The PDF Reader That Started a Bigger Question
Matthew Berman's recent experience installing Adobe Acrobat for a simple PDF task highlights a growing frustration. What should have been a basic operation—reading and annotating a document—became a frustrating gauntlet of extra features, persistent prompts, and account login friction.
This ubiquitous bloatware problem led Berman to a striking conclusion: why not build a custom PDF tool? His Folio example demonstrates the appeal of bespoke software: a lightweight reader with only the specific functions he needed, including notes, annotations, e-signing, and the ability to ask PDF questions.
The wider implication is profound. AI-generated applications are making such tailored software plausible for individuals, shifting the "build versus buy" calculus. While a working demonstration like Folio showcases the power of autonomous agents, it's crucial to distinguish between a functional prototype and a production-ready, fully supported replacement.
Yet, the convergence of increasingly capable AI models, plummeting token costs, and sophisticated agentic workflows means that the barrier to creating custom software is rapidly dissolving. This shift suggests a future where individuals routinely generate their own tools, precisely configured for their unique needs, rather than adapting to off-the-shelf solutions.
Three Shifts Made Building Feel Cheaper Than Buying
Berman identified three critical shifts that made building feel cheaper than buying. First, models like Anthropic’s Claude 3.5 Sonnet / Opus achieved sufficient reasoning capabilities to handle complex coding tasks. Second, token costs dropped dramatically, making extensive generation affordable. Finally, autonomous agent workflows emerged, allowing models to iterate and refine code with minimal human intervention.
This convergence fundamentally alters the build-versus-buy calculation. The cost of generating a custom application and maintaining it now frequently undercuts the recurring fees of commercial subscriptions, which can range from $10 to over $50 per month. This equation, however, is dynamic, depending on an application’s scope, usage, and upkeep requirements.
Berman demonstrated this new reality with practical examples. He rapidly prototyped an Excel-style browser application, achieving near feature parity with a $10 token spend, compared to Microsoft Excel’s $180 single license or $7-10 monthly subscription. He also developed a YouTube analytics utility, showcasing how hyper-personalized tools can be built precisely to individual specifications without bloat. These examples illustrate the power of rapid, bespoke software creation.
The Clone Can Look Right—and Still Fall Short
Cloning existing software interfaces has become remarkably simple. Projects like ArtCraft, featuring PhotoCraft, FilmCraft, and VectorCraft, exemplify open-source attempts to replicate familiar creative workflows. These were generated by pointing an AI at existing software and iterating until an exact visual and functional clone was produced.
Despite the impressive visual fidelity, a significant 90/10 gap often emerges. While AI excels at matching menu structures and interface patterns, reproducing deep, nuanced behavior proves more challenging. Critical elements like accurate color management, codec support, cross-application file compatibility, GPU performance, and long-term stability often fall short.
Beyond technical hurdles, practical and legal questions arise. Generated code, even for personal use, still requires security review and ongoing maintenance to prevent vulnerabilities. Furthermore, close imitation of proprietary interfaces and functionality can invite legal scrutiny regarding copyright, trade-dress infringement, and reverse-engineering concerns. Explore the code for PhotoCraft and other ArtCraft projects directly at the PhotoCraft GitHub Repository.
Enjoying this? Get one like it in your inbox each morning.
one email a day · unsubscribe in two clicks · no third-party tracking
SaaS Won’t Vanish; Its Moat Will Move
SaaS is not vanishing, but its competitive moats are shifting. Narrow utilities are most exposed to this new dynamic. Basic PDF readers, simple converters, and lightweight analytics apps with limited feature sets are ripe for replacement by bespoke, AI-generated alternatives.
Conversely, large enterprise platforms remain robust. Their value is rooted in complex features like:
- Collaboration tools
- Cloud synchronization
- Deep integrations with other business systems
- Stringent compliance and security certifications
- Dependable support infrastructure
These platforms offer more than just a familiar interface; they provide a comprehensive ecosystem that is difficult for a single person—or even a small team—to replicate and maintain.
Experiment with a small personal app or an established open-source alternative like ArtCraft’s PhotoCraft, FilmCraft, or VectorCraft. Test its functionality thoroughly, protect any sensitive data, and account for ongoing upkeep before canceling a subscription. The future of software is more personal, but established services still offer critical reliability.
Frequently Asked Questions
Can AI agents build a usable app from a prompt?
They can often produce working prototypes and focused utilities, but complex apps still need testing, refinement, and human review.
Is building software with AI cheaper than paying for SaaS?
It can be for a narrow, stable use case. Include the cost of testing, hosting, security, and ongoing maintenance before comparing.
Can an AI-generated clone replace professional software?
Not necessarily. A clone may imitate the interface while missing reliability, edge-case handling, performance, or compatibility.
What software is most vulnerable to AI-built alternatives?
Simple, single-purpose utilities are easier to reproduce than tools whose value depends on collaboration, compliance, or cloud infrastructure.