The Deep Feed

01 — Lenny's Newsletter

The Designer's New Frontier

Why the rise of AI is a moment of radical simplification for product creation

By Lenny Rachitsky · 10 min read

Editor's note: As AI begins to write code and generate assets, the role of the designer shifts from builder to architect of intent.

The traditional design process is under siege. For decades, the job of a product designer involved a heavy lift of pixel-pushing, prototyping, and managing the friction between an idea and its digital manifestation. But as generative models begin to handle the heavy lifting of execution, the role is undergoing a violent transformation. Ian Silber, the head of product design at OpenAI, argues that we are entering the most significant era for designers in history, even if the industry has yet to catch up to the speed of the engineers. While engineers have used AI to increase their output tenfold, design teams often remain stuck in old workflows, trying to apply traditional methods to a medium that is fundamentally changing.

The Death of the Pixel-Pushing Era

In the old world, design was about the arrangement of elements. In the new world, design is about the orchestration of capabilities. When a model can generate a user interface or a piece of marketing copy in seconds, the designer's value moves upstream. It is no longer about how a button looks, but about the logic of the interaction and the intent behind the user's journey. Silber suggests a counterintuitive approach for modern teams: 'Just do less.' This isn't an invitation to laziness, but a call to focus on the high-leverage decisions that AI cannot make—the direction, the taste, and the fundamental understanding of why a product should exist in the first place.

The designer's job is moving from the execution of details to the curation of intent.

This shift creates a new hierarchy of skills. The ability to use a tool is becoming a commodity. The ability to possess a unique point of view, to understand the psychological needs of a user, and to invent entirely new ways of interacting with technology is where the competitive advantage lies. We are moving toward a 'super app' future where ChatGPT and similar interfaces act as the primary layer between humans and the digital world. In this context, the designer is not just making a tool; they are designing the very way humans experience intelligence.

Where humans maintain the advantage:

User empathy and psychological insight
Original invention and creative leaps
The ability to hold a consistent point of view
Defining the 'why' behind the product

The tension in the industry right now is between those who see AI as a threat to their craft and those who see it as an expansion of their agency. The former will focus on defending their territory in the realm of execution. The latter will embrace the vacuum left by automation to focus on the strategic and philosophical questions of product experience. The winners will be those who stop trying to compete with the speed of the machine and start leveraging it to reach higher levels of abstraction.

Key Takeaway

Design is shifting from the labor of making to the discipline of deciding.

02 — The Marginalian

The Void of Actuality

Finding reality in the gaps between signals

By Maria Popova · 12 min read

Editor's note: A philosophical look at why the present moment is often lost in the noise of our memories and expectations.

We live our lives in the wake of signals. Most of what we perceive as 'reality' is actually the delayed transmission of events that have already passed. Like the light from a distant star that reaches us long after the star has died, our consciousness is often populated by the ghosts of previous moments. We react to memories, we plan for futures that do not yet exist, and we mistake these projections for the present. The art historian George Kubler proposed a different way to view this: the concept of 'actuality.' Actuality is not the signal; it is the void between the signals. It is the silent, infinitesimal gap between the ticks of a clock, the moment where nothing is happening, yet everything is possible.

The Paradox of the Present

The difficulty of inhabiting the present lies in the fact that our brains are designed to process change and permanence. We understand time by observing the succession of events, which inherently pushes us out of the 'now' and into the 'then' or the 'next.' Kubler argues that the instant of actuality is a rupture—a void interval that slips through time. To truly experience it, we must recognize that the present is the only plane upon which our existence is actually projected. Everything else is just a relay of information. The tragedy of human consciousness is that our ability to remember and predict—the very things that make us intelligent—is what prevents us from being fully present.

Actuality is the eye of the storm: a diamond through which the possibilities of the present are drawn into the past.

There is a strange economy to the present moment. To live in actuality is to minimize possibilities. While the future is a sprawling map of infinite choices, the present moment only allows for one action. When we choose, we collapse the wave of possibility into a single, unchangeable event. This is why the present can feel so heavy and so fleeting. It is the point where the 'could be' becomes the 'was.' We spend so much time trying to maximize our experiences that we often bypass the very moments that constitute our lives.

Characteristics of Actuality:

It is the void between events
It is the instant of choice
It is non-signal; it is the gap where signals land
It is the only direct experience we possess

If the present is so elusive, how do we reclaim it? Kubler suggests that art might be our most effective tool. Art does not merely record signals; it creates a laboratory for the possible. It allows us to pause the relentless flow of events and examine the structure of the moment. By engaging with art, we are not just looking at a representation of the past, but participating in a way of seeing that respects the weight and the emptiness of the now. We find ourselves not in the noise of the signal, but in the quietude of the actual.

