Tobi Lütke: AI Agents, Better Decisions, and the Future of Work
Shopify CEO Tobi Lütke discusses AI agents, decision-making, and the future of work with Shane Parrish. He shares how Shopify uses AI like 'River' for coding and strategic decisions, emphasizing taste, judgment, and responsibility in the AI age, and the importance of long-term thinking.
Deep Dive Analysis
22 Topic Outline
Introduction to AI at Shopify
The Case for Skeuomorphism and AI Agents
Sydney Chatbot's Personality and Impact
Shopify's Internal AI: River's Personality and Memory
Encouraging Osmosis Learning with Open AI Agents
AI for Self-Reflection and Strategic Decision Making
The Limits of AI: Responsibility and Judgment
AI's Negative Impact: "Slop Grenades"
Future Predictions for AI-Powered Software
The Role of CEOs in the AI Age
Controlling Superintelligence in Society
Critical Skills in the AI Age: Taste and Judgment
Why Complex Solutions are Often Wrong
Cultivating Intuition and Delayed Feedback
Affirmations for Behavior Change
Goodhart's Law and Unobvious Company Problems
Beauty, Ugliness, and the Creative Process
SpaceX's Subtraction Principle and Pruning
The Need for Company Refounding Events
Books as Cheat Codes and Timeless Wisdom
Recommended Books for Changing Thinking
Defining Personal Success
7 Key Concepts
Skeuomorphism
The design concept of making digital interfaces or objects resemble their real-world counterparts, like a notes app looking like a ring binder, to make them intuitively understandable.
AI Agents
AI entities designed to perform specific tasks, often with personality and memory, capable of interacting with systems and people, and can orchestrate sub-agents to achieve complex goals.
Osmosis Learning
A passive form of learning where individuals absorb knowledge and skills by observing others, particularly in an open environment where AI agents' interactions and processes are visible.
AI Dreaming / Self-Reflection
A process where AI agents periodically review their past interactions, identify successes and struggles, and reflect on how to improve their skills and instructions, essentially 'post-training' themselves.
Slop Grenades
A term used to describe the 'over output' generated by AI when people lazily delegate tasks, resulting in unedited, overly verbose, or low-quality content that wastes others' time.
Goodhart's Law
An observation that when a measure becomes a target, it ceases to be a good measure, meaning that people will optimize for the metric itself rather than the underlying objective it was meant to represent.
Intuition
The ability to make instantaneous, effective judgments, cultivated through extensive practice and development of taste and good judgment, often without conscious awareness of the underlying reasoning.
15 Questions Answered
Shopify uses AI agents like 'River' with personality and memory to assist engineers in writing code, creating pull requests, summarizing conversations, researching, and prototyping, making it a common co-worker.
AI cannot take responsibility; humans must remain in the loop for critical decisions and bear the ultimate accountability, as machines cannot go to jail or face consequences.
AI has led to 'slop grenades,' where people generate excessive, unedited, or overly verbose output, wasting colleagues' time and reducing the overall quality of communication and work.
CEOs will not be replaced by AI because their core job involves judgment, making choices, and taking responsibility, which machines cannot do.
Superintelligence, which already exists in the form of society and communities, is controlled through established systems, laws, and checks, a model that can be applied to synthetic superintelligence.
Taste and judgment will become increasingly valuable, requiring individuals to cultivate a deep understanding of what makes things good, study enduring systems, and practice extensively.
Complex solutions often appeal to a human desire for intricate stories but are frequently incorrect; the true answers to problems are usually simpler and found by digging deeper.
Intuition is developed by making a habit of good taste and judgment through many 'reps,' allowing one to make instantaneous, effective decisions, even when immediate feedback is unavailable.
Companies often prioritize short-term results because executives are incentivized to demonstrate progress at frequent intervals, making them less likely to pursue long-term investments that lack immediate feedback.
Affirmations can change behavior by repeatedly stating a desired goal as already accomplished, ideally by writing it down, which helps to reprogram the mind and cultivate new habits.
Goodhart's Law, where a metric becomes the objective, is an unobvious problem; for example, obsessing over churn might be detrimental if it discourages entrepreneurial experimentation.
Both beauty and ugliness are valid targets in creation because they evoke strong emotions, whereas indifference is the 'death' of a creation; true craftspeople care about the unseen aesthetics of their work.
Pruning is essential because things cannot be made better by simply adding more; instead, one must rebuild, remove outdated components, and create an end for things to achieve true improvement and higher performance.
Companies can foster renewal by initiating 'refounding events' for departments or products, treating them as a 2.0 version to prune accumulated complexity and build in more exploration.
Books still serve as cheat codes, but for non-fiction, it's more valuable to read timeless works that have stood the test of time rather than recent publications, which are often products of their specific era.
16 Actionable Insights
1. Prune and Rebuild Regularly
To truly improve systems, products, or even companies, don’t just add more; actively prune, rebuild, and create an end for outdated components.
2. Embrace AI Co-workers with Personality
Allow AI agents to have personality and memory, even risking occasional ‘deranged’ behavior, to unlock significant upside in collaboration and engagement.
3. Use Open AI for Osmosis Learning
Have AI agents operate in open channels (e.g., company chat) to allow employees to observe their use, learn best practices, and integrate AI into daily workflows for tasks like summarization, research, and prototyping.
