# The New Lean Startup

Sid Bendre, Oleve | AI Engineer World's Fair 2025 | 13:26

Source: https://www.youtube.com/watch?v=pQz-PgA1eJw
Channel: AI Engineer (https://www.youtube.com/@aiDotEngineer). Summarised by AIE Talks.
Page: https://aietalks.com/talks/the-new-lean-startup
Published: 2025-07-01
Tags: agents, product-strategy, team-adoption, workflows

## TL;DR
- Oleve built a profitable portfolio of consumer AI products with a four-person team by combining strong generalists, clear ownership, and reusable technical and operational playbooks.
- The company splits engineering into harvesters, who own products and their metrics, and cultivators, who build shared AI infrastructure and automation for the rest of the company.
- Oleve is moving from human-led tools to workflow automation and then to coordinated agent systems that can run parts of research, growth, operations, and product development.

## Summary
Sid Bendre presents Oleve as an example of a smaller company building and scaling several consumer software products with a tiny team. He describes the principles behind that approach: hire highly capable generalists, prioritize profit, give every person a KPI, improve repeated processes, consolidate workflows into flexible tools, and turn lessons into reusable blueprints. Oleve organizes engineers as harvesters and cultivators. Harvesters own individual products, while cultivators build shared AI infrastructure and automation. Bendre also describes a three-stage path from tools that assist people, to systems that automate complete workflows, to multiple agents coordinated by an autonomous decision-making system. Examples include using LaunchDarkly to route model traffic, alter ingestion processes, and run experiments without code pushes. The talk ends with Oleve's ambition to let strategic people command agent systems across products and business units, while humans contribute judgment, taste, and direction.

## Key ideas
### Tiny teams can scale profitable consumer products
[00:01](https://www.youtube.com/watch?v=pQz-PgA1eJw&t=1s)
Bendre opens by arguing that successful companies are becoming smaller, delaying fundraising, and reaching profitability earlier, partly because of AI tooling. He presents Oleve as an example. The company launched Quizard AI on January 26, 2023, after a TikTok video produced a million views and 10,000 users in less than 30 hours. It reached its first million dollars in ARR and profitability within nine months. Its later product, Unstuck AI, reached a million users in under nine weeks and generated more than a quarter billion social views in a month. Bendre says the company had only four people at the time described.

### Oleve hires for complementary strengths and gives people direct metric ownership
[03:37](https://www.youtube.com/watch?v=pQz-PgA1eJw&t=217s)
Oleve's hiring rule is to hire the right person or not hire at all. Bendre wants 10x generalists with several complementary strengths. Product engineers may combine full-stack development, product thinking, and strong computer science fundamentals. Marketers may code, and designers may build. The company also has a profit-first mentality. Bendre says profit gives the company a clear way to make decisions and maintain focus. Every employee owns a KPI, and decisions are checked against that metric. He says this reduces micromanagement because people focus on moving their metric week over week.

### Repeated work should become a better process and then a reusable blueprint
[04:40](https://www.youtube.com/watch?v=pQz-PgA1eJw&t=280s)
Oleve treats problems and failures as system failures. For any repeated process, the team asks how to improve the next run and what went wrong in the previous one. This creates feedback loops for both operations and engineering. The company also follows a rule Bendre calls 'don't learn it twice.' It invests in technical playbooks, code-complete templates, internal libraries, and shared modules so lessons from one product can be reused elsewhere. Shared infrastructure includes model-provider integrations, notifications, and an experimentation layer. Bendre connects this compounding approach to Unstuck's rapid growth, which used learnings accumulated from Quizard.

### A feature management tool can become a flexible control layer for infrastructure
[05:53](https://www.youtube.com/watch?v=pQz-PgA1eJw&t=353s)
Oleve uses LaunchDarkly beyond its usual feature-management role. The company places it between its language-model calls and uses it as a manual traffic load balancer. That lets the team reroute traffic between model providers when it encounters rate limits or changes its priorities. Bendre says this was especially useful when Azure OpenAI quotas were difficult to increase. Oleve also uses LaunchDarkly to change the order of file-ingestion processes on the fly when a third-party service fails. An experimentation layer around the tool allows the team to modify user interfaces and test paywalls without pushing new code.

