LLenny's Podcast
← All frameworks
LeadershipBoris Cherny

Under-Resource to Force Automation

Deliberately under-staff projects and hand out unlimited tokens so people are forced to automate with AI.

Difficulty
Moderate
Time to result
~months to results
Steps
4
Confidence
85%

A management framework for the AI era: intentionally put fewer people on a project so they are forced to offload work to AI ('Claudify'), while removing the cost friction by giving engineers as many tokens as they want. The counterintuitive claim is that you get more out of AI tooling with fewer people, and that early cost-cutting kills the experiments that produce your best ideas.

Origin

Boris Cherny's management approach on the Claude Code team at Anthropic; advice he gives to CTOs.

Core principles

  • 01Under-resourcing forces creative automation instead of throwing headcount at problems
  • 02Speed is a real advantage — 'if you can do something today, do it today'
  • 03Give engineers maximum tokens up front; don't optimize cost until an idea proves out
  • 04Token cost per experimenting engineer is small relative to salary, so freedom is cheap
  • 05Optimize (cheaper models, Haiku/Sonnet) only after something works and scales

How to run it

  1. 1

    Staff lean on purpose

    Put a small team — sometimes one engineer — on a project so the only way to ship fast is to automate heavily with AI.

    Pro tip Lean staffing plus intrinsic motivation to ship fast is what drives people to Claudify their own work.

  2. 2

    Remove token friction

    Give engineers effectively unlimited tokens so no one hesitates to try a crazy idea for cost reasons.

    Pro tip For an individual experimenting, token cost is low relative to their salary — treat it as cheap optionality.

    Watch out Don't cost-optimize at the start; premature optimization kills the experiments that generate breakthrough ideas.

  3. 3

    Let ideas prove themselves

    Allow engineers freedom to try ideas that would otherwise be 'too crazy.' Hold them accountable to cut losses fast on the ~80% that fail.

    Watch out Psychological safety to fail must be paired with accountability to abandon bad ideas quickly.

  4. 4

    Optimize only at scale

    Once an idea works and its token consumption becomes large, that's the moment to optimize — e.g. move from Opus to Sonnet or Haiku where it still works.

In the wild

The memory-leak debug

Boris debugged a Claude Code memory leak the traditional way — heap snapshots in a special debugger. A newer engineer instead just told Claude 'it seems like there's a leak, can you figure it out.' Claude took the heap snapshot, wrote its own just-in-time analysis tool, found the issue, and shipped a PR faster than Boris could.

Illustrated how leaning on AI beats manual effort, and why forcing that habit matters even for veterans.

Perk of unlimited tokens

At Anthropic everyone can use a lot of tokens, and some engineers spend hundreds of thousands of dollars a month in tokens. Companies are starting to offer 'unlimited tokens' as a hiring perk.

Freedom to over-spend on tokens surfaced ideas that cost-capped teams would never have tried.

Common mistakes

Cost-cutting too early

Optimizing token spend at the start removes the freedom engineers need to try crazy ideas, so you never discover what works.

Throwing headcount at problems

Fully staffing a project removes the pressure to automate, so the team defaults to manual work instead of Claudifying it.

Is it for you?

Best for

Engineering leaders and CTOs deciding how to staff and budget AI-augmented teams

Not ideal for

Mature products at scale where token spend is already large and cost optimization is the priority

From the transcript

you want to under-resource things a little bit at the start

10:30

then people are kind of forced to Claudify

24:00

Start by just giving engineers as many tokens as possible.

26:00

don't try to cost cut at the beginning

26:00

From the episode

Head of Claude Code: What happens after coding is solved

Boris Cherny