January 2025
January 31, 2025
The Compound Effect of Small Optimizations
This month reinforced a fundamental truth: small, intentional changes compound into dramatic productivity gains. Every optimization builds on the last, creating momentum that's greater than the sum of its parts.
Discovery: AI at 30,000 Feet
Here's a travel hack that changed my flight experience entirely: Meta's Llama LLM through WhatsApp works with basic in-flight messaging. Suddenly, those hours in the air become thinking time rather than dead time. Simple shift, profound impact.
The Build vs. Consume Paradigm
I've started tracking my build-to-consume ratio religiously. The data doesn't lie – we consume exponentially more than we create. This month I resumed time logging, not for micromanagement, but for clarity. What gets measured gets managed, and building requires intentional protection from consumption's gravitational pull.
Age as Competitive Advantage
Getting older brings an unexpected gift: ruthless clarity about priorities. Time becomes finite, not infinite. Energy becomes precious, not renewable. Focus becomes surgical, not scattered. This isn't limitation – it's liberation from the noise.
The Reading Renaissance
I've optimized my entire reading system for maximum absorption. Vertical scrolling on Kindle across all devices – iPhone, Mac, PC – transforms the reading experience. The iPhone app handles EPUB/PDF uploads beautifully. Small friction removed, big behavior change enabled.
AI-Powered Learning Synthesis
NotebookLM creates custom podcasts from your learning materials. The concept is brilliant – turn written knowledge into commute-friendly audio. The execution is inconsistent, with noticeable hallucinations where it discusses topics beyond the source material. But even imperfect AI tools can accelerate learning when used thoughtfully.
Tactical AI Applications
ChatGPT excels at breaking down PDF books by chapter. This solves a real problem: managing LLM context limits. Instead of overwhelming the model with entire books, feed it specific sections. Want to discuss Chapter 17? Extract just Chapter 17. Precision beats volume.
The Patience Trade-Off
Amazon Delivery Day shipping: slower delivery in exchange for credits toward Kindle and Audible purchases. For non-urgent items, this is pure arbitrage – trading time for money, then investing that money back into knowledge acquisition.
Tools That Stick
Switched from iTerm2 to Ghostty for terminal work. This reflects a deeper trend: my CLI usage has increased dramatically over the past year. When you live increasingly in the command line, terminal optimization becomes crucial infrastructure.
Ancient Wisdom, Modern Application
A year of reading "The Daily Stoic" has been transformational. Stoicism isn't philosophy for philosophy's sake – it's a practical operating system for navigating uncertainty and maintaining focus on what you can control.
Technical Deep Dive
The RFB project forced me into CloudFormation, Packer, YAML, and system architecture. These aren't just technical skills – they're leverage multipliers. Understanding infrastructure as code means understanding how to scale solutions beyond manual processes.
AI as Writing Partner
Amazon's internal "Cedric" LLM has become my writing collaborator. Drafting, structure brainstorming, tone refinement, filler word elimination, email composition – it handles the mechanical aspects, freeing me for strategic thinking. This raises serious questions about standalone tools like Grammarly when AI becomes embedded everywhere.
End-to-End Ownership
I'm deliberately building complete systems: product, Python, SQL, deployment, frontend, pipelines. Previously, I'd partner with specialists for unfamiliar areas. Now I'm pushing into discomfort zones. Full-stack understanding creates full-stack leverage.
Premium Tool Strategy
Invested in Perplexity Pro (annual via Xfinity offer) and resubscribed to Cursor Pro. These aren't expenses – they're force multipliers for acceleration. The cost of premium tools is trivial compared to the opportunity cost of moving slowly.
The Reasoning Revolution
Testing o3-mini and R1 reasoning models was revelatory. Their step-by-step thinking and constraint handling capabilities are remarkable, especially for synthetic data generation. We're witnessing the emergence of AI that shows its work, not just its conclusions.
Speed as Competitive Advantage
Typing speed increasingly determines productivity with LLMs. Input/output becomes the bottleneck, not generation speed. Exploring voice-to-text for potentially higher throughput. When AI removes thinking bottlenecks, physical interface speed matters more.
Workflow Acceleration
Raycast has become essential workflow infrastructure. These micro-optimizations – faster app switching, instant calculations, quick searches – accumulate into significant time savings throughout the day.
Note-Taking Philosophy Shift
Revised my entire approach: notes for ideation, not retention. Searchability trumps complex PKM systems with backlinks and hierarchies. Moved to UpNote after experimenting with Workflowy and Obsidian. Simple, searchable, fast. Sometimes the best system is the one you actually use.
Git Discipline in the AI Era
With LLM-generated code, frequent snapshots and disciplined workflow management become crucial. It's similar to managing outsourced work – clear requirements, explicit expectations, comprehensive tests. AI amplifies both good and bad development practices.
Project-Based Learning
The RFB (Regular Flyer Buddy) project exemplified full-stack development. Phase 1: local full-stack app. Phase 2: AWS deployment. Each phase built skills that compound into the next challenge.
Post-Project Reflection
Interacting with LLMs after completing AI-assisted projects accelerates understanding dramatically. Asking about architectural choices and design decisions reveals the "how" and "why" behind generated code. This meta-learning approach transforms AI from black box to teaching assistant.
Tool Discovery
RepoMix (repomix.com) converts GitHub repositories to XML, perfect for feeding entire codebases into LLM context. These utility tools that bridge different AI workflows are incredibly valuable.
Multi-Model Strategy
Different LLMs excel at different tasks. Claude Projects for structured thinking and artifacts. Cursor for cost-effective coding. ChatGPT for follow-up questions without disrupting primary context. Specialization beats generalization when you can afford multiple tools.
Complexity Management
Four hours debugging Docker taught me a crucial lesson: complex, detailed prompts often fail spectacularly. Step-by-step, incremental approaches work better, especially for deployment tasks. Break down complex requests into simple, sequential steps.
The Communication Effect
I interact with LLMs approximately 5x more than with humans. This constant interaction seems to be improving my ability to ask better questions and communicate more clearly. AI becomes a training ground for human communication skills.
Synthetic Data Success
Reasoning models excel at creating synthetic data. Built a complete synthetic data generation engine using these models, fulfilling a long-standing project idea. Sometimes the best time to build something is when new capabilities make it suddenly feasible.
Forward-Looking Architecture
Began designing a new forecasting architecture for work. The key insight: build for tomorrow's capabilities, not just today's requirements. Architecture decisions create or constrain future possibilities.
The thread connecting all these observations: small optimizations compound, AI amplifies existing processes, and intentional tool selection creates sustainable competitive advantages. January wasn't about revolutionary changes – it was about evolutionary improvements that set the foundation for bigger leaps ahead.