Claude Code's founder's 7 key insights at the Sequoia Conference

Summary: A Ying

Boris Cherny, founder of Claude Code, shared at the Sequoia Conference, and the information was extremely dense. Many viewpoints I heard for the first time in their entirety. This guy indeed has a pretty accurate understanding of AI.

I’ll share my own summary.

01 Code is no longer scarce

For many mainstream development scenarios, manually writing code has begun to become an inefficient task.

In the past, delivering a feature, engineers would sit down, think through how to implement it, then type out the code line by line. In this process, the engineer’s greatest value was: whether they could write, how well they could write, and how quickly they could write.

Now, the way of working has changed.

The same feature, engineers’ work is more like: first clarify the requirements, break it into several parts assigned to an Agent, set acceptance criteria, then check whether the results produced by the Agent are correct; if not, adjust prompts and run it again.

AI can now handle most coding tasks. Of course, not 100%, there are still many large and complex codebases, obscure languages, or special environments where current models still underperform.

Overall, the value of engineers has shifted from: can you write code? to: can you break down tasks? can you clearly communicate goals? can you verify results? can you manage Agents?

This change is very much like the Industrial Revolution.

Before the Industrial Revolution, a blacksmith did everything—from forging, polishing, to assembly—all by himself. Skilled blacksmiths naturally commanded higher value.

Later, assembly lines appeared. Each worker was responsible for a single process, and overall output increased by dozens or hundreds of times compared to manual work.

At this point, the valuable role in factories was no longer the craftsman who was best at a specific process, but the person who could design, manage, and keep the assembly line running smoothly.

Workers didn’t disappear, but their roles changed.

Software engineering is now experiencing a similar shift. Code itself is no longer a scarce resource. The ability to write code is becoming a basic skill, like using PowerPoint.

What is truly scarce is the ability to break vague requirements into clear tasks, select the best solution from several options provided by an Agent, and coordinate a group of AIs to complete a task.

Many veteran engineers initially find this hard to accept. Writing code by hand has been the reason many have loved this profession for decades.

Handing this over to machines is not just a change in work style for many, but a reshaping of identity.

But trends are trends.

02 Like Gutenberg’s printing press

Coding is transforming from a specialized skill into a fundamental ability. This can be compared to the invention of the printing press in 15th-century Europe.

Before the printing press, only about 10% of Europeans were literate. These people were often employed by illiterate nobles, specializing in reading and writing for others.

Then the printing press was invented. In 50 years, the number of books published in Europe exceeded the total of the previous thousand years, and book prices dropped by about 100 times. After a few centuries of supporting infrastructure (education systems, economic structures), the global literacy rate rose to today’s 70%.

Boris believes that AI’s impact on software is an accelerated version of the printing revolution. Software will be fully democratized within decades, becoming something anyone can master.

Eventually, coding will become as natural as sending a text message.

03 What skills are most important?

When the barrier to writing code is lowered to the extreme by AI, what truly distinguishes a person’s ability is their product sense, their deep understanding of a specific field.

For example, two people are tasked with creating a product for doctors. One is a fast coder, the other has worked in hospital IT for several years.

In the past, the coder was more likely to succeed because they could implement the idea.

Now, anyone can implement an idea. At this point, the person who truly understands hospital workflows is more valuable. Because they know which features doctors will actually use, and which are just plausible-sounding.

In other words, once AI levels the playing field for execution, the gap in judgment becomes more prominent.

This directly redefines the meaning of “generalist.”

In the past, a generalist usually meant an engineer who could do iOS, web, and backend development—essentially a full-stack within engineering.

The future generalist is interdisciplinary full-stack.

Some understand product, design, and engineering. Others understand product, data science, and engineering. Such combinations were nearly impossible before, as each required long-term specialized training.

But now, AI lowers the execution barriers for each, allowing a person to span multiple fields while maintaining depth.

The Claude Code team is like this. Engineering managers, PMs, designers, data scientists, finance, user research—all writing code.

Designers can run interaction prototypes themselves to show the team, no longer just creating mockups for engineers to implement.

Finance can build analysis tools on their own, running complex financial models without waiting for BI. User research colleagues start analyzing data themselves, taking over tasks that previously required data team support.

Everyone’s expertise remains deep. But with AI assistance, coding has become a shared language.

04 The moat of SaaS is crumbling

Over the past decade, several axioms in the SaaS industry have been widely accepted.

