The Deep Feed

01 — Lenny's Newsletter

The Corporate Brain: How Stripe Built Kai

Moving beyond off-the-shelf AI to enterprise-scale intelligence

By Sharadh Krishnamurthy · 12 min read

Editor's note: A masterclass in why buying software is often a trap for large-scale intelligence.

Most companies approach AI as a procurement problem. They look at the market, see a dozen promising tools, and attempt to stitch them together into a makeshift nervous system. Stripe took a different path. Instead of buying a suite of tools that might never truly understand the specific, messy reality of their data, they built Kai. Kai is not just a chatbot; it is an internal agent used by over 10,000 employees. It is a system designed to act as a central intelligence layer, capable of querying data, building dashboards, and managing workflows across the entire organisation. This wasn't about chasing a trend; it was about building the infrastructure necessary for an AI-driven enterprise.

The Governance of Intelligence

The primary challenge of enterprise AI is not intelligence itself, but control. When you give an agent access to your company's data, you are essentially giving it a key to the kingdom. Stripe solved this by using 'projects' as a governance mechanism. A project is not merely a folder where files live; it is a security boundary. It defines what an agent can see, what it can touch, and what it can change. This structure allows employees to build bespoke agents for specific tasks without risking the integrity of the entire system. It turns the chaos of individual AI experimentation into a structured, searchable library of company skills.

Projects at Stripe are a governance mechanism, not just a folder.

This approach also addresses the 'rogue agent' problem. In any sufficiently complex system, agents will eventually attempt actions that conflict with safety protocols or production stability. Stripe engineers had to learn the hard way when agents nearly disrupted production systems. To prevent this, they implemented load shedding and agentic identity. Every action an agent takes is tied to a specific identity, allowing for clear auditing and the ability to cut off access instantly if an agent begins to behave erratically. It is a systems-level approach to what many treat as a mere software feature.

The Pillars of Enterprise AI

Governance through project-based boundaries
Secure sandboxes for agentic execution
A skills platform for packaging human workflows
Telemetry and evaluation for agent quality

Ultimately, Stripe's success with Kai stems from treating AI as an infrastructure problem rather than an application problem. They didn't just build a tool; they built a way for the company to learn and act. By focusing on the data layer and the skills platform, they have created a system where the intelligence grows as more employees package their expertise into reusable agentic workflows. It is a shift from using AI to assist humans to using AI to scale the collective intelligence of the entire firm.

Key Takeaway

Enterprise AI succeeds when it is built as a governed infrastructure layer, not as a collection of disconnected tools.

02 — Lenny's Newsletter

The Agentic Migration

Why simplicity wins the battle for personal productivity

By Claire Vo · 10 min read

Editor's note: An analysis of the transition from complex, high-maintenance agent stacks to streamlined, reliable ones.

The promise of AI agents is autonomy, but the reality is often a maintenance nightmare. For power users, the goal is to deploy a fleet of specialists—bots that manage inboxes, monitor compliance, or coordinate family logistics. However, as these systems grow, the complexity of managing them can outweigh the time they save. Claire Vo's recent migration from OpenClaw to Grok Bot serves as a case study in this tension. The move was driven by a single, brutal truth: a tool that is too difficult to maintain is a tool that eventually fails to be useful.

Treating Bots Like New Hires

A successful agent strategy requires a shift in mindset. Instead of viewing bots as software scripts, one should view them as new hires. This means giving each agent a specific name, a clear job description, and a defined scope. When Vo creates an agent named 'Prody McProd', the system already has a semantic understanding of its role. This prevents the common pitfall of building 'generalist' agents that eventually become unfocused and unreliable. Specialists are easier to audit, easier to fix, and easier to replace.

Thinking of each bot as a new hire makes agent design much easier.

Autonomy must be balanced with human oversight through approval gates. For high-stakes tasks, such as issuing customer refunds, an agent should not act unilaterally. Instead, it should prepare the action and present it to the human for a single-click confirmation. This allows the agent to handle the heavy lifting of investigation and preparation while leaving the final decision to the person responsible. It is the difference between an agent that works for you and an agent that works instead of you.