### Harvester and cultivator engineers have different ownership boundaries
[07:25](https://www.youtube.com/watch?v=pQz-PgA1eJw&t=445s)
Oleve's engineering structure borrows from Palantir's model and divides work between harvesters and cultivators. Harvesters are product engineers who own a product end to end. They work with product metrics, run A/B experiments, build features, and work with marketing. Their responsibility is to build products that people want and pay for. Cultivators are AI software engineers who build the company's agentic operating system. They create automation for areas such as marketing, design, and product, with infrastructure that can help the whole company ship and scale across markets.

### AI tools should multiply strong people rather than cover missing ability
[08:38](https://www.youtube.com/watch?v=pQz-PgA1eJw&t=518s)
Bendre says Oleve thinks of tools as a way to turn a 10x person into a 100x person. The company does not frame them as a substitute for hiring people who meet its standards. Oleve pays for a range of services that help with script writing, campaign analysis, operations, code generation, and communications. Bendre describes the result as giving each person a chief of staff inside the company. This tooling is part of the broader effort to let a small team handle work that would otherwise require more specialized staff.

### Oleve is moving from assisted tasks toward coordinated autonomous workflows
[10:06](https://www.youtube.com/watch?v=pQz-PgA1eJw&t=606s)
Bendre describes three stages of automation. First, human-led tooling adds dashboards, scraping tools, and prompt-chain agents inside existing team workflows. Second, workflow automation takes over complete processes and frees people to work on more interesting or higher-leverage tasks. Third, Oleve wants to coordinate multiple agents under one decision-making system. Current examples include market-research agents that scan for promising categories and acquisition targets that fit the company's strategy. Growth systems monitor content, trigger feedback loops, and help manage creator and influencer relationships.

### The long-term operating model puts human strategy above agent execution
[12:11](https://www.youtube.com/watch?v=pQz-PgA1eJw&t=731s)
Oleve's longer-term vision is for one strategic person to direct specialized agents. That person sets the objective, while agents execute and the system improves over time. Bendre extends the idea from managing a team to running an entire product, several business units, or a portfolio of apps. Humans would be hired for strategic insight, talent, and taste, while agents handle much of the company's execution. He describes this as a portfolio of one-person billion-dollar companies. The talk also briefly introduces Trellis, a framework Bendre says is designed to scale reliable AI user experiences to five million users in systems intended to go viral.

## Notable quotes
- "We either hire right or not at all." (03:40)
- "Profit is power and profit is focus." (04:00)
- "Does this move your KPI?" (04:19)
- "Don't learn it twice." (05:19)
- "They set the objective, the agents execute, and the system improves over time." (12:11)

## Tools & references mentioned
- Oleve
- Quizard AI
- Unstuck AI
- LaunchDarkly
- Azure OpenAI
- OpenAI Codex
- Palantir
- Trellis
- Duolingo
- Photomath
- TikTok

## Who should watch
- You are building several products with a small engineering or operations team and need a way to reuse what one product teaches you.
- Your team is deciding how to divide product ownership from shared AI, automation, and infrastructure work.
- You want concrete examples of feature flags, workflow automation, and agents being used to reduce recurring manual work.

## Related talks

- [Tiny Teams](https://aietalks.com/talks/tiny-teams) (Britney Walker, CRV & Eric Simons, StackBlitz & Sid Bendre, Oliv & Max Broer Herbas, Gum Loop & Grant, Gamma & Vic Paruturi, Data Lab & Alex Duffy, Every, 3:37:25)
- [Stop Ordering AI Takeout: A Cookbook for Winning When You Build In House](https://aietalks.com/talks/stop-ordering-ai-takeout-a-cookbook-for-winning-when-you-build-in-house) (Jan Siml, 10:45)
- [Small AI Teams with Huge Impact](https://aietalks.com/talks/small-ai-teams-with-huge-impact) (Vikas Paruchuri, Datalab, 17:36)
- [Building AI Products That Actually Work](https://aietalks.com/talks/building-ai-products-that-actually-work) (Ben Hylak, Raindrop & Sid Bendre, Oleve, 18:42)
- [Rethinking Team Building: How a 30-Person Startup Serves 50 Million Users](https://aietalks.com/talks/rethinking-team-building-how-a-30-person-startup-serves-50-million-users) (Grant Lee, Gamma, 18:06)