The first is switching costs. Once a company adopts your system, it accumulates years or even decades of data, configurations, fields, and permissions.

Migrating to another system involves exporting and importing all this data, which is often so painful that companies hesitate.

The second is workflow lock-in. Employee operations, cross-department collaboration, approval processes—all are built around this SaaS.

Switching systems isn’t just about data transfer; it’s about rebuilding the muscle memory developed over years.

Together, these form the deepest moat for SaaS. But with sufficiently powerful models, the logic begins to change.

Regarding switching costs: previously, moving from one SaaS to another required months of engineering work to align fields and replicate data structures.

Now, simply feed the interfaces and data structures of both sides into a model, let it figure out the mappings, and optimize step by step. What used to take months might be done in days.

Regarding workflow lock-in: more interesting. The reason workflows lock in customers is because these processes are complex, implicit, and human-dependent.

The tacit understanding in employees’ minds about who approves whom, when to escalate, can’t be directly transferred.

But models like Opus 4.7 excel at understanding, dissecting, and reconstructing complex workflows in new environments. Sometimes, the reconstructed version might even be better than the original.

Thus, the moat built on data and process accumulation is dissolving.

For SaaS providers, this might be bad news. But for SaaS users and teams building the next-gen SaaS, it’s a real opportunity window.

05 The best era for entrepreneurs

In the next 10 years, truly industry-disrupting startups could be ten times more numerous than in the past decade.

The reason is simple.

Small teams can produce products on par with or better than large corporations using AI. Conversely, large companies trying to fully leverage AI are actually at a disadvantage.

Why?

A company with over ten years of history has developed a complete set of business processes, roles, collaboration habits, training systems, and KPIs. These are assets and barriers in the past.

But integrating AI means rethinking all that: restructuring workflows, retraining staff, facing internal resistance, coordinating multiple departments and approvals.

A three-person startup from day one treats AI as a default foundation. No legacy baggage to dismantle, no habits to change, no need to persuade anyone. Clarify the plan today, run a demo tomorrow, launch for users the day after.

This speed difference has always existed for startups. They have speed advantages over big companies. But AI amplifies this gap many times over.

Why?

Because the stronger the AI, the greater the leverage a single person can wield in the same amount of time. A small team mastering AI today might produce output equivalent to ten people yesterday; tomorrow, it could match thirty.

But the organizational weight of big companies doesn’t lighten; it actually becomes heavier because they need to digest AI. The more powerful AI is, the larger the acceleration gap between small teams and big organizations.

This is what Boris calls negative assets. It’s not that big companies lack money, people, or willingness—it’s that their past profit engines are now stuck on the path where AI can unlock value.

06 MCP will not die

MCP will not die.

After Skill became popular, many thought MCP was no longer needed. The founder of OpenClaw shares this view.

But Boris disagrees. He believes MCP will become the software connectivity layer in the AI era.

In the past, internet software connected via APIs.

But the core issue with APIs is that they are designed for engineers. To use an API, you need to read documentation, apply for tokens, write code, align fields, handle exceptions. Basically, APIs are built for human developers.

MCP is different. It allows models to connect directly, understanding and calling without a programmer translating for them.

So Boris calls API the Human Developer Interface, and MCP the Model Interface Protocol. One is for humans, the other for models.

This is very similar to the past. In the mobile internet era, all services were expected to be API-enabled. In the AI era, all services will be MCP-enabled.

07 Computer Use remains important

Many now argue that Computer Use might not work anymore.

The reasons are reasonable: it consumes tokens, runs slowly, and is unstable. It looks more like a flashy demo, still far from practical use.

But Boris sees a different layer.

He emphasizes that Computer Use solves a major pain point in AI deployment: in the real world, many systems have no APIs or MCPs.

Especially in enterprise environments.

Once inside a company, you see that many core systems are very old. ERP, OA, financial systems, internal approvals, supply chain backends, custom systems. Many lack open interfaces, documentation, or automation. They sit there, manually operated by countless employees every day.

Why not just build APIs for them?

Because it’s often impossible. The vendors who built these systems may no longer exist. IT departments lack the motivation or budget to rebuild.

Business units can’t afford to pause for half a year or a year. These systems will never wait for a perfect API to save themselves.

In the short term, major models will still focus on enhancing their own Computer Use capabilities.