Should You Be Token-Maxxing?

Applications for SR007 close in just two weeks! It only takes five minutes. Apply now.

This Week’s Big Ideas 💡

🗽 The Boston and New York Tech Week calendars are live! It’s the biggest East coast Tech Week yet, with 1,500+ events happening across both cities. Boston is from May 26–31 and NYC is right after from June 1–7. Check out the calendar and build your own personalized schedule here.

🇺🇸 Tom Hammer explains how speedrun supports founders from around the world.

🤖 Fareed Mosavat says agent-native applications are coming.

💼 Join our talent network for more opportunities. If we see a fit for you, we’ll intro you to relevant founders in the portfolio

Should you hire two engineers or give one engineer unlimited tokens and nicotine?

The answer to this question is an existential one for startups to answer. Most seed stage startups only have 12-18 months to live. So deciding whether to spend $300,000 a year on agents or using that cash to hire a second engineer is a do or die decision.

So what is a founder to do? Should you be token-maxxing?

The case for going long on tokens

The structural arguments for token-maxxing are that tokens scale linearly with throughput. You can 10x your fleet on a Tuesday afternoon. You cannot 10x your headcount in a day. Additionally, tokens are pure variable cost, with no equity dilution, no demands for green juice and complimentary massages, and no managing someone through a PIP. If the models get better in six months (as they have been) your engineering pipeline gets better.

It is also significantly faster. Founders are building more product in 12 weeks than they used to build in a year. Non-technical founders are shipping enterprise-ready platforms in just a couple of months. The upper bound of what a small team can do has moved, and it has moved fast.

Andrey Starenky at Sentra (SR006) puts a number on it. His team uses Claude Code, Cursor, and OpenCode in whatever combination each engineer prefers, with cloud agents wired into Slack to triage bugs and ship small features the moment a complaint comes in. He pegs the productivity boost at a minimum of 5-10x. The annualized Cursor bill alone, he told us, “is more than the salary of an average engineer.”

The aggressive use cases are even more lopsided. Freddie Wollen from Coalition Systems (SR006) runs agents overnight, attacking his own architecture in search of security flaws. “One workload where spraying tokens at the problem seems like the right answer,” he calls it. The Rust compiler and a strong test suite give the agents a closed feedback loop, so they iterate without supervision while the team sleeps. No human team would do this work, at this scale, for any price.

The point is that the upper bound of what a small team can do has moved, and it has moved fast.

The case for going long on people

Now the other side, just as hard.

When token-maxxing, it can be exceptionally hard to understand what caused the AI to go haywire, and to fix it going forward. With people you can just say “stop doing that or you are fired.”

Simo Rachidi of Safeworld (SR006) builds robotics, where weirdly stupid bugs have physical consequences. He has postponed hiring around internal tools, QA automation, and lightweight full-stack work, but draws a hard line on the roles that carry accountability with tasks like robotics safety, systems architecture, simulation fidelity, and customer trust. “AI gives every engineer more hands,” he says. “It does not replace ownership.”

Token costs are also a moving target, and they have moved in both directions. Frontier model pricing has been reset multiple times in the last two years. A $300K-a-year agent budget in Q1 can become a $500K budget by Q3 if the wrong tier gets repriced. Hiring a person commits you to a salary you can model. Hiring a fleet commits you to a vendor’s roadmap.

The complication

It may be that this is a false binary.

It’s possible founders are not actually foregoing hires because of tokens. Token spend is up. Headcount is also up. Founders are doing both at the same time, or doing neither for unrelated reasons like the funding environment or the difficulty of recruiting in 2026.

Jackie Lunger from Panorama (SR006) told us that her team’s first instinct when agentic tools arrived was to spin up hundreds of agents and let them cook. They coded fast and made the wrong choices. “Coding is actually the easy part,” she says. “The hard part has always been knowing why you are doing what you’re doing.” Now the team invests heavily in planning, and a single agent with clear instructions can one-shot the work. “Spending more tokens usually just means the clarity of thought wasn’t there when the task was started,” Jackie says. “More tokens signal less precision.”

Simo from Safeworld arrives at the same place from a different direction. “I would not call us token-maxxers,” he says. “I would call us failure-mode-maxxers. We use AI aggressively, but the thing we are really maximizing is confidence before deployment.”

If that hypothesis holds, then “token maxing versus hiring” is a false binary, and the real question is something more specific. Which categories of work should be automated, and which categories require a person who can be looked in the eye?

We can all stop asking “tokens or people” and start asking “what kind of work am I actually trying to get done.”

You are receiving this newsletter since you opted in earlier; if you would like to opt out of future newsletters, you can unsubscribe immediately.This newsletter is provided for informational purposes only, and should not be relied upon as legal, business, investment, or tax advice. You should consult your own advisers as to those matters. This newsletter may link to other websites and certain information contained herein has been obtained from third-party sources. While taken from sources believed to be reliable, a16z has not independently verified such information and makes no representations about the enduring accuracy of the information or its appropriateness for a given situation.References to any companies, securities, or digital assets are for illustrative purposes only and do not constitute an investment recommendation or offer to provide investment advisory services. Any references to companies are for illustrative purposes only; please see a16z.com/investments. Furthermore, this content is not directed at nor intended for use by any investors or prospective investors, and may not under any circumstances be relied upon when making a decision to invest in any fund managed by a16z. (An offering to invest in an a16z fund will be made only by the private placement memorandum, subscription agreement, and other relevant documentation of any such fund which should be read in their entirety.) Past performance is not indicative of future results.Charts and graphs provided within are for informational purposes solely and should not be relied upon when making any investment decision. Content in this newsletter speaks only as of the date indicated. Any projections, estimates, forecasts, targets, prospects and/or opinions expressed in these materials are subject to change without notice and may differ or be contrary to opinions expressed by others. Please see a16z.com/disclosures for additional important details.