AI Engineer World’s Fair 2025 - Tiny Teams

Key Concepts

• Tiny Teams: Small, highly effective teams achieving significant results.
• Generalist vs. Specialist: Hiring individuals with broad skillsets and adaptability versus those with narrow expertise.
• Player-Coach Model: Leaders who actively contribute to the work while also mentoring and guiding the team.
• Product-Led Hiring: Attracting talent through the appeal and utility of the product itself.
• Profit-First Mentality: Prioritizing profitability as a key decision-making metric.
• Super Tools: Leveraging existing tools in unconventional ways to maximize their utility.
• Harvester and Cultivator Model: Organizing engineering teams into product-focused (Harvesters) and infrastructure-focused (Cultivators) roles.
• Benchmarks as Memes: The idea that benchmarks, like memes, spread and influence the development of AI models.
• Benchmark Saturation: The phenomenon where models become overly optimized for specific benchmarks, diminishing their real-world relevance.
• High Agency: Individuals who take initiative, ownership, and solve problems independently.

Stack Blitz's Bolt.net: Scaling with a Small Team

The Unexpected Success of Bolt

• Stack Blitz, after seven years as a company with $0.7 million ARR, launched Bolt.net.
• The team was less than 20 people at launch.
• Initial expectations were modest, aiming for $100,000 ARR by year-end.
• Bolt exceeded expectations, more than doubling ARR in a short period.
• The growth was remarkably smooth, despite the product being a stripped-down MVP.

Team Structure and Philosophy

• Small Teams with More Context: Emphasizing a small number of people with broad understanding and agency.
• Empowerment and Speed: Enabling team members to build and implement without excessive bureaucracy.
• Low Burn Rate: Maintaining a lean operation to extend runway and increase opportunities for product market fit.
• Shared Core Values: Prioritizing low ego, high trust, user obsession, and resilience.
• Analogy to the Movie 300: The feeling of being a small, aligned team surrounded by overwhelming demands.

Prioritization and Firefighting

• The Fire Truck Analogy: Prioritizing high-impact areas and key infrastructure during periods of intense demand.
• Hard Decisions: Accepting that some issues will have to be left unresolved.
• Focus on the Essential: Concentrating on 10% of the tasks that yield the majority of the results.

Independent Thinking and Community Engagement

• Avoiding Hive Mind: Encouraging independent decision-making and critical evaluation of industry trends.
• Leading from the Front: Actively engaging with the community to build trust and gather feedback.
• Weekly Office Hours: Hosting live sessions to showcase progress and address user concerns.
• Community Building: Creating spaces for users to learn from each other and receive support.
• Hackathon.dev: Hosting the world's largest hackathon to drive product adoption and gather feedback.

AI-Powered Support

• Parah Help: Implementing AI-powered support tools to automate ticket resolution.
• SAM AI Assistant: Using Parah Help's SAM to resolve 90% of support tickets automatically.
• Custom AI Models: Training internal models to assist users within the product experience.

Key Takeaways

• Don't hire an army, hire a small number of Spartans.
• Focus on building a strong, aligned team with shared values.
• Prioritize ruthlessly and focus on the most impactful tasks.
• Engage with the community and build trust.
• Leverage AI to automate support and improve the customer experience.

Alie: The New Lean Startup

The Shift Towards Leanness

• Companies are becoming smaller, rounds are getting delayed, and profitability is being attained earlier.
• AI tooling is driving this shift, enabling small teams to generate significant ARR.
• The age of bloated teams and endless hiring rounds is over.

Alie's Success

• Building a portfolio of consumer software products.
• Scaling to $6 million ARR profitably with a team of four.
• Generating over half a billion views across social media.

Key Milestones

• January 2023: Launching Quizard AI mobile app with a viral TikTok video.
• Early Scaling: Utilizing the initial Codex model for AI outputs, cycling through multiple accounts.
• Fall 2023: Moving to New York City and launching a back-to-school campaign, hitting $1 million ARR and profitability.
• Spring 2024: Achieving number six in the education charts alongside companies like Dualingo.
• New Product: Unstuck AI: Reaching a million users in under nine weeks and generating over a quarter billion views.
• Third Product Launch: Launching a profitable product outside the education domain in three weeks.

