
Building a 100 Million Computation Solution in a Week with Brendan Hellweg (Holly)
Brendan Hellweg, co-founder of Holly, joins Greg Shove to explain how his company is helping local governments solve job design and collective bargaining problems that have historically required expensive consultants, and why AI finally makes that possible at scale.
Brendan came to Holly through public service, not software. He worked in the Baltimore mayor’s office, helped build a 300-person COVID-era public health program, and met his co-founder Sherry at the joint Harvard Business School and Kennedy School program, where they discovered a shared love of improv and a shared conviction that local government hiring and compensation processes were broken. What started as a pivot away from a pothole-reporting app became Holly: software that helps HR directors at cities, counties, and states figure out what to pay people, what to require for a role, and how to prepare for union contract negotiations.
In this episode, Brendan and Greg discuss:
- Why Holly pivoted away from its original pothole-reporting idea once they realized government agencies already had enough demand, just not enough staff to service it
- How improv principles like “yes, and” shape Holly’s culture and cross-team collaboration
- Why government sales cycles can actually move fast when agencies urgently need a solution, and how Holly grew from one customer to around 100
- Why benchmarking job data across thousands of distinct government job titles is a problem only AI can realistically solve at this scale
- Building a solution for a 2 million computation compliance problem in about a week, and shipping it to every other California agency facing the same law
- How Holly thinks about inference spend as a percentage of payroll rather than a fixed budget, and why they don’t use token leaderboards
- The question of PM-to-engineer ratios as engineering output accelerates, and what that means for where bottlenecks actually show up
- Running the company as a “portfolio of bets,” including a $1.1 million grant-funded project exploring the future of public sector work with Virginia and Oklahoma
If you’re building AI-native software for a traditionally slow-moving industry, or trying to figure out how much of your team’s output should come from inference versus headcount, this conversation is a grounded look at what that actually requires.
Altri episodi di "Supercompanies with Greg Shove"



Non perdere nemmeno un episodio di “Supercompanies with Greg Shove”. Iscriviti all'app gratuita GetPodcast.