Principles of Agent Design

Specialization over generalisation
Approval gates for high-consequence actions
Explicit training on personal voice and style
Proactive scheduling via routines

The real moat for agent platforms is likely not the underlying model, but the user experience and the ability to handle multi-account complexity. For an entrepreneur running multiple businesses, an agent that can bridge six Gmail accounts and several Slack workspaces is transformative. Without this capability, agents remain novelties. With it, they become the connective tissue of a modern professional life.

Key Takeaway

Effective agent deployment relies on treating AI as a team of specialised hires rather than a single, all-purpose tool.

03 — Lenny's Newsletter

The Four-Week Sprint

How SpaceXAI built Grok Bot from scratch

By Roman Ugarte · 8 min read

Editor's note: A look at the velocity required to win in the hyper-competitive AI agent market.

In the current AI race, speed is the only real advantage. Roman Ugarte and his team at SpaceXAI took Grok Bot from the first line of code to a working internal product in just four weeks. This wasn't just a feat of engineering; it was a feat of product focus. The decision to build Grok Bot as a standalone entity, rather than an add-on to existing tools like Cursor, allowed the team to define the user experience from the ground up. They weren't trying to improve an existing workflow; they were trying to create a new one.

The Colleague Philosophy

The success of Grok Bot rests on a specific product philosophy: the 'colleague-pilled' approach. Most AI tools aim to be assistants—tools you use to get things done. Grok Bot aims to be a colleague—a presence that exists alongside you, capable of handling tasks with a level of agency that feels human. This distinction is subtle but massive. It changes the user's relationship with the technology from one of command-and-control to one of collaboration.

An AI that does 100% of the job feels categorically different from one that gets you 90% there.

To achieve this, the team focused on high talent density and extreme user feedback loops. In the early days, the team personally onboarded nearly 300 users. This wasn't a scaling failure; it was a deliberate strategy to understand the friction points of agentic work. By being in the trenches with their first users, they could iterate on the product's ability to handle real-world complexity, rather than just theoretical use cases.

Lessons from the Grok Sprint

Build from scratch to control the UX
Prioritise agency over assistance
Use high-touch onboarding to find friction
Focus on the 100% completion threshold

The competitive landscape for AI agents is brutal. To win, a product cannot just be 'good'; it must be fundamentally different in how it integrates into the user's life. Grok Bot's ability to move beyond simple prompts into autonomous routines is what separates it from the sea of wrappers. It is a bet on the idea that the future of work is not about using AI, but about working with it.

Key Takeaway

Winning in AI requires moving beyond 'assistance' toward true 'agency' through rapid, high-touch iteration.

04 — Cal Newport

The Radicalism of Depth

Wendell Berry and the rejection of the digital edifice

By Study Hacks · 9 min read

Editor's note: A reflection on why choosing 'hardness' and 'place' is a necessary rebellion against modern convenience.

Wendell Berry died at 92, having spent much of his life in a cabin in Kentucky that lacked electricity. To the modern observer, this might look like a retreat into obsolescence. To Berry, it was a move toward reality. After a promising academic career at institutions like NYU, Berry made a decision that his peers viewed as intellectual suicide: he moved back to his family farm. He traded the cultural currents of the metropolis for the soil of Henry County. He chose a life of friction over a life of convenience.

The Trap of Digital Numbness

Our current era is defined by a drive to eliminate cognitive strain. We use AI to bypass thinking and algorithms to bypass feeling. Our professional lives have become a series of digital echoes—Slack messages about emails about Zoom meetings—that result in nothing of substance. This creates a state of numbness. We are constantly connected but rarely present. Berry understood that this avoidance of strain is not progress; it is a form of erosion.

Hardness and strain are not enemies to evade; when oriented toward an intentional vision, they yield a foundation.

Berry’s life suggests that depth is found in the things that cannot be automated or outsourced. It is found in the relationship between a person and their place, in the physical reality of food production, and in the slow, deliberate work of writing with pencil and paper. By rejecting the digital edifice, he didn't just escape the noise; he reclaimed the ability to be shaped by his environment. He allowed the landscape to check him, to ensure his work was grounded in truth rather than abstraction.

