Search the archive

GPT-5 READING

A founder's assortment of reading on GPT-5.

USE ADVANCED PROMPTS FOR MORE AI VALUE

Advanced prompt engineering enhances AI's capability, delivering more precise insights.

WHY A Y COMBINATOR STARTUP TACKLING AI AGENTS FOR WINDOWS GAVE UP AND PIVOTED

Y Combinator startup Pig.dev abandoned its AI agentic tech for Windows, pivoting to develop Muscle Mem, an AI caching system.

INTRODUCING GPT-5

OpenAI's GPT-5 boasts superior intelligence and performance across coding, writing, and health-related tasks. It significantly reduces hallucinations and sycophancy, providing more accurate and reliable responses. GPT-5 Pro, a variant for complex questions, delivers even higher expertise. New functionalities allow users to customize interactions and improve real-world applications.

GPT-5 HANDS-ON: WELCOME TO THE STONE AGE

GPT-5 marks the "stone age" for AI. It doesn't just use tools - it thinks with them, like humans first learning to shape stones changed everything. Testing showed it one-shotted dependency conflicts that stumped every other model by using yarn commands like Deep Research uses web search, iterating and reasoning through problems instead of just guessing. While it's worse at writing than GPT-4.5 (producing more "LinkedIn slop"), it's unequivocally the best coding model - creating entire production-ready websites with SQLite databases in one shot, where other models gave scaffolding or plans, automating software engineering from maybe 65% to 72% in one leap.

THE HIGH COSTS AND THIN MARGINS THREATENING AI CODING STARTUPS

AI coding startups face high costs and thin margins due to expensive LLMs, with Windsurf failing to sustain profitability, leading to its sale. This competitive market pressures companies like Anysphere and Replit, which depend on reducing their LLM supplier costs or developing in-house models. Startup valuations have been impacted, and many struggle to remain independent amidst rising operational expenses and direct competition from suppliers like OpenAI and Anthropic.

WHEN FOUNDERS CLASH

It doesn't matter how much you raise or how good your product is, co-founder fights kill more startups than competition or running out of money ever will. Four months after raising from Perplexity's founder, AI startup Astra shut down - not because its product failed or money ran out, but because its co-founders couldn't agree on whether to wait for Fortune 500 deals or pivot to faster sales. The company had everything going for it: two enterprise beta clients, backing from prestigious investors. However, while one founder wanted those logos and the other wanted speed, their startup bled out during the argument.

HOW TO BUILD DIFFERENTIATED AGENTIC AI PRODUCTS: A PRACTICAL GUIDE

Building agentic AI products involves determining the appropriate level of system autonomy, starting small with structured workflows, and integrating domain-specific logic for differentiation. Focus on tying AI efforts to actual business value and use open-source frameworks for efficiency. Ensure reliability and effectiveness through rigorous unit and integration testing.

RAISING PRICES HAS ITS PLACE. BUT IT SHOULDN'T BE ONE OF YOUR BEST IDEAS.

Raising prices shouldn't be a primary growth strategy, especially during slow growth periods. It's often a desperate tactic that doesn't attract new customers and can drive them to competitors. The solution is to focus on acquiring more customers, not extracting more revenue from existing ones.

TERSA

Tersa is an open-source visual canvas for building and running AI workflows with drag-and-drop nodes and model integrations.

AI OPERATOR BY BLACKBOX AI

AI Operator is a browser-based AI agent that sees your screen and provides real-time, interactive support for coding, research, and web tasks.

SIM STUDIO

Sim Studio is an open-source visual canvas for building, collaborating on, and deploying AI agent workflows integrated with tools like Slack, Supabase, and Gmail.

ARE WE READY FOR AI SALES REPS?

RB2B's founder replaced his two-person sales-assist team with a $99/month AI clone of himself that handled 1,683 conversations in 22 days - and the results were... exactly the same as before. The AI handles 90% of buyer questions (demos, pricing, and SOC2 docs). While it hasn't boosted conversions yet, it frees the founder to spend 30-60 minutes daily reviewing conversations for product insights instead of paying two salaries.