The Lean Playbook

• Operating Principles:
  • Hiring: Hiring 10xer generalists with multiple complementary spikes.
  • Profit First Mentality: Prioritizing profits as a key decision-making metric.
  • KPI Ownership: Aligning everyone in the company to a KPI.
  • Continuous Process Refinement: Continuously improving processes based on feedback and failures.
  • Super Tools: Building compounding benefits by investing in technical playbooks and operational blueprints.
• Organizational Structure:
  • Harvester and Cultivator Model: Organizing engineering teams into product-focused (Harvesters) and infrastructure-focused (Cultivators) roles.
• AI Tooling Augmentation:
  • Using AI to automate tasks and augment the capabilities of the team.

Super Tools: Launch Darkly

• Intended Use Case: Feature management platform.
• Extended Use Cases:
  • Manual Traffic Load Balancer: Rerouting traffic to different LLM providers based on rate limits.
  • On-the-Fly Infrastructure Changes: Changing the prioritization of ingestion processes.
  • UI Modifications and Paywall Experiments: Running experiments without code pushes.

Key Takeaways

• Hire senior generalists who can work across the stack.
• Prioritize profits and focus on moving key metrics.
• Continuously refine processes and leverage super tools.
• Organize teams into product-focused and infrastructure-focused roles.
• Use AI to automate tasks and augment the capabilities of the team.

Gum Loop: Scaling Automation with a Small Team

The Gum Loop Journey

• Founder Max Broer Herbas spent six months building various software projects.
• Created the first UI for AutoGen, an open-source framework for AI agents.
• Realized that users needed a way to define complex workflows for AI agents.
• Pivoted to a workflow automation tool.
• Joined YC, raised a seed round, and then a series A.
• Scaled with under 10 people, focusing on hiring exceptional talent.

Hiring Approach

• Be Super Picky: Only hire candidates that are exceptionally exciting.
• Product-Led Hiring: Attract talent through the appeal and utility of the product itself.
• Work Trials: Bring candidates on for several days to assess their fit and skills.

Internal Operations

• Almost No Meetings: Prioritize deep focus time for building.
• Let People Build: Empower team members to take ownership and build without excessive oversight.
• Automate Everything Internally: Use Gum Loop to automate various business processes.

Culture

• What If We Built It Today?: Encourage rapid iteration and shipping.
• Make It Fun: Organize retreats and activities to offset the intensity of building.
• Be Intentional About Company Culture: Document values and expectations in a company handbook.

Key Takeaways

• Hire exceptional talent and give them the space to build.
• Automate everything possible to maximize efficiency.
• Create a fun and supportive culture.
• Be intentional about company values and expectations.

Benchmarks as Memes

The Power of Benchmarks

• Benchmarks, like memes, spread and influence the development of AI models.
• Benchmarks define what model providers are trying to get their models good at.
• A single person can come up with an idea for a benchmark that shapes the future of AI.

The Problem with Benchmarks

• Benchmarks are getting saturated.
• Models are becoming overly optimized for specific benchmarks, diminishing their real-world relevance.
• Benchmarks can be gamed or manipulated.

The Opportunity

• Create new benchmarks that are more relevant and meaningful.
• Shape the future of AI by defining what models should be good at.
• Empower people by giving them agency in the development of AI.

Characteristics of a Great Benchmark

• Multifaceted: Allows for a variety of strategies to succeed.
• Reward Creativity: Encourages innovative solutions.
• Accessible: Easy to understand for both models and people.
• Generative: Can be used to train future models.
• Evolutionary: Gets harder as models improve.
• Experiential: Mimics real-world situations.

AI Diplomacy: A Case Study

• A benchmark that uses the board game Diplomacy to assess the social and strategic abilities of AI models.
• Models must negotiate, form alliances, and betray each other to win.
• Revealed interesting personality traits of different models.
• Highlights the need for squishy, non-static benchmarks.

The Importance of Trust

• Benchmarks can help build trust in AI by showing people how it works and what it can do.
• People need to feel like they have a role in the development of AI.
• Benchmarks can empower people by giving them agency in the process.

Key Takeaways

• Benchmarks are powerful tools that can shape the future of AI.
• Create benchmarks that are relevant, meaningful, and empowering.
• Focus on building trust in AI by involving people in the process.