Key Takeaway

True presence is found not in the accumulation of experiences, but in the recognition of the singular moment of choice.

03 — Simon Willison

The Trust Deficit

Why marketing cannot fix a crisis of credibility

By Simon Willison · 5 min read

Editor's note: A blunt assessment of why AI companies are failing to win over the public.

There is a growing sentiment that the public is turning against artificial intelligence. While many industry leaders point to bad PR or poor messaging as the culprit, the reality is much more systemic. As Dario Amodei, CEO of Anthropic, suggests, we are not facing a communication problem; we are facing a crisis of trust. For decades, the relationship between the tech industry, governments, and the public has been defined by suspicion. People do not believe that the companies building these powerful tools have their best interests at heart. They suspect that the benefits will be concentrated at the top while the risks are distributed among the many.

The Failure of Optimism

The standard industry response to skepticism is to launch glitzy marketing campaigns that promise a utopian future. They speak of AI curing cancer, solving climate change, and ushering in an era of unprecedented prosperity. To a skeptical public, these claims feel like clichés. They sound like the same empty promises that have been made by every major industry for a century. When a company says its technology will save the world, the immediate reaction is not inspiration, but a search for the catch. The gap between the promise and the reality is where trust goes to die.

Saying that AI will cure cancer is more a cliché than it is inspiring. The thing that will work is actually curing cancer.

The criticism that AI companies have failed to deliver on their promises is entirely valid. We have seen massive investment and incredible technical leaps, but the tangible, widespread benefits to the average person remain elusive. The industry has focused on the capability of the models rather than the utility of the outcomes. Until the technology moves from being a novelty or a tool for the elite to being a genuine driver of human well-being, no amount of clever messaging will bridge the gap. Trust is not built through words; it is built through verifiable, useful results.

The components of the trust crisis:

Decades of institutional skepticism
The gap between hype and utility
The perceived centralization of power
The failure to deliver on social promises

If AI companies want to win back the public, they need to stop talking about what the technology *could* do and start showing what it *is* doing. This requires a shift from marketing to accountability. It means focusing on solving real-world problems that people actually care about, rather than chasing the next technical benchmark. The era of 'AI confidence theater' must end if the industry is to survive its own hype. Credibility is earned in the trenches of application, not in the polish of a press release.

Key Takeaway

Trust is a product of utility, not a product of persuasion.

04 — The Marginalian

The Empathy Reflex

Oliver Sacks and the choice to see humanity

By Maria Popova · 11 min read

Editor's note: A reflection on the neurological and moral capacity to choose compassion over retaliation.

In his memoir *On the Move*, the neurologist Oliver Sacks recounts a series of encounters that reveal the fundamental tension between our primal instincts and our capacity for empathy. Sacks describes being targeted by aggressive drivers—men who used their vehicles as weapons to intimidate and endanger him. His initial reaction was a raw, human fury. He felt the urge to retaliate, to strike back at the perceived injustice of the attack. This is the biological baseline: a defensive, aggressive response to a threat. But Sacks's life work was dedicated to understanding the complexities of the human mind, and his personal life reflected that same commitment to looking deeper than the surface of behavior.

The Moment of Recognition

In one instance, Sacks pursued an aggressive driver into a cul-de-sac, armed with a monopod and ready to strike. He was driven by a sense of righteous indignation. However, the moment he reached the car, the narrative shifted. He did not find a monster or a predator; he found four terrified teenagers. In that instant, his rage evaporated, replaced by a profound sense of their vulnerability. The sight of their fear broke his momentum. He realized that his 'enemy' was simply a group of young people caught in a moment of foolishness and terror. The transition from the desire for vengeance to the impulse of compassion was instantaneous and total.

When I saw their youth, their helplessness, their fear, my fist opened and the monopod fell out of my hand.

This is the essence of empathy: the ability to step outside of one's own threatened ego and recognize the shared humanity in another. It is not a passive feeling, but an active, often difficult, cognitive and emotional choice. It requires us to resist the urge to categorize others as 'enemies' or 'others' and instead to see the underlying vulnerability that drives much of human conflict. Sacks's experience suggests that empathy is not the absence of anger, but the ability to let empathy override anger at the critical moment of contact.

The mechanics of the empathy shift:

The initial trigger: perceived threat and rage
The confrontation: the physical meeting of the 'other'
The observation: seeing the vulnerability instead of the intent
The resolution: the collapse of the defensive stance

Sacks's life was a testament to the idea that understanding the brain does not diminish the soul; rather, it provides the tools to navigate its most difficult impulses. By studying the neurological basis of behavior, he gained a deeper appreciation for the fragility of the human condition. Empathy, then, is both a biological capacity and a moral discipline. It is the choice to see the person behind the action, a choice that is most necessary when we are most inclined to strike.

