Azeem Azhar's AI-Driven Research Workflow Enhances Deep Thinking

The subtle power of deliberate friction in an AI-accelerated world, as revealed by Azeem Azhar, offers a profound counter-narrative to the relentless pursuit of speed. This conversation unpacks not just tools, but a philosophy for navigating technological advancement: how to harness AI's acceleration without sacrificing depth, how to build unique advantages by embracing tasks others avoid, and why understanding the "why" behind technology is paramount. Anyone seeking to build durable skills and competitive moats in the age of AI will find strategic insights here, moving beyond mere tool adoption to a more intentional approach to learning and work.

The Unseen Benefits of Slowing Down

In a landscape increasingly dominated by the promise of AI-driven speed and efficiency, Azeem Azhar introduces a compelling counter-argument: the strategic advantage of deliberate friction. While AI tools like ChatGPT, Claude, and Gemini can rapidly synthesize information, Azhar emphasizes that true understanding and unique insights often emerge from a more considered, even slower, engagement with material. This isn't a rejection of AI, but a nuanced integration, where the power of these tools is amplified by a human capacity for deep thinking, cultivated through practices that intentionally slow down the learning process.

Azhar illustrates this with a striking personal example: his use of a fountain pen that requires dipping into an ink bottle. This seemingly archaic practice serves a critical purpose. "If I am writing and every five words or 10 or 15 words I have to stop to dip the pen in ink," he explains, "I’m slowing myself down. I’ll get very haptic in the experience and look at what I’m writing and force myself to cross out mistakes, and feel frustrated about mistakes, so therefore slow my thinking down even more." This deliberate interruption forces a deeper engagement with the text, fostering a more profound understanding and a more considered output. This contrasts sharply with the frictionless consumption of information often facilitated by AI, which can lead to a superficial grasp of complex topics.

"In a world where I can move really quickly, I will slow myself down. I’ll get very haptic in the experience and look at what I’m writing and force myself to cross out mistakes, and feel frustrated about mistakes, so therefore slow my thinking down even more."

-- Azeem Azhar

This principle extends beyond writing. Azhar highlights how academic rigor, involving the sustained engagement with a peer-reviewed paper or a book over several hours, builds a capacity to grasp complex arguments and underlying concepts. This deep, active learning, he suggests, makes one "better at using the the ais." The ability to ask more sophisticated questions of AI, and to critically evaluate its outputs, is a direct consequence of this deeper cognitive work. The danger, as he implies, is that an over-reliance on AI for quick answers can bypass the development of these foundational reasoning skills, leaving users with a vast but shallow pool of knowledge.

Cultivating Competitive Moats Through Intentional Tooling

Azhar's approach to leveraging technology is not about adopting every new tool, but about building a personalized, effective toolkit that aligns with his intellectual goals. This involves a sophisticated understanding of how different AI models serve distinct purposes and a willingness to develop custom solutions. He notes that his team at Exponential View builds internal research tools, such as "Prism," which functions as a personalized Bloomberg terminal for analyzing vast amounts of research. This custom infrastructure allows him to query information, identify emerging themes, and receive curated insights, freeing up his human time for higher-level strategic thinking.

This deliberate construction of tools highlights a key aspect of competitive advantage: building systems that others are unwilling or unable to replicate. "We build our own tools," Azhar states, drawing a parallel to historical artisans who crafted their own instruments. This philosophy suggests that true innovation lies not just in using existing tools, but in understanding their limitations and creating bespoke solutions that offer a unique edge. The investment in custom tooling, while requiring significant effort, yields a durable advantage because it is deeply integrated into Azhar's workflow and tailored to his specific needs.

"You know if you were a painter in the 16th or 17th century you went and made your own ink... we build our own tools within my team at exponential view."

-- Azeem Azhar

Furthermore, Azhar employs AI models strategically, understanding their strengths and weaknesses. He uses Manus for broad, "find color" type research queries that don't require deep argumentation, running them overnight to gather anecdotes and case studies. For more complex analysis or coding tasks, he might turn to Gemini Pro or Claude, recognizing Claude's particular aptitude for coding and text refinement. This layered approach, where cheaper or more specialized models are used for less demanding tasks, and more powerful (and expensive) models are reserved for critical functions, demonstrates a sophisticated understanding of cost-efficiency and resource allocation. This "good enough" principle, applied judiciously, allows for scalability and sustainability in his workflow.

The Evolving Grammar of Interaction

The conversation also delves into the emergent "grammar" of interacting with AI, a concept that underscores how our relationship with technology shapes our thinking and capabilities. Azhar observes that as users become more comfortable and confident with AI tools, they naturally push them harder, developing new communication styles and expectations. This is akin to how the advent of emoji and features like Instagram Reels transformed social media interaction.

He highlights his own evolution, moving from "hunting and pecking" with ChatGPT two years ago to now pushing more advanced models like Gemini 3 harder, even in areas outside his expertise. This progression suggests that mastery of AI is not a static endpoint but an ongoing process of adaptation and learning. The "new grammar" emerging involves more sophisticated prompting, iterative refinement, and a deeper understanding of how to elicit nuanced responses.

"There is a new grammar, a new behavior emerging... I don't know whether you get to a satisfying of these tools... maybe they are like personal motivation maybe they're like I haven't yet run my best time I haven't yet done the most pull ups I'll ever do in my life and if that's the case I can imagine people wanting to have better and better systems."

-- Azeem Azhar

This evolving interaction model has significant implications for education and professional development. Azhar's team is already incorporating AI into their workflows, and he is beginning to teach students how to use these tools responsibly and effectively. The challenge lies in navigating the rapid pace of AI development and ensuring that users develop not just proficiency, but a critical understanding of AI's capabilities and limitations. The emphasis on owning the output, regardless of the tool used, reinforces the idea that human judgment and accountability remain paramount. As AI continues to integrate into our lives, understanding and shaping this new grammar of interaction will be crucial for unlocking its full potential while mitigating its risks.

Key Action Items:

• Embrace Deliberate Friction: Intentionally slow down your engagement with complex information. Use methods like longhand note-taking or structured reading to deepen comprehension. (Immediate Action)
• Develop a Layered AI Strategy: Don't rely on a single AI model. Experiment with different tools (e.g., Perplexity for quick answers, Claude for coding, DeepSeek for cost-efficiency) and assign them tasks based on complexity and cost. (Ongoing Experimentation)
• Build Custom Workflows: Identify repetitive knowledge-management tasks and explore building or adapting tools to automate them, creating a personalized "information dashboard." (Long-term Investment: 6-12 months)
• Practice Sophisticated Prompting: Move beyond basic queries. Learn to prompt AI with multiple roles, specific constraints, and iterative feedback loops to elicit more nuanced and valuable responses. (Immediate Action)
• Own Your AI-Assisted Output: Treat all AI-generated content as your own. Critically review, edit, and fact-check everything before using it, ensuring accountability and accuracy. (Immediate Action)
• Invest in Foundational Reasoning: Prioritize understanding underlying concepts and arguments over simply getting quick answers. This will enhance your ability to use AI tools more effectively and critically. (Ongoing Practice)
• Explore "Good Enough" Models: For many common tasks, less advanced or cheaper AI models are sufficient. Identifying these can significantly reduce costs and free up resources for more critical applications. (Immediate Action)