The Deep Feed

Success is a trap. For most founders, the goal is to build something that lasts, yet the very act of succeeding creates a new set of pressures that actively work against the original mission. This is what Eric Ries calls financial gravity. Once a company reaches a certain scale, the primary objective often shifts from solving a problem or fulfilling a purpose to satisfying the quarterly demands of investors and the stability requirements of a large workforce. This shift is not a sudden event but a slow, predictable slide into mediocrity. The original spark that ignited the company is replaced by a desire to protect the current state, leading to a culture of risk aversion and bureaucratic bloat.

The Governance Gap

The statistics are grim. Roughly 80% of venture-backed founders are forced out within three years of their company going public. This happens because the governance structures designed to manage growth are rarely the same ones designed to protect a mission. When a company goes public, the fiduciary duty to shareholders often clashes with the founder's original vision. If the mission requires long-term thinking or unconventional methods that might hurt short-term margins, the board will almost always side with the margins. Without specific legal protections, the company becomes a vessel for capital rather than a vehicle for an idea.

Success won’t protect you—it instead makes you a bigger target.

To combat this, a new breed of company is emerging, using specific legal and structural tools to insulate themselves from the whims of the market. Look at Anthropic or Costco. These organisations do not just hope to stay true to their values; they bake those values into their legal DNA. By using structures like Public Benefit Corporations, they change the rules of the game. They create a legal mandate that allows them to prioritise their mission even when it conflicts with immediate profit maximisation. This isn't about being 'nice'; it is about survival in an era where mission-driven companies can move faster and more decisively than those bogged down by conflicting interests.

Building an Incorruptible Structure

Building a company that withstands any era requires more than just a good product. It requires an understanding of how power and money move through an organisation. You must decide early on how much control you are willing to trade for capital. If you want to remain the architect of your vision, you cannot rely on the standard venture capital playbook. You need to implement governance that survives the transition from a scrappy startup to a global entity. This might mean dual-class share structures, specific board compositions, or legal filings that take only a few pages but change the entire trajectory of the business.

Ultimately, the goal is to build something that is not just profitable, but resilient. A company that can survive the transition to public markets without losing its soul is a rarity. Most will fail this test. They will become the very things they sought to disrupt: large, slow, and ultimately uninspired. The founders who succeed are those who treat governance as a core product feature, not an afterthought of the legal department.

There is a pervasive sense of guilt in modern technical work. We are surrounded by tools that are faster, better, and more efficient than anything we could build ourselves. If you are writing a custom script to parse a file when a standard Unix utility like `awk` could do it in one line, you feel like a failure. You feel like you are wasting time. This guilt stems from a misunderising of what it means to be a practitioner. We often mistake the use of tools for the mastery of a craft. But there is a fundamental difference between using a tool and understanding the problem the tool solves.

The Trap of Idle Study

The trap is believing that you should only build what is not already built. This leads to a state of perpetual, shallow study. You spend your time reading documentation and watching tutorials, trying to find the 'perfect' implementation that someone else has already perfected. But this is a form of procrastination. It feels like work, but it doesn't push you toward the frontier. You are simply standing on the shoulders of others without ever learning how to balance. True progress requires the friction of implementation.

You need to reinvent a couple of wheels to get to the edge of what we know about wheel-making.

Andrew Quinn argues that the goal is not to reinvent a thousand wheels, but to reinvent a few. By building a search-and-replace tool from scratch, or by implementing a basic database engine, you encounter the specific, messy problems that the creators of the 'standard' tools had to solve. You learn why certain trade-offs were made. You see where the existing tools fail. This direct engagement with the mechanics of a problem propels you toward the true frontier much faster than any amount of reading ever could. The struggle is not a waste of time; it is the engine of expertise.

In highly developed fields like mathematics or computer science, this might mean reinventing twenty or thirty fundamental concepts. But the principle remains: you cannot understand the edge of a field if you have never fought your way to it. The tools we use are meant to be foundations, not ceilings. When you build your own version of a tool, you aren't just making a copy; you are building the intuition required to use the next generation of tools more effectively.

We have built a digital world that prizes speed above almost everything else. Nowhere is this more evident than in WebRTC, the protocol that powers our real-time video and audio calls. WebRTC is designed with a single, uncompromising goal: keep latency low. To achieve this, it is programmed to be aggressive. If the network gets congested, the protocol doesn't wait for the missing data to arrive; it simply drops the packets. It would rather give you a distorted, garbled audio stream than a five-hundred-millisecond delay. In the world of voice calls, we have accepted that 'broken' is better than 'slow'.

The AI Mismatch

This design philosophy is running headlong into the reality of Large Language Models (LLMs). When you interact with a voice-based AI, you aren't just listening to a person; you are waiting for a complex reasoning process to complete. The value of the interaction lies in the accuracy and depth of the response. If the network drops a packet during an AI's response, you don't just get a momentary glitch in audio; you get a garbage response. A broken sentence from an AI can render the entire computation useless. In this context, the 'real-time' priority of WebRTC is actually a liability.

I would much rather wait an extra 200ms for my slow/expensive prompt to be accurate.

As Luke Curley points out, users are willing to trade a small amount of latency for a massive increase in quality. We are paying for the intelligence, not just the speed. If a prompt takes an extra fraction of a second to arrive but results in a coherent, helpful answer, that is a successful transaction. WebRTC, however, is hard-coded to prevent this. It is built on the assumption that retransmitting data is a failure. For the next generation of AI-driven interfaces, we need protocols that understand the difference between a human stutter and a computational error.

The technical difficulty of changing this is significant. Major platforms like Discord have attempted to work around these limitations, but the protocol itself is fundamentally built for 'real-time or else'. As we move toward more sophisticated agentic interfaces, the industry will need to develop new ways to handle data delivery—ways that treat accuracy as a first-class citizen alongside latency. We need a protocol that knows when to wait.