$46 billion. Ben Horowitz, co-founder of a16z: Don't aim for large-scale in AI just yet. First, get these things right.

a16z manages $46 billion in assets and is one of the most active AI investment institutions globally.

Behind companies like OpenAI, Databricks, Figma, and Cursor that are changing the world, there's always its presence.

However, many people only know the name of this company but are unaware of its unique investment logic.

Ben Horowitz, the co - founder of this fund, started as a product manager at Netscape and is the author of the Silicon Valley entrepreneurial bible "The Hard Thing About Hard Things". He focuses on the psychological building of CEOs.

On September 11, 2025, during this 90 - minute in - depth conversation, he didn't discuss model parameters, nor did he shout about making quick money with AI. What he talked about were "how to prevent early - stage AI companies from failing", "how CEOs can build judgment", and "how top investors assess a person's potential ceiling".

After organizing this conversation, we extracted his core judgments for AI entrepreneurs:

It's not about strategic planning, but about real - world inflection points; it's not about choosing a path, but about avoiding going astray.

In the next round of competition among AI companies, what matters is not the growth story, but:

Whether you've done these things right first.

Thing One: Make a decision first. Don't wait around for a better option

The biggest problem for a CEO isn't a lack of intelligence.

It's that you know exactly where the problem lies but hesitate to take action to fix it.

Ben Horowitz shared a personal decision - making experience: When his company was only 18 months old and its revenue in the past 12 months was less than $2 million, he made a decision that everyone thought was crazy - to go public directly.

He said:

"There's no doubt that this was a bad idea. But the other outcome would have been even worse."

The media mocked him. The Wall Street Journal wrote a long article saying he was immature, and the cover of Business Week was simply titled "The Hellish IPO". But if he hadn't made this decision, the company would have gone bankrupt.

"I knew the media would scold me, and I knew I wasn't ready. But at that moment, if I delayed any longer, it wouldn't be a matter of going public too early; it would be that the company couldn't survive."

Ben summed up such moments in one sentence:

You have to stare into the abyss and then say: That side is a little better. Let's go that way.

This is his definition of a CEO. It's not about giving orders but about having the courage to take that step.

✅ Sometimes, a CEO's task is to choose a path with no good answers

Ben said that most people look for the right option when making decisions. But in actual entrepreneurship, you often face two bad choices.

If you change the product architecture, it might delay for half a year and make financing more difficult;

If you don't change it, the product direction is wrong, and the cost will increase as time goes on.

Which path to choose? Neither has a good result. The worst thing is to be stuck in the middle, making no changes and just waiting to see.

This is the most dangerous behavior for a CEO: hesitation.

Ben said that many failed CEOs didn't fail because of wrong decisions but because they didn't make decisions. The longer you hesitate, the more uneasy the whole company will become. The team will start to secretly speculate: Does the boss already see the problem but just doesn't want to admit it?

If you're not brave enough to make a decision, your subordinates will make decisions on their own.

But they don't have the CEO's overall view, and this will only make the situation worse.

✅ So why do people hesitate? Because they're afraid of being disliked

He gave a common example:

The CEO wants to replace an under - performing manager, but that person is an old employee;

He's not a bad person, but he no longer fits the current rhythm of the company;

But just thinking about having a conversation with him makes the CEO hesitate, worrying about the company atmosphere and what others will think.

Ben's advice is: Don't wait. Go have the conversation.

But don't say "You're not suitable". Instead, speak with facts:

"You're a good engineer, but if you only communicate with the engineering department and can't get the finance and operations departments to cooperate with you, you don't qualify as a CTO. If you're willing to learn, I can help you. If you don't want to change, then I have to replace you."

He said that some CEOs aren't unaware of the problem but don't know how to bring it up. But the longer they delay, the bigger the problem will become.

His advice to early - stage CEOs is just one sentence:

When you see a trend, move forward. Don't wait for everyone to agree with you.

Because if you wait for everyone to agree, then you, as a CEO, aren't really playing a role. The real value is that you take action when others are still hesitating.