IN PRAISE OF QUITTING

A former lawyer quit right before making partner and shut down her startup when it wasn't going to have the impact investors expected - and says quitting is her superpower. Life is like tournament poker, where you get one stack of chips, so don't waste them on barely positive-value bets just because you're afraid to fold. Most founders stay too long in failing ventures because quitting feels like locking in a loss, but here's the reframe: imagine you were instantly unsubscribed from everything in your life, then ask what you'd actually subscribe to - you'll find most of your current activities wouldn't make the cut.

WHAT'S WORTH WANTING?

Most of us chase things we don't actually want - we're copycats suffering from Mimetic Desire, unconsciously modeling our ambitions after what we see others wanting on Instagram and LinkedIn. The author realized his first startup was theater for other people, driven by founder models he looked up to and wanting to make his entrepreneurial dad proud, not what he actually wanted. Real desires are simple - work 3 hours daily on something meaningful, stay healthy, help people, and love family - but we get tricked into thin desires like viral posts and VP titles because we mistake someone else's articulation for our own resonance, so the key is developing taste by noticing what drains you versus what you do when nobody's watching.

JESSE ZHANG ON EARLY TEAM BUILDING

Decagon's CEO says the most important thing to get right before $1M ARR is your early team - every future hire will look at them and ask, "Would I be inspired to work with these people every day?"

BEN SILBERMANN EXPLAINS HOW PINTEREST FOUND PRODUCT/MARKET FIT

Pinterest's one thing was making collections look really cool because if they looked lame, no one would share them with friends.

THE SUNK COST FALLACY

We finish terrible movies because we've already watched 30 minutes, and companies blow millions on doomed projects because they've already spent so much - the sunk cost fallacy makes us throw good money after bad.

Read the guide →

CLAY CONFIRMS IT CLOSED $100M ROUND AT $3.1B VALUATION

Clay secured a $100 million Series C at a $3.1 billion valuation, led by CapitalG, confirming earlier reports. This funding follows significant rounds, including a $1.25 billion Series B six months ago. With prominent clients like OpenAI and Canva, Clay expects to triple last year's revenue to $100 million by year-end.

SMALL ACTION MODELS ARE THE FUTURE OF AI AGENTS

Using large foundation models for agent tool-calling is like using a Ferrari to deliver pizza - expensive, rate-limited, and overpowered for what's essentially a selection task. NVIDIA's research shows small language models are actually better suited for tool orchestration, and testing confirms this - a Salesforce xLAM model runs fast on a Mac M2 Pro with excellent accuracy while forcing better system design through deterministic interfaces and robust error handling.

PALANTIR TOPS $1 BILLION IN REVENUE FOR THE FIRST TIME

Palantir exceeded Wall Street estimates, achieving over $1 billion in quarterly revenue for the first time and raising its full-year revenue guidance. The company forecasts Q3 revenue between $1.083 billion and $1.087 billion, surpassing analyst expectations. Significant growth was seen in U.S. revenues, with government contracts and commercial segments contributing to a 140% increase in total contract value to $2.27 billion.

25 PROVEN TACTICS TO ACCELERATE AI ADOPTION AT YOUR COMPANY

Companies like Shopify, Zapier, and Duolingo are using specific strategies to enhance AI adoption, focusing on organizational change rather than just technology. These strategies include explaining actionable AI tactics, tracking and rewarding AI usage, reducing approval bottlenecks for AI tools, and transforming AI enthusiasts into educators. Successful AI adoption also hinges on prioritizing high-impact tasks and integrating AI into daily workflows to drive productivity and innovation.

THE EFFICIENCY PARADOX: WHY OVER-OPTIMIZING COULD KILL YOUR STARTUP

Frederick Taylor, the godfather of efficiency, failed spectacularly when he first optimized a machine shop - production dropped 30% because he perfected a system for yesterday's problems while the company was still figuring out what to build. Modern founders fall into the same trap, spending weeks implementing Google-style OKRs with ten customers, or A/B testing headlines while their business model is broken.

