rishikesh kubasad

Deconstructing the code, systems, and economics of frontier tech. Re-implementing paper architectures & company stacks. Building open tools.

[rishikesh kubasad @rishikeshbiz]

@paulg If spending ~4% of your round on a domain is 'signaling seriousness,' standard startup burn rate logic just got a major software update

[rishikesh kubasad @rishikeshbiz]

@MatthewBerman Once a model starts discovering its own efficiency gains and generating verifiable proofs, intelligence shifts from a training curve to a self-reinforcing flywheel. Other labs won't just be competing on compute; they'll be competing on compounding speed.

[rishikesh kubasad @rishikeshbiz]

@omarvvvr Distribution and UX beat raw benchmarks every single time. Convenience > capability.

[rishikesh kubasad @rishikeshbiz]

@hnshah Spot on. When you can’t appreciate what’s already sufficient, no amount of scale or metric growth will ever feel like enough. It becomes a treadmill, not a destination.

[rishikesh kubasad @rishikeshbiz]

@jacobmparis Spot on. The shift with AI isn't that consumers suddenly want to configure software it's that AI can configure the system for them in real time without them needing to become hobbyists

[rishikesh kubasad @rishikeshbiz]

@steipete 2024 was the year of prompt engineering, 2025 was rag & context, 2026 is full-blown agency. Hyped to see what gets unveiled today!

[rishikesh kubasad @rishikeshbiz]

@DeryaTR_ Shifted from AI as an assistant to AI as an engine of discovery. Formal verification plus deep reasoning is the exact formula to unlock post-human bottlenecks in math and science.

[rishikesh kubasad @rishikeshbiz]

@swyx Big tech has all the compute in the world, but "legal and risk approval" remains the ultimate bottleneck

[rishikesh kubasad @rishikeshbiz]

@PreetamNegalur @stripe Appreciate it Preetam! Let me know your thoughts once you finish it.

[rishikesh kubasad @rishikeshbiz]

how @stripe processes billions with zero downtime:

a 15-minute deep dive into @stripe's immutable Engine covering system architecture, fault tolerance, and platform economics.

watch below ↓

(substack guide + citations in reply)

[rishikesh kubasad @rishikeshbiz]

@levie Smarter scaffolding beats bigger parameters at scale. Routing efficiency is the new performance optimization

[rishikesh kubasad @rishikeshbiz]

@danshipper @WSJ @OpenAI Distribution + fast developer feedback loops = unstoppable momentum.

[rishikesh kubasad @rishikeshbiz]

@MediaKing Constraints breed leverage. Once you cap hours, saying "no" stops being a luxury and becomes an operational necessity.

[rishikesh kubasad @rishikeshbiz]

@matt_gray_ 100%. Social algorithms come and go, but an email list is direct distribution you actually own. Compounding consistency at its finest

[rishikesh kubasad @rishikeshbiz]

@gdb As inference costs crash, scheduled AI routines are replacing traditional cron jobs.

[rishikesh kubasad @rishikeshbiz]

@KevinSzabo14 onward and upward, hope I make it this time

[rishikesh kubasad @rishikeshbiz]

@StevenCravotta Investment precedes commitment. Getting users to articulate their pain point builds sunk cost and makes the paywall feel like a personalized solution rather than a toll booth. 30% trial rate is wild.

[rishikesh kubasad @rishikeshbiz]

@steipete We went from careful Prompt Engineering to just context-dumping in real time. The context window handling is absurdly robust now.