Don't Just Look at Now—Think Two Steps Ahead | Mike Krieger, Instagram Co-Founder & Anthropic CPO

Key Concepts

Entrepreneurship, Angel Investment, Product Development, AI, LLMs, Machine Learning, Personalization, User Feedback, Model Training, Team Building, Startup Culture, Prototyping, Design Thinking, User Intent, Consolidation, Long-Horizon Interaction, Agency, Independence, Self-Improvement, Relationships, Learning, Evolution.

When to "Call It" on a Venture

One key piece of advice for entrepreneurs is to regularly check in with investors. Entrepreneurs often feel pressured to continue a failing venture due to perceived investor expectations. However, investors may recognize when a company has exhausted its options and support the decision to shut down. It's crucial to have open communication and not view the closure as a tragedy, but rather a learning experience.

Mike Krieger's Background and Career

Mike Krieger is currently the Chief Product Officer at Anthropic, an AI company. He co-founded Instagram and served as its CTO from 2010 to 2018. He also co-founded Artifact, an AI news startup, before joining Anthropic. Krieger credits his early interest in technology to his father, who introduced him to computers at a young age. He enjoyed disassembling and reassembling software, sparking a lifelong passion.

Symbolic Systems Program at Stanford

Krieger attended Stanford University and studied Symbolic Systems, a program combining computer science, design, philosophy, and psychology. The program emphasized solving problems for people through technology and the importance of understanding the context around software development. Key takeaways from the program included:

1. Design Thinking: Everything built should solve a problem for someone.
2. Prototyping: Build prototypes and gather feedback early and often.
3. Teamwork: The value of a good team and the right partners.

Instagram's Origin Story

In 2009, the mobile landscape was still nascent. Cameras on phones were improving, but not great. Facebook's mobile app was a basic recreation of the website, and there were few new social products. Krieger reconnected with Kevin Systrom, who was building Burbn, a location-sharing app with photos and videos. Krieger was excited by the potential for mobile devices to create personal connections. Three key moments led to Instagram's creation:

1. Reconnecting with Kevin Systrom and aligning on ideas.
2. Krieger obtaining his work visa.
3. Realizing Burbn wasn't working and pivoting to focus on photo sharing.

The art of building great products, especially with emerging technologies, involves identifying technologies ready for broader adoption and building a user-friendly product around them. Look for early adopters and the rate of change in the technology.

Facebook Acquisition of Instagram

Facebook acquired Instagram after it reached 100 million registered users. Facebook's mission was to help Instagram grow without forcing it onto their infrastructure. Facebook aimed to prioritize Instagram as they would an open graph partner, allowing them direct access to Facebook's code. Before the acquisition, Instagram was a 13-person team with revenue ideas but no concrete implementation. Joining Facebook allowed them to scale while preserving a startup culture focused on experimentation and rapid iteration ("move fast and break things").

The Value of Team

The value of the team is paramount. A good team consists of talented individuals without excessive ego, who are willing to be generalists and are passionate about the product. Fluidity within the team, where members can contribute across different areas (e.g., backend engineers building iOS/Android components), is crucial. Great product breakthroughs come from team members who are close to the details and proactively generate ideas.

Artifact: An AI News Startup

In 2021, Krieger co-founded Artifact, an AI news startup, based on the observation that despite the rise of machine learning, few products felt truly personal. The goal was to combine cutting-edge machine learning with good product design to create personalized experiences. They started with news and articles, aiming to connect people to relevant content. The product spent two years in private beta, which was deemed too long, leading to team exhaustion by launch. After a year, the team observed a lack of energy and difficulty shifting the product's trajectory. Two key issues contributed to Artifact's challenges:

1. Poor user experience on the mobile web pages linked from the app (ads, formatting issues).
2. The product required significant user data to provide truly personalized recommendations, leading to high bounce rates from new users.

A key lesson learned from Artifact is that users won't adopt a product solely based on its underlying technology or intelligence. It must solve a problem and be useful from the start.

Applying Artifact's Lessons to Anthropic

At Anthropic, Krieger emphasizes making AI models useful even for users unfamiliar with LLMs. This involves lowering the barrier to entry and providing concrete use cases. They also focus on gathering user intent and personalizing the experience more quickly, such as through onboarding questions.

Determining When to Shut Down Artifact

To decide when to shut down Artifact, Krieger and Systrom made a list of remaining ideas they wanted to try. They prioritized three, tested them, and assessed whether they changed the company's trajectory. When those efforts failed to shift the direction, they decided to move on. It's important to be concrete with either a date or a set of projects to avoid getting stuck in a failing venture.

Joining Anthropic

Joining Anthropic offered a combination of established company and zero-to-one opportunities. While the company had launched models and cloud.ai, there was still room to build new products like mobile apps and Claude Code (an agentic coding tool).

Claude's Character and Personality

Anthropic has a dedicated team focused on Claude's character, philosophies, and responses. They use internal evaluations and user feedback to understand how Claude's personality is evolving. This includes determining Claude's verbosity and ensuring it recognizes its limitations.

Challenges of Building Products with Evolving Models

Building products alongside model development is challenging because both are moving targets. Models evolve up to a week before launch, requiring product teams to be adaptable. It's crucial to be upfront about the capabilities, limitations, and risks of the models.

Gathering User Feedback on Claude

Anthropic gathers user feedback through thumbs up/thumbs down ratings and written explanations for each response. This feedback is aggregated and analyzed to identify themes and improve the model. They do not train on user data or conversations, but use the feedback to guide model improvements.

Consolidation in the AI Landscape

Krieger expects consolidation in two areas:

1. Model Development: Due to the high cost of building and training advanced models, the field will likely narrow to a few major players. Anthropic is well-positioned due to partnerships with Amazon and Google.
2. AI Apps: The current explosion of AI-related startups will likely consolidate as successful ideas gain traction and business models evolve.

Future of AI: Long-Horizon Interaction, Agency, and Self-Improvement

Key areas for AI development include:

1. Long-Horizon Interaction: Models need to understand users beyond single conversations, considering their relationships, moods, and challenges.
2. Agency and Independence: Models should be able to take direction, perform tasks, and conduct research over extended periods.
3. Self-Improvement: Models will be able to provide insights about users themselves, acting as personal coaches or advisors.

Advice for Young People

Krieger advises young people to:

1. Not worry about every step making logical sense, trusting that seemingly unrelated projects can connect later.
2. Remember that relationships built early in your career will recur and are worth investing in.

Krieger's Future at Anthropic

Krieger hopes Anthropic will be his last job, seeing the opportunity to build interesting products alongside research for many years. He emphasizes the importance of continuous learning, evolution, and working with interesting people.

Entrepreneurship and AI

Entrepreneurship is about identifying ways to make the world better and feeling empowered to create that change. AI will empower people to ask and answer the question of how the world could be different and better.