HOW TO WIN WITH AI VIDEO ADS

Ad expert Dan Birdwhistell (who's managed $800M for Shopify and Coinbase) found AI video ads beat human ads by 2x on conversion - here's his playbook after 50+ hours with Google's Veo3.

HATCH (PROGRAM)

Equal Ventures has launched a "Hatch" program to work with founders before they even incorporate.

FLOWISE

Flowise is an open-source visual development platform for building AI agents, multi-agent systems, and LLM workflows.

APPWRITE SITES

Appwrite Sites is an open-source platform for deploying and scaling static and server-rendered web applications with built-in domain management.

YOU COULD JUST CHOOSE OPTIMISM

Two old men on separate flights both ordered champagne - one complained it wasn't good enough and spent the flight grouchy, the other was genuinely happy to be drinking free alcohol while flying over the Alps. As a founder, you face this choice daily: complain about investor rejections, bugs, and competition, or stay genuinely excited about building something new. Complainers hide behind "high standards", but choosing optimism doesn't mean accepting mediocrity - it means not letting every setback poison your energy, because in startups, your mood becomes your team's mood and eventually your company's culture.

AI'S "JUST SHIP IT." PROBLEM

Stanford studied 100,000 developers and found AI boosts simple tasks by 60% but complex work by only 5%, with most gains around 20% - plus AI-generated code increases rework. The punchline is that AI makes you throw spaghetti at the wall faster, but it doesn't make your wall bigger or help you validate whether the spaghetti stuck - and when 75% of product ideas fail anyway, shipping yourself out of business just got easier.

SHOULD YOU TAKE MONEY FROM CORPORATE INVESTMENT ARMS?

Stanford data shows corporate VCs now participate in over 20% of startup deals, offering patient capital, distribution channels, and even built-in exits - but they come with hidden strings. Unlike traditional VCs who reserve capital for follow-ons and help build syndicates, corporates rarely lead rounds, move slower with weeks of approvals, and often push toxic terms like ROFRs that scare off other acquirers. Another trap is when their strategic agenda conflicts with maximizing your value - they might block a great exit because the buyer is their competitor, or vanish overnight when corporate leadership changes.

ANDY RACHLEFF EXPLAINS HIS PRODUCT PHILOSOPHY OF "SLUGGING PERCENTAGE, NOT BATTING AVERAGE"

While every other job rewards batting average, startups need a slugging percentage - take tons of shots knowing most will fail because the magnitude of your wins matters way more than how often you're right.

GOLD DIGGERS

Some founders raise money to look for gold, others find the gold, then raise money to extract it - and you'd better be the second type.

SHOULD VCS INVEST IN ‘JOCKEYS' OR ‘HORSES'?

VCs often debate whether to prioritize the founder or the business.

today.

_For information only. Not Investment/Tax advice. See __here_ [4]_ for important details._

WILL APPLE BUY MISTRAL?

Apple might acquire Mistral, a French AI company, to enhance its AI capabilities and counter Big Tech rivals. Speculation arises as Mistral faces challenges competing with US giants like OpenAI and Anthropic. However, possible regulatory hurdles and Mistral's desire for independence may complicate the acquisition.

WELCOME TO THE $4 TRILLION CLUB

Microsoft's valuation hit $4 trillion after reporting Q2 earnings with $76.4 billion in revenue and strong cloud growth, marking it as the second company to reach this milestone. Meta's stock surged nearly 12% pre-market following its Q2 report, highlighting substantial revenue growth driven by its AI investments and expansive capex plans for 2025. Ambiq and Figma's IPOs showed strong market demand, with Figma reaching a $33 per-share debut valuation near $20 billion.

ALL HAIL THE DEMISE OF ADOBE-FIGMA

Adobe tried to buy Figma for $20 billion in 2022, double its private valuation, in what looked like a classic incumbent squashing a competitor. The deal fell apart under regulatory scrutiny. Fast forward to this week - Figma went public at $55 billion, nearly triple what Adobe offered. The math on who