A weekly informal conversation with Aishwarya Khanduja and Alex Komoroske riffing on that week’s Bits and Bobs.

Week of May 26, 2025

Descript Episode • Google Drive Link • Bits and Bobs Discussed

AI as plumbing - Debate between AI as invisible infrastructure vs. getting hands dirty with messy implementation details
Open-endedness in ecosystems - How browsers created magic through unlimited linking possibilities, unlike closed systems
The Iron Triangle limitation - Can't have untrusted code, sensitive data, and network access simultaneously in current systems
Personal systems of record - Why we lack them and how they could become personal context engines
AI therapy concerns - Risks of surveillance, manipulation, and lack of human accountability in AI-powered therapy apps
LLM patience creating magic - How AI's ability to do infinite research work could enable powerful personalized assistance
Human-in-the-loop approach - Keeping humans involved in decision-making while AI handles research and options
Knowledge management dead ends - Why systems like Notion hit limits due to maintenance burden and lack of real-world connection
Activation energy as key metric - Lower barriers to action determine whether tools become useful or abandoned
App paradigm limitations - How current app models are inherently closed-ended compared to open web systems

Week of May 19, 2025

Descript Episode • Google Drive Link • Bits and Bobs Discussed

Second and third order impacts of technology vs. first order optimization
Personal context engines vs. engagement-maximizing dossiers
Presence → Intuition → Noticing → Intention → Execution → Alignment feedback loop
Adaptive efficiency vs. optimization that breaks resilient systems
The steamroller metaphor - slow-moving problems that crush you by the time you notice
The apprenticeship evaporation problem and tacit knowledge transfer
Gall's Law: complex systems evolving from simpler working systems
Parasocial apprenticeships as the new model for skill acquisition
Simpson's paradox in AI adoption - individually rational, collectively destructive
Pro-social vs. anti-social AI relationships

Descript Episode • Google Drive Link • Bits and Bobs Discussed

Intentional technology as an extension of human agency that aligns with aspirations rather than shallow wants
The four requirements for intentional tech: human-aligned, privacy by default, pro-social, and open-ended
The connection between modern tech concerns and transcendentalist philosophy (Thoreau's "living deliberately")
"Chatbots are a feature, not a paradigm" - the limitations of chat for structured, long-lived tasks
Orchestration overload - the burden of maintaining multiple personal knowledge systems
Ambient Smart Environments (ASEs) as behavioral scaffolds that increase metacognition
Context vs. content distinction - treating context like undifferentiated content leads to inappropriate sharing
Personality differences in organization - organizing ahead of time vs. just-in-time organizing
Three types of innovation: informative, transformative, and formative
The "T-shaped" model of expertise forming fractal patterns when multiple domains interact

Descript Episode • Google Drive Link • Bits and Bobs Discussed

Coactive computing as "working jointly" - technology as a trusted extension of human agency
The distinction between alignment with intentions vs. engagement maxing
"Engagement maxing" as a gravity well that consumer businesses fall into
Shallow engagement vs. deep engagement that leads to growth and change
The "Coco Melon effect" - optimizing for constant attention at the expense of metacognition
Technology that helps humans become "more themselves" rather than more average
Intelligence as a "mass noun" - flowing like water or sand rather than being centralized
Chat interfaces as the wrong affordance for many tasks
Data-facing techniques vs. user-facing applications in AI systems
The problem of "context collapse" when a single memory system tries to handle multiple domains
The "dossier" problem - when AI systems maintain hidden information about users
The rich, nuanced understanding that comes from shared human experiences
Privacy concerns related to accidental embarrassment and context-appropriate information
The multiple perspectives of truth - "we contain multitudes" across different contexts

Descript Episode • Google Drive Link • Bits and Bobs Discussed

Private intelligence vs. AI - the importance of having AI on your terms and in your control
The ChatGPT account lockout problem - why digital homes need to be on "your turf"
Hidden AI memory features - concerns about undisclosed information AI systems maintain about users
LLMs raising teachers' expectations - opportunities to redefine education beyond just preventing cheating
The Zeigarnik effect - how offloading thinking to AI affects our tacit knowledge and intuition
Decentralization and centralization in adaptive tension - how they naturally coexist and balance each other
The "Iron Triangle" of security - untrusted code, network access, and sensitive data challenges
Convex vs. concave systems - why systems with human agents are inherently destabilizing
Socratic methods in education - potential for AI to promote growth mindset learning
Building diverse high-trust teams - importance of shared understanding across different perspectives

Descript Episode • Google Drive Link • Bits and Bobs Discussed

Topics:

The relationship between algorithms and user wellbeing - 47% of Gen Z regret TikTok's invention, 50% regret Twitter's
Content vs. art - algorithms recommend content that centers users rather than challenging them
"Alienation from software" - users can't modify or fix the apps they rely on daily
The walled garden problem - Apple's ecosystem prevents running software outside the App Store
The "Coasian floor" - features below a certain economic threshold won't be implemented by companies
The power of lock-in - iMessage "blue bubbles" keeping users in Apple's ecosystem despite preferences
The concept of "infinite software" - moving beyond supply constraints to bespoke software on demand
The need for human-centered, pro-social software versus current aggregator models
"Meta apps" - clusters of customized mini-apps addressing personal use cases
Security threats co-evolving with ecosystems - systems only get targeted when valuable enough
The "Iron triangle" of security - choosing only two among untrusted code, sensitive data, and network access
How algorithmic content amplification increases polarization, even on non-political topics

Descript Episode • Google Drive Link • Bits and Bobs Discussed

Topics:

The shift from "big tech" to "better tech" that considers societal implications
Chat as a feature, not the entire system - "LLM pixie dust" infused into normal UIs
Tools extending human agency vs. anthropomorphized entities
Human-centered design vs. AI-human interaction paradigms
Prompt injection as a fundamental problem for AI systems
"Apps are about isolation, not integration"
LLMs making generalists almost as good as specialists
Specialists reaching local maxima vs. generalists finding global maxima with "epistemic humility"
Enlightened technology embracing more than just the CS lens
The power to transcend paradigms as the most effective leverage point
The balance between acting and understanding in technology development
The Zeigarnik effect - having "tabs open" in your mind drives creative innovation