software development itself is becoming prompt-driven. The gap between “I need software that does X” and “I have software that does X” has collapsed from months of work to minutes of prompting. That shift fundamentally changes what it means to build a defensible software business.
Isn’t all software just abstractions?
A fair critique is that every wave of software has been about layering domain knowledge on top of existing technology. The difference is that in the “before times,” encoding that knowledge was genuinely hard. Good engineers and product managers had to translate messy human processes into thousands of lines of brittle code. Whole disciplines like domain-driven design emerged to capture the nuance.
Promptware changes this dynamic. Today, much of that translation can be compressed into a single well-crafted prompt without the translation overhead of before. The art isn’t gone, but it’s shifted: the barrier to building something that looks like “real software” is dramatically lower, which makes defensibility the central question.
A Quick History Lesson
Think about conversation intelligence. Gong built a valuable business analyzing sales calls because transcription and analysis were expensive and technically complex. Fast forward: Granola, Fireflies, Otter, and even OpenAI’s own ChatGPT Record have commoditized the entire category. What once required a huge engineering team can now be replicated with a decent prompt and a weekend hack.
This same playbook is running across verticals—sales coaching, medical scribing, education feedback. Promptware everywhere.
Why Promptware Is Exploding
Three forces explain the boom:
AI adoption is mainstreaming. Buyers who once balked at giving employees raw model access are fine with “AI for legal” or “AI for healthcare”, especially if it comes stamped with SOC2 or HIPAA.
Implementation is trivial. What used to take 20 engineers now takes a few API calls and some clever prompts.
Verticals sell better. “AI that understands me and my industry” is an easier pitch than “good luck writing your own prompts.”
That’s why I see two or three decks a week that look exactly the same: same foundation model, different wrapper.
The Coming Commoditization
Promptware can generate millions in ARR today. But the advantage is fleeting:
Easy to clone. If your moat is a prompt, someone can rebuild you over a weekend.
Incumbents are catching up. CRMs, EHRs, and legal platforms are all baking AI in natively. Why pay for a standalone tool when it’s bundled into what you already use?
Users are getting smarter. As prompt literacy spreads, the “premium” for pre-packaged prompts will shrink fast.
Distribution still wins. Companies that already own workflows don’t need the best AI, just “good enough” AI and their existing channel.
In other words: promptware makes for a great demo and small business, not necessarily a durable venture-scaled business.
When Promptware Becomes Real Software
The companies that survive will evolve beyond prompts. Four common traits:
Proprietary data. Training on unique datasets—case outcomes, medical records, financial filings—that compound over time.
Workflow integration. Embedding directly into systems customers can’t live without (Salesforce, Epic, SAP).
Network effects. Tools that improve as more people use them (e.g. developer feedback loops in Claude Code).
The Investor Lens
When I see promptware pitches, I ask three questions:
What proprietary data will you own in 12–18 months?
Does this product save or make money in a measurable way? (“Saves time” isn’t enough.)
Who’s the economic buyer, and do they have to buy it or is it a nice-to-have?
The Path Forward
Promptware isn’t useless, it’s a necessary stage in the AI cycle. These companies are teaching markets, solving real problems, and generating revenue. But the advantage is temporary and I’m not sure everyone realizes this.
The winners will use this window to build data moats, embed deeply, and create switching costs. The others will optimize prompts while the ground shifts under their feet.
So the question I keep asking founders is simple:
👉 Are you building a company or just a clever prompt?