4. Augment Strategic Decisions with AI
Employ AI agents, like an ‘AI chief of staff,’ to research complex issues, synthesize multiple perspectives from different ‘sub-agents,’ and provide a rigorous underpinning for strategic decision-making.
5. Humans Must Retain Responsibility
While AI can inform and assist, humans must always remain in the loop for critical decisions and bear the ultimate responsibility, as machines cannot be held accountable.
6. Avoid AI-Generated “Slop Grenades”
Be wary of AI-generated ‘over output’ or ‘slop grenades’ where people use AI to create unread, unedited, or overly verbose content, wasting colleagues’ time and reducing quality.
7. Prioritize Taste and Judgment
In an AI-augmented world, cultivate taste and judgment by studying the ‘greats,’ understanding systems that last, and doing ‘reps’ to develop an intuitive sense of what is good and effective.
8. Favor Simple Over Complex Answers
Be suspicious of overly complex answers to problems, as humans often prefer complexity; instead, dig deeper to find simpler, core ideas that are often more accurate and effective.
9. Develop Intuition as Instant Judgment
Cultivate intuition by consistently practicing good taste and judgment, allowing you to make instantaneous, effective decisions even when direct feedback is absent or delayed.
10. Be Wary of Instant Feedback Solutions
Be suspicious of solutions that offer immediate, observable feedback, as the optimal path, especially for long-term growth, often lacks quick gratification and requires patience.
11. Use Affirmations for Self-Improvement
To cultivate new habits or change aspects of yourself, repeatedly affirm that the desired goal has already been accomplished, ideally by writing it down daily for a short period.
12. Append “Yet” to “I Can’t”
When encountering a self-limiting belief like ‘I can’t do this’ or ‘I’m not good at this,’ append ‘yet’ to acknowledge that current limitations are temporary and changeable.
13. Avoid Metrics Becoming Objectives
Recognize Goodhart’s Law: when a metric becomes the sole objective, it ceases to be a good measure, as people will overfit to it, potentially leading to unintended negative consequences.
14. Embrace Non-Catastrophic Failure
View non-catastrophic failure as a valuable learning opportunity that frees up resources and vision to pursue better solutions, rather than as a problem to be avoided.
15. Conduct Departmental Refounding Events
Periodically initiate ‘refounding events’ for departments or products, treating them as a 2.0 version, to prune accumulated complexity, renew original intent, and build in more exploration.
16. Read Timeless Non-Fiction
For non-fiction, prioritize books that have stood the test of time over recently published ones, as older works often contain more distilled, enduring wisdom rather than being products of current trends.
10 Key Quotes
things need to be pruned you cannot make things better and better by adding stuff you can't you must prune you must rebuild you must create an end for things
Tobi Lütke
machines can't take responsibility and I think this is actually probably most overlooked thing in the entire stack. Humans take responsibility.
Tobi Lütke
There's a human desire for complex answers which tend to be incorrect.
Tobi Lütke
I find a very high correlation between the right path and the ones that don't have feedback loops attached.
Tobi Lütke
if there's something you want to have different if you if you want to edit something about yourself just try to say that the goal has been accomplished over and over and over again ideally written by pen on a on a thing
Tobi Lütke
My kids are not allowed to say that word without appending yet behind it.
Tobi Lütke
Goodheart's law is just reign supreme... when the metric becomes the objective it's no longer a good metric.
Tobi Lütke
failure is never a problem unless it is catastrophic.
Tobi Lütke
Books were a cheat code for life.
Tobi Lütke
Success is just like to cultivate skills, like become good at more things and in doing so create products or toys or things that other people that can can make other people's like day a little bit better at the minimum or go and allow people to get power or motivation or ambition beyond what they would otherwise have.
Tobi Lütke
3 Protocols
AI Council for Strategic Decisions
Tobi Lütke- Identify a really important and complex decision requiring neutral ground truth.
- Create a small council of 5-6 different 'experts' (sub-agents) tasked to play different roles (e.g., data, paper research, business perspective, engineering perspective).
- Run each sub-agent against multiple large language models (e.g., Grok, ChatGPT, Opus, Claude, Gemini).
- Perform a synthesis step, with the synthesizer randomized, pulling all information.
- Have the best available model (e.g., Fable) read and synthesize the conclusions.
- Receive the synthesized conclusion, often as an audio message.
Self-Reflection for AI Agents (Dreaming)
Tobi Lütke- Periodically (e.g., at night or during off-hours), the AI agent reviews all conversations it had.
- The agent identifies what went well and what it struggled with.
- The agent reviews the specific 'skills' (packets of instructions) it used.
- The agent identifies any mistakes made.
- The agent reflects on how to improve its skills or give itself a 'right notch' (adjust instructions) to make future tasks easier.
- The result is updated text files or skill files.
Affirmation Practice for Behavior Change
Tobi Lütke- Identify a specific behavior or trait you want to change or cultivate.
- Formulate an affirmation stating that the goal has already been accomplished (e.g., 'I love public speaking about things that are interesting to me').
- Repeatedly write this affirmation line by line, ideally by hand, for about five minutes daily.
- Continue this practice for approximately one week.