Building Bespoke Personal Software to Bypass Enshittified Ecosystems

The move toward AI-native hardware and personal software agents represents a shift from passive consumption to active, personalized systems. While the industry is currently fixated on foundational model wars and IPO-driven talent acquisitions, the real competitive advantage lies in integrating AI into physical and operational workflows. This conversation shows that the most durable value will not come from superficial coding or generic LLM wrappers, but from the ability to bridge the gap between high-level AI reasoning and low-level system execution, such as medical imaging or custom fitness management. Readers should look past the hype of the latest model releases and focus on the personal software trend: the ability to build bespoke tools that bypass the bloated, subscription-heavy software ecosystems currently dominating the market.

The Hidden Cost of Enshittification in Personal Software

We are seeing a systemic shift where legacy software, once useful, is becoming unusable due to aggressive monetization and enshittification. The speakers note that calorie-tracking apps, formerly straightforward utilities, are now so cluttered with conversion prompts that they are effectively broken for the end user.

The non-obvious consequence here is the rise of the personal software model. Instead of relying on third-party apps, users are beginning to build their own bespoke systems using LLM-based agents. This is a defensive move against a software ecosystem that prioritizes subscription growth over utility.

"I have found this app now almost unusable because A, I think they are just trying to convert everybody but there is something going on where there is just like an inch unification of that kind of software. I am really now seriously considering... that personal software thing is like I guess I could probably just do this myself."

-- Speaker (Gavin)

Why Midjourney’s Pivot Matters More Than the Model

Midjourney’s move into medical hardware, specifically a full-body ultrasonic CT scanner, is a classic example of a company using non-dilutive capital to solve a moonshot problem that public companies, constrained by quarterly earnings and IPO timelines, cannot touch. While critics focus on the fidelity of early sample images, they miss the systemic implication: a company that has not raised venture capital is free to pursue physical, high-risk, high-reward hardware that creates lasting market separation.

This creates a competitive moat that is invisible to the current AI arms race. While competitors are crushing coding benchmarks to satisfy public market expectations, Midjourney is building a new imaging modality. The downstream effect is a potential paradigm shift in early disease detection, moving from reactive, high-radiation scans to casual, frequent, and safe monitoring.

The Maintenance Pause and the Illusion of Progress

The current state of AI development is defined by a strange, temporary stasis. As developers wait for access to high-tier models like Fable 5, they have shifted their focus from ambitious, foundational refactoring to routine maintenance and backlog pruning.

This creates a hidden dynamic: the industry is currently waiting in the wings. The implication is that when access to these models is restored, there will be a sudden, massive surge in sophisticated implementation. The conventional wisdom that AI coding is just for vibes fails here; the reality is that professional engineering teams are already deeply embedded in these workflows, and their current pause is a strategic holding pattern, not a lack of interest.

"None of them seem to be implementing like major foundational stuff or refactoring important components of their of their builds because they are all waiting in the wings... But the moment fable comes back then they will tackle the more ambitious stuff."

-- Speaker (Gavin)

Key Action Items

• Migrate to Personal Software: Over the next quarter, identify one utility app you pay for that has become bloated. Replace it with a custom Claude or Codec-based prompt system. This creates a long-term advantage by removing dependency on predatory subscription models.
• Implement Token Compression: Integrate tools like Headroom into your LLM workflows immediately. This reduces operational costs and allows for the processing of significantly larger datasets without losing accuracy, providing a structural edge in efficiency.
• Shift from Vibe Coding to System Integration: Stop focusing on simple prompt outputs. Invest time in learning the Model Context Protocol (MCP) to connect your AI agents directly to your software engines, such as Unreal Engine. This pays off in 12 to 18 months by allowing agents to execute complex, multi-step builds autonomously.
• Adopt Benchmaxing for Health: Use existing open-source prompt libraries to turn your LLM into a dedicated, private fitness trainer. This avoids the enshittification of commercial health apps and provides a highly personalized, data-driven routine.
• Monitor Non-Public R&D: Pay closer attention to companies like Midjourney that are not beholden to IPO timelines. Their ability to take risks on hardware, such as medical devices, suggests they are playing a different game than the foundational model providers, which may yield higher long-term dividends.