Building and scaling AI-native developer tools with ruthless prioritization and high-agency culture

Cannibalize your own product every 6-12 months to stay ahead.

The team deliberately obsoletes its current form factor so that the next version makes the old one look 'silly' or 'dumb'.

Why: In fast-moving AI markets, incremental improvements are insufficient; only radical leaps maintain differentiation.

"We should be cannibalizing the existing state of our product every six to 12 months. Every six to 12 months, it should make our existing product look silly. It should almost make the form factor of our existing product look dumb."

Hire only when the team is 'dehydrated'—truly underwater for a function.

No role is filled until existing members are demonstrably overloaded; this prevents politics and manufactured work.

Why: Excess headcount creates internal lobbying for non-critical projects, diluting focus and velocity.

"I want the company to almost be like this dehydrated entity. Every hire is like a little bit of water, and we only go back and hire someone when we're back to being dehydrated."

Reward the individual who delivers the most impact with the fewest people.

Team size is not a status symbol; value is measured by outsized output per contributor.

Why: Aligns incentives with lean execution and discourages empire-building.

"The person that is the most valuable at the company is the person that can do the most crazy project out there with as few people as possible, and that's what you should be rewarding internally."

Screen for high agency and willingness to work hard, not just intelligence.

Candidates are asked directly how hard they are willing to work; ‘I work smart’ is considered a red flag.

Why: In a collaborative group project, one low-agency member lowers the bar for everyone.

"If we have many smart people at our company that also work hard, what's the differentiate are going to be? Are you just going to pull them down?"

Build a new interface when the old one caps AI leverage.

Forking VS Code into Windsurf tripled autocomplete acceptance because the UI could surface richer interactions.

Why: Model quality plateaus quickly; user-interface innovation unlocks the next order-of-magnitude improvement.

"Immediately by porting over to Windsurf, our acceptance rate tripled. Same ML models, it just tripled."

Invest in enterprise sales early if large customers signal strong demand.

Codeium hired a VP of Sales in late 2023 and scaled to 80+ GTM people after Fortune 500 pilots flooded in.

Why: Large companies rarely adopt via self-serve; dedicated sales unlocks revenue and deeper product requirements.

"If you want to sell to the Fortune 500, it is very hard to do that purely by swiping a credit card."

Use specialized models for planning vs. retrieval vs. editing to maximize quality and minimize cost.

Sonnet handles high-level planning; custom smaller models retrieve relevant code chunks and apply precise edits.

Why: Sending 100-million-line codebases to a frontier model is infeasible; hybrid orchestration keeps latency and cost low.

"We try to make the anthropic-based model… do as much of the high-level planning as possible… we actually make sure we run models to actually chunk up the entire codebase."

Start with small, explicit prompts to learn the product’s hills and valleys.

New users should request minor refactors first; large sweeping changes increase the blast radius of mistakes.

Why: Builds user intuition for what the AI reliably handles and where human review is still critical.

"Start by making smaller changes… learn what the hills and valleys of the product are."