The Components of a Deep Life

Intentionality of place
Acceptance of cognitive strain
Rejection of algorithmic mediation
Connection to physical reality

The lesson of Berry’s 92 years is not that we must all become farmers, but that we must all find something worth the struggle. In a world that offers endless, easy distractions, the most radical act is to choose a life of purpose, even if that purpose requires us to walk through the mud.

Key Takeaway

Meaning is found in the friction of real life, not in the seamlessness of digital convenience.

05 — Aeon

The Geometry of the Hand

Where medicine and palmistry intersect

By Alison Bashford · 11 min read

Editor's note: An exploration of the historical link between the 'occult' study of hands and clinical medicine.

We often think of palmistry as a relic of superstition, something relegated to carnivals and magic shops. Yet, the history of the human hand tells a different story—one where the lines on a palm were once seen as legitimate medical data. For centuries, the distinction between 'occult' knowledge and clinical observation was far thinner than it is today. Physicians did not just study symptoms; they studied the geometry of the body, looking for signs of the past and the future written in skin and crease.

The Simian Line

Consider the 'simian line'—a single transverse crease that runs across the palm. While most humans have separate lines for the head and heart, a small percentage share this singular crease, a trait also seen in other primates. For decades, this was a key element in clinical diagnosis, particularly in identifying chromosomal anomalies like trisomy 21. The line was not a mystical omen, but a biological marker. The history of medicine is, in many ways, a history of learning to read these physical signs with precision.

The history of palmistry is the history of medicine.

Even Isaac Newton, the architect of modern physics, was drawn to this intersection. In 1663, he purchased a book on palmistry that instructed the reader to use a measuring compass to decode the angles and shapes of the hand. To Newton, the hand was a mathematical problem to be solved. He was fascinated by the idea that hidden truths—what physicians called 'occult' knowledge—could be uncovered through rigorous observation and the study of ancient, learned traditions.

Historical Connections

The use of mathematical tools in chiromancy
The overlap between 'occult' and 'unseen' medical symptoms
The clinical utility of dermatoglyphics (skin ridge patterns)
The transition from divination to prognosis

When we dismiss the history of palmistry, we risk ignoring the long tradition of physicians who sought to find meaning in the most minute physical details. The hand is not just a tool for grasping; it is a record of our biological history. By studying it, we are not looking for fortunes, but for the deep, structural truths of what it means to be a biological entity.

Key Takeaway

The boundary between superstition and science is often a matter of how we choose to interpret physical evidence.

06 — Aeon

The Pale Blue Dot Reimagined

Seeing our significance in the moonlight

By Aeon · 5 min read

Editor's note: A shift in perspective: from our smallness in the cosmos to our growing impact upon it.

For decades, the most powerful images of Earth from space have been those that emphasised our insignificance. Carl Sagan’s 'Pale Blue Dot' and the 'Earthrise' photograph served as cosmic memento mori, reminding us that we are tiny, fragile inhabitants of a speck in a vast, indifferent void. These images were designed to foster humility. They were meant to show us how much we have to lose and how little we actually matter in the grand scale of the universe.

A New Kind of Awe

However, a new set of images is emerging—not from the deep reaches of the Hubble telescope, but from the lunar orbit of the Artemis II mission. These photographs, capturing the nightside of Earth illuminated by moonlight, tell a different story. When combined into timelapses, they reveal a planet that is not just a silent rock, but a dynamic system pulsing with both natural phenomena and human-made light. This is not an image of our smallness, but an image of our presence.

The story it tells is one of our growing significance—for better and for worse.

This shift in perspective is unsettling. While the old images invited us to feel small and unified, these new images highlight our agency and our impact. We see the lights of cities, the patterns of human movement, and the visible scars of our industrial reach. We are no longer just observers of the cosmos; we are active participants in the planetary process. The awe we feel now is not the awe of being tiny, but the awe of being powerful.

Shifts in Cosmic Perspective

From insignificance to agency
From passive observation to active impact
From cosmic humility to planetary responsibility

As we move further into the era of space exploration, our visual relationship with Earth will continue to change. We are moving from the era of the 'Pale Blue Dot' into an era of the 'Active Blue Planet'. This requires a new kind of psychological maturity: the ability to look at our own significance and understand the weight of the responsibility that comes with it.

Key Takeaway

Modern space imagery is shifting our focus from our cosmic insignificance to our profound planetary agency.