Key Takeaway

Empathy is the ability to see the vulnerability behind the aggression.

05 — Simon Willison

The Hallucination Hack

Using generative error as a creative tool

By Simon Willison · 4 min read

Editor's note: A technical insight into how 'hallucinations' can actually solve classification problems.

In the world of Large Language Models (LLMs), 'hallucination' is usually a term of derision. It refers to the moment a model confidently asserts a falsehood or invents a fact that does not exist. For developers building classification systems, hallucinations are the enemy of precision. If you ask a model to categorize a product into one of 2,000 existing tags, the model often struggles. It tries too hard to fit the data into the rigid, predefined boxes you have provided, often resulting in errors or a refusal to categorize. But there is a way to turn this weakness into a strength.

Breaking the Constraints

The problem with traditional classification is the constraint of the existing vocabulary. When a model is forced to choose from a list, it loses its generative power. A more effective approach, as explored by Doug Turnbull, is to let the model 'hallucinate' first. Instead of providing a list of tags, you ask the model to invent new, descriptive categories that best fit the content. You encourage it to imagine what the classification *should* look like, rather than what it *is* allowed to be. This allows the model to use its full linguistic capability to describe the essence of the item.

Don't classify. Hallucinate!

Once the model has generated these novel, 'hallucinated' tags, the second step is to map them back to reality. This is done using vector embeddings. By comparing the semantic meaning of the model's invented tags to the actual list of existing tags, you can find the closest matches. This hybrid approach combines the creative, descriptive power of generative AI with the structural rigor of a controlled taxonomy. You are essentially using the model's ability to imagine to bridge the gap between raw data and structured information.

The 'Hallucination Hack' workflow:

Prompt the model to invent novel categories without a predefined list
Use descriptive language to guide the 'shape' of the tags
Generate these 'hallucinated' tags for the input data
Use vector embeddings to map invented tags to the real taxonomy

This technique represents a broader shift in how we work with AI. We are moving away from treating models as rigid databases and toward treating them as reasoning engines. The goal is not to suppress the model's tendency to wander, but to direct that wandering toward a useful end. When we stop fighting the model's inherent nature and start building systems that incorporate its strengths, we find solutions that were previously impossible.

Key Takeaway

Generative error can be a bridge to better structured data if you use it to expand possibilities before applying constraints.

06 — Simon Willison

The Browser-Based API

Democratizing access to local and remote models

By Simon Willison · 3 min read

Editor's note: A look at a new tool that simplifies testing various AI endpoints directly in the browser.

The fragmentation of the AI ecosystem is a significant hurdle for developers. Between proprietary APIs like OpenAI's and local setups like LM Studio, testing different models and reasoning settings often requires complex, multi-step environments. Simon Willison's 'CORS Chat' addresses this friction by providing a lightweight, browser-based interface that can communicate directly with any OpenAI-compatible API endpoint, provided it supports CORS headers. This allows for rapid prototyping and testing without the need for heavy local infrastructure or complex backend configurations.

Testing at the Edge

The utility of such a tool lies in its simplicity. It allows a developer to quickly switch between a massive, cloud-hosted model and a smaller, local model running on their own hardware. This is essential for understanding the trade-offs between latency, cost, and intelligence. By managing multiple chat sessions and custom headers within a single web UI, the developer can observe how different models handle the same prompt in real-time. It turns the browser into a laboratory for model comparison.

A web UI for exercising an OpenAI-Responses-compatible chat endpoint.

One of the more interesting technical details is the tool's ability to handle streaming content. As the model generates tokens, the UI updates progressively. This includes the ability to detect and render SVG images as they are being streamed, providing a visual feedback loop that is crucial for testing multimodal capabilities. The ability to persist conversations locally in the browser and export them as JSON further enhances the tool's utility for debugging and documentation.

Key features of CORS Chat:

Direct browser-to-API communication via CORS
Support for custom headers and multiple sessions
Local conversation persistence
Progressive rendering of streaming SVG images

While it may seem like a niche utility, tools like this are vital for the democratization of AI development. They lower the barrier to entry, allowing anyone with a browser and an API key to experiment with the cutting edge. As the ecosystem continues to expand with more open-source and local-first models, the ability to seamlessly jump between these different modes of operation will become a standard requirement for any serious practitioner.

Key Takeaway

Reducing the friction between testing and execution is the fastest way to accelerate AI experimentation.