Here, Ben Horowitz isn't teaching "how to judge how much a company is worth". What he's really talking about is the first thing:

Make a decision first. Don't wait for everyone to agree, and don't wait for a better option to appear.

Technology isn't the biggest challenge; procrastinating on making a decision is.

Early - stage AI companies are most afraid not of making mistakes but of not taking action.

Thing Two: Hire people who can take action, not those who need to be trained

The second most common mistake startups make isn't unclear direction, insufficient funds, or an inability to understand the market. It's hiring a group of people who can't create value.

Ben Horowitz shared a real - life scenario where a CEO asked him for help:

"Ben, my CTO is a jerk."

"Do you want to fire him?"

"No... He's very good at technology, and I don't want to fire him."

"So you want me to teach you how to communicate with a jerk and keep him from quitting, right?"

"Yes."

This is a problem many early - stage founders face: how to manage difficult - to - get - along - with people.

But Ben believes the real question is: Can this person help the company move forward?

✅ The key isn't whether the person is nice, but the management leverage

Ben explained this problem clearly: As a CEO, you can't push the company forward on your own.

If your team members only start to act when you're constantly watching, urging, and giving advice, you're not building a team; you're being a "temporary team leader".

A real manager is someone who can mobilize the team and make the organization move forward on its own. When you say one thing, they can do three; even if you don't say anything, they know where to go.

This is the kind of person you really need.

Ben calls this standard "managerial leverage".

It means that when you hire this person, it's not to help them grow or train them. Instead, after they join, the company as a whole becomes "lighter", the direction becomes clearer, and the pace becomes faster.

If there's no leverage, the company will be like you're pulling it alone, and everyone will wait for your instructions.

What kind of person has "managerial leverage"? Ben used the example of Databricks. There was an engineer named Ali in the team who had never been a CEO, but Ben quickly found that although this person was low - key, he had real leverage ability.

He never waited for instructions. Instead, he actively proposed why to do something, how to execute it, and how the progress was. Moreover, he not only had strong technical skills but also understood customers, promoted business, and facilitated the collaboration of different teams.

Ben immediately called Ali over and said to him:

"You're no longer the VP of engineering. You're the CEO now because you can mobilize the whole organization. You have leverage."

After Ali took over as CEO, the pace of Databricks completely changed.

Ben recalled: "When I invested in them at the beginning, I never expected that there was a natural - born CEO hidden within the company."

✅ People who need your constant guidance aren't suitable to lead a team

Ben said that many early - stage founders have a misunderstanding: They think leading a team means helping everyone grow.

This is correct, but it's not suitable for the early stage.

He was very specific:

"You're not the VP of engineering, not someone who teaches everyone how to write code."

"You're the CEO, and your responsibility isn't to train people; it's to choose the right people."

"Choose those who already know how to do things."

His view is very straightforward:

"Don't fantasize that you can train an ordinary person into a top - notch manager, especially when you're busy as hell and looking for money everywhere."

Ben's advice to CEOs is very clear: Don't look at whether a person is smart or willing to learn. Look at whether the company becomes clearer and easier after they join. If you find that you have to spend a lot of time reminding and guiding them, you've made a mistake.

He said that those people you think "can be trained" often end up making you worry every day and cleaning up their messes.

What you really need to find are people who can make the team run more smoothly from the start.

In the end, the second thing is to hire ready - to - work people and don't count on training them slowly.

Thing Three: Integrate the product into the process, not just attach an interface

Many AI entrepreneurs start with one goal: find a model, put a shell around it, make an application, and then get financing.

Ben Horowitz directly poured cold water on this idea: This path won't work.

In the conversation, he clearly pointed out:

"There are still people saying that they can build a company by 'putting a layer of GPT on it'. This idea first emerged in the 1980s. At that time, people thought that companies like Salesforce were just simply wrapping a layer around a database. But in fact, truly successful companies have done in - depth business reconstruction, which is far more complex than it seems on the surface."

