a16z · August 28, 2026

Ben Horowitz on AI Infrastructure, Agents and the Mythical Man-Month

Ben Horowitz says the mythical man-month no longer protects a lead. Hiring a thousand engineers still wrecks a company, but $3 billion of chips can put a new model in the race. He said it while introducing a16z's Machine Age infrastructure fund, on a panel with partners Martin and Raghu, published on the a16z channel in August 2026.

Speaker
Ben Horowitz
Where
a16z
Published
Length
54 min

Ben Horowitz: Co-founder of Andreessen Horowitz, author of The Hard Thing About Hard Things.

Recording: Why Top Founders Are Racing Into AI Infrastructure

Key points

  1. 1

    Every layer of the stack needs rebuilding

    Horowitz opens with a pattern: each major new technology forces a new infrastructure, and this one reaches further than any before it. His list runs from chips and system software down to power and the copper in the wires. For a founder, the opening sits in the physical stack below the model, built for an older style of computing.

    “not only do we need new chips, new system software, we need new ways of doing power, we need to replace copper”
    Ben Horowitz, a16z, 01:38
  2. 2

    Money now buys back a lead

    The old rule said you could not catch a company with a two-year lead by hiring, because a thousand new engineers wrecks your company. Horowitz says that rule broke. A rival can now spend $3 billion on a chip cluster and join the race, which is how he explains Grok and Kimi appearing from nowhere. Everyone, he says, is still adjusting to money working on almost any problem.

    “But it's not hiring a hundred thousand engineers. It's taking $3 billion and like lighting up a magnificent cluster”
    Ben Horowitz, a16z, 17:31
  3. 3

    Agents are a new type of employee

    Asked how a16z brings agents into the firm, Horowitz treats them as a second workforce with its own learning curve. He lists how they fail: they waste tokens and money, forget things, make things up, behave badly and create security holes. They can also be very productive. The open work is fitting them in beside the people they work with.

    “they can burn a lot of tokens and spend a lot of money and get nothing productive done”
    Ben Horowitz, a16z, 23:21
  4. 4

    Make people superhuman first

    Horowitz does not claim a16z has solved agents. He says the firm has not automated itself and he is not phasing out staff. His target is to make each person superhuman while keeping the bots from wrecking the place.

    “how do we make all our humans superhuman um without like wrecking the place um because the bots got out of control”
    Ben Horowitz, a16z, 24:14

How it compares

How this recording lines up with what Ben said before and with other operators Gavel cites.

Builds on

Horowitz's culture page defines culture as a set of actions, built from rules clear enough to obey or break today. Here he brings the same management problem to agents. a16z spent years learning to work with human employees and now has to learn a second kind. He offers no agent rules yet, so this reads as the start of that work.

Ben Horowitz: Culture Is a Set of Actions

Builds on

YC partners argue that blitzscaling is over and AI replaces headcount as the way to grow. Horowitz agrees that hiring was never how you caught a leader. He adds the other side: at the model layer, money spent on compute now closes gaps that hiring never could.

YC Partners: AI Revolution, What Nobody Else Is Seeing

Agrees with

Jason Lemkin says AI agents are not plug-and-play and need training and steady oversight, enough to justify someone who owns them. Horowitz's list of how agents fail, from wasted tokens to invented facts to security holes, makes the same case from inside a venture firm.

Source: Jason Lemkin on managing AI agents, Gavel corpus

Related

More talks

All talks