The Signal Layer: What to Build When Anything Can Be Built — Lena Hall, Akamai

Lena Hall argues that while AI has brought unprecedented abundance in output, speed, and leverage, this universal access means the cost and value of “average” work have plummeted to zero. The same AI that makes you fast also makes your competitors fast, enabling them to replicate features almost instantly.

The Convergence Machine and the Scarcity of Vision

AI acts as a “convergence machine,” answering from common knowledge derived from past data. When asked questions like “what users want” or “what to build,” it provides competent, confident, yet identical answers to everyone. This means that to differentiate, one needs a unique vision—a point of view on where things are going—and then leverage AI for execution. The one decision AI cannot make is “what to point at.” Therefore, the new critical job is deciding what AI makes and providing the unique reason for customers to choose your version.

Hall observes that while there’s an abundance of innovative products, many in the market sound identical. To stand out, companies must master two critical aspects of what she calls the “signal layer”:

1. Knowing Your Signal (Build Side): Clearly defining what you’re building, why it’s unique, and how it avoids being merely “average.” This encompasses the product, code, and roadmap.
2. Emitting That Signal Without Distortion (Ship Side): Ensuring that what customers believe about your product accurately matches your intent. This involves content and go-to-market strategies.

The Graded vs. Ungraded Problem

AI agents have drastically improved coding efficiency (e.g., solving high 80s% on benchmarks vs. a fraction two years ago). This highlights a core principle: “Anything you can measure, you can train against.” Tasks with free graders (like compilers or test suites) can be automated and optimized by AI until they “win.” Code automation was a primary target due to its checkability. However, the most buildable thing is rarely the most valuable. AI will build whatever it’s pointed at but offers no insight into where to point.

The true challenge, then, shifts from implementation to “pointing.” Paul Graham’s advice suggests finding what people genuinely want by experiencing the need yourself, building for yourself and friends before the market forms or surveys can identify it. These “weird, specific, genuinely embarrassing ideas” are the unique signals AI won’t proactively generate.

Taste is Trainable, Judgment is Not (Mostly)

Hall refutes the idea that “good taste” will save you, explaining that taste is “preference under feedback,” which is precisely what AI systems learn and can imitate. Broad good taste is no longer a differentiator.

What resists training and remains durable are two narrower forms of judgment:

1. Judgment about things that have not happened yet: There’s no data for future events.
2. Judgment inside a relationship the model cannot observe: The model has consumed all data about your customer but has never met them or understood their specific history and context.

The Hamming Problem in the Age of AI

Drawing on Richard Hamming’s observation about great scientists working on important problems, Hall notes that a problem is “important” only when you have a “reasonable attack” on it. While Hamming lamented the rarity of such attacks, AI has now provided “an attack on everything.” The new scarcity is knowing which problem is actually worth attacking. This judgment comes from deep, personal engagement with a domain, understanding its “battle scars,” and possessing insights that lie in the delta between AI’s training data and what should exist.

Protecting Your Signal from Distortion

AI’s “convergence machine” effect extends to content, leading to a proliferation of identical-sounding posts and articles. Readers quickly recognize and skip AI-generated generic content. While AI can efficiently automate content, it fills any gaps with “sameness.”

Therefore, there are two distinct ways to use AI for content:

1. Average Prompt, Average Output: This efficiently automates irrelevance, adding another indistinguishable drop to a sea of similar content.
2. Specific Point of View, AI for Convergence: Provide your unique insight, the “real story” AI cannot generate, then let the machine handle formatting, drafting, and optimization.

Hall identifies three key points where the signal can distort on its journey to the customer:

1. Source Distortion (Startups): Founders, deeply understanding their product, often compress its essence past legibility, assuming context their audience lacks. They focus on technical coolness, deleting the “customer pain” from the narrative.
   • Fix: Begin the story with the user’s pain that the product alleviates.
2. Organization Distortion (Big Companies): As the signal travels through layers of management, legal, and sales, it gets “rewound towards the average.” This isn’t due to incompetence but the difference in investment; founders sweat “unaverageable details” due to personal stake, while others ship to spec.
   • Fix: Reattach the signal to the outcome, like a founder, by implementing a “thin signal layer” in go-to-market engineering to preserve the original intent across handoffs.
3. Machine Distortion: AI remixes careful claims (with clear scope and evidence) into broader, less nuanced statements (e.g., a “narrow eval” score becoming a general “promise”).
   • Fix: Engineer the signal so that the “limit is welded to the claim.” For instance, a monitoring tool’s unique signal (“tells you what not to wake up for”) should be articulated with its inherent constraints (“stays quiet on anything it can’t tie to a real user impact and shows you everything it silenced so you can overrule it”). Ensure these limits cannot be easily edited out by AI in different content formats.

The Ultimate Scarcity: Trust

All efforts in building, shipping, and protecting the signal converge on one goal: earning trust. In a world of infinite, identical alternatives, trust is the only differentiator. Trust has no grader, no benchmark, and cannot be fully automated; it is granted slowly through relationships and consent.

Producing averageness is not free; it costs resources and makes a company harder to choose. Hall concludes that while AI has shifted the locus of value, the ultimate scarcity is trust. Therefore, when you can build anything, you should build trust by having strong conviction, defining your signal clearly, protecting it from distortion, and aggressively using AI for everything else.

• Tags:
• Go-to-Market
• Feature
• AI Strategy
• Organizational Design
• Innovation
• Trust
• Decision Making
• Product Differentiation
• Customer Insight
• Value Creation
• Hamming Principle