He emphasized that the problem with "putting a shell" isn't about technology but about the mindset:

If you just connect ChatGPT, users are still copying and pasting.

You don't understand how users work and what rhythm they need.

You haven't really integrated AI into the work process.

What you make at most is a tool, not a real product.

✅ Cursor isn't just an additional interface; it's a reconstruction of the process

Ben gave the example of Cursor, an AI programming tool company.

a16z is the main investor in Cursor. On the surface, it's just an AI programming assistant. But Ben explained:

"There are 14 models running behind Cursor, not just one. They've re - understood how engineers think and write code. It's not about letting the model simply answer questions but about truly integrating AI into the developer's thinking process."

For example:

When you're writing a function, Cursor will automatically guess what you're trying to accomplish;

It can refer to the files, code style, and variable names in your current project;

It will also offer modification suggestions, generate comments, and even help you plan the next step.

This isn't just adding an AI input box; it's about integrating AI deeply into the entire work process. What Cursor does isn't just enhancing search; it's thinking ahead for you.

Cursor demonstrates in - depth technical integration in the development scenario. But in enterprise - level applications, integrating into the process has another dimension: understanding business logic. You can't solve the problem of business definition just by relying on a model. Ben gave a very specific example:

"You think the definition of a customer is very simple, right?"

"But in an enterprise, a customer could be a person, a company, a department, or a certain contract structure."

"If you ask 10 enterprises, each will have a different definition of 'customer'."

And the AI application you're developing must understand these details and match the organizational structure and business process in the real - world scenario.

✅ Databricks' success relies on integration ability, not the model

Ben invested in Databricks and witnessed its entire process of growing from a big - data processing technology team to an AI data platform.

The core of Databricks isn't their model. It's their ability to integrate the originally scattered things like enterprise data, permissions, processes, and analysis results so that they can function properly.

That is to say, the real value isn't what model you use but whether you can handle the real business of the enterprise.

And these tasks usually can't be clearly described in a single prompt.

They involve permissions, historical data, context, project status, and delivery rhythm. If you don't understand these, it's very difficult to develop a usable AI application.

Ben said that this is the core reason he values companies like Cursor and Databricks:

"They don't win by relying on a single model. Instead, they understand which model to use in what scenario."

Not all AI applications need to develop their own models, but all applications have to answer one question:

Is what you're doing really what the users need?

Stop simply putting a shell on it. If you want to do it, integrate deeply into the business, execute efficiently, and deliver stably.

The third thing is to make the product close to real - world tasks, not just wrap it around the model.

Thing Four: Look at what a person can achieve, not the mistakes they've made

Ben Horowitz said that when he invests, he doesn't look at how many mistakes a person has made but at whether they can achieve what others can't.

This statement is most clearly reflected in the case of Adam Neumann.

Everyone knows about the failure of WeWork. The media coverage was overwhelming, with documentaries, TV shows, and critical articles... It once became a representative of the collapse of an entrepreneurial myth.

But Ben invested in his new company, Flow.

He said that this was the most criticized investment he'd ever made.

"Some people said we were stupid, and some said we were sexist and racist. It's quite rare for an investment to be labeled with so many things."

However, he believes that Flow will become one of our most successful investments.

Why?

Ben's answer is:

"We shouldn't judge a person's ability to succeed based on their worst moment. You should look at the best part of what they've done and see if others can match it."

Looking back at the process of Adam founding WeWork, his abilities in brand building, community creation, and space design aren't what ordinary founders can achieve. The problem was the loss of control in later - stage management and decision - making, but this doesn't deny his early - stage product creativity.

Ben's investment logic is that you can't ignore the fast - paced part of a person's journey just because they've stumbled.

✅ This way of evaluating people isn't the first time Ben has used it

He also gave another extreme but real example: Shaka Senghor.

This person committed a serious crime in Detroit and served 19 years in prison, 7 of which were in solitary confinement. After getting out of prison, he became a writer and a mentor and received attention from mainstream media, becoming a typical case of life transformation.

Ben is very familiar with him and has written about him in his book.