Hostile Dark Patterns

The essence of SaaS lock-in, dark patterns designed to separate you from your cash.

Hostile Dark Patterns

At times it may seem like this post is a conspiracy theory. We all face the reality that SaaS platforms create an economic bribery model that cannot be underestimated. Anyone who has worked at a SaaS company understands two fundamental business drivers:

1. The enticing benefits of cloud computing;

2. Switching costs.

Customers love it when you maximize #1. They abhor it when you optimize #2 to chain the exit doors.

These typical examples of user-hostile dark patterns align with the concerns raised about SaaS (Software as a Service) platforms. I suspect there are many more.

1. Lock-in Mechanisms (Optimizing Switching Costs):
   SaaS platforms often design their systems to make it difficult or costly for users to leave. For example, they might store user data in proprietary formats incompatible with other tools, requiring significant effort or expense to export and migrate elsewhere. This creates a dependency that traps users, even if the service becomes unsatisfactory or expensive.

2. Hidden Data Ownership Traps:
   The article hints at users "handing over their most precious digital assets" to SaaS companies. A dark pattern here could be unclear or buried terms of service that imply the platform retains rights to user data, leaving users unaware that they no longer fully own or control their content. This lack of transparency exploits user trust.

3. Gradual Price Creep (Bait-and-Switch Pricing):
   This dark pattern lures users in with low initial costs or free tiers, only to increase prices incrementally once they are deeply invested. Switching makes it financially unsustainable, even as the service’s value diminishes.

4. Forced Continuity:
   SaaS platforms might auto-renew subscriptions without clear reminders or make cancellation processes intentionally convoluted—think hidden menus, multi-step confirmations, or requiring users to contact support directly. This keeps users tied to the service against their will, amplifying the "switching costs" the article critiques.

5. Disappearing Act Risk:
   A lack of user agency. A dark pattern here is the absence of clear data export options or warnings about service discontinuation, leaving users blindsided and scrambling to recover their data when a platform shuts down unexpectedly.

This is why I aggressively lean into file-over-app.

User-hostile dark patterns have driven me and countless others to seek ways to insulate, sometimes eliminating dependencies on SaaS platforms that could vanish without notice or slowly raising rates beyond reasonable cost-benefit ratios.

This is what software could be: private, secure, delightful, free of dark patterns, fully aligned with the user, where you retain full control and ownership of your data in simple, universal formats, and where tools can be extended and composed. —Andrej Karpathy

In my world of note-taking and sense-making, file-over-app became a conscious movement sometime in 2019. Steph Ango supercharged my interest in 2023 and has expanded ever since, dominating my thoughts daily. The most important takeaway from Steph’s article:

It’s the plain text files I create that are designed to last. Who knows if anyone will want to read them besides me, but future me is enough of an audience to make it worthwhile.

To say file-over-app is a movement may be premature. Vastly, the circles I run in seem content with handing over their most precious digital assets to SaaS companies. There’s little evidence this will change anytime soon.

At its core, Steph asserts …

File over app is a philosophy: if you want to create digital artifacts that last, they must be files you can control, in formats that are easy to retrieve and read. Use tools that give you this freedom. File over app is an appeal to tool makers: accept that all software is ephemeral, and give people ownership over their data.

Despite aggressively adopting this mantra, I still have much work remaining to insulate myself from hostile dark patterns fully. I have a plan, which is to say, I still don’t have it dialed in. But I can safely say that my plan is sound. It is validated daily by subtle and mostly fleeting moments of delight.

A few days ago, my team met to discuss several QA issues on an AI project. Earlier that day, through automated AI tests, I discovered a previously unknown gateway to jailbreak a customer-facing AI agent. The details of the issue have not yet been formalized in the issue tracker. They were well-documented but scattered across several Slack, Teams, and Email apps.

Using a single, fully localized AI-driven search app, I summarized them in seconds and brought the entire team up to speed with examples and a deep awareness of my discovery. If I had been forced to rely on several SaaS platforms to do this, I would have failed to brief the team during the meeting. I instantly performed a keyword search that identified dozens of references from five different apps. Then, I used this prompt to distill what I knew about this topic.

Summarize the conversations concerning jailbreak and identify the patterns discovered in QA tests.

Pardon the redactions. This was generated in ~500ms from a local search/AI summation using ScreenPipe and Ollama.

This is a file-over-app example, vastly unlike the usual and customary examples, such as using Obsidian to create and manage note-taking in your file system. Despite this unique approach, it is 100% open-source and easily maintained. I fully control it and I’m free to alter how it generates files, organizes them, and how I archive them (if at all). It ticks all of Steph Ango’s file-over-app boxes.

How it Works

ScreenPipe takes a snapshot of my desktop every 30 seconds and converts the images to text (using OCR). It also provides an API to query the filesystem of these documents and captures all audio from videos you play to Zoom calls you attend. All of it is transformed into text files.

Ollama, a locally installed LLM-agnostic chat app, provides an interface that ScreenPipe can access to share its search results. Formulating insights from this data becomes effortless and fast.

There are no external dependencies, and no SaaS systems are involved. Best of all, no LLM token costs or latency to generate AI output except for the processing capacity of your PC. I’m using Apple Silicon, so I enjoy relatively fast throughput. Mileage may vary.

App-less Note-Taking and Sense-Making

With record-the-world capability and cost-free, localized generative AI, we’re on the threshold of a new pattern: digital work free of early binding our knowledge through hand-crafted notes and observations.

If we have the genetic material of every conversation, every website visited, and every word or sound we’ve ever heard, why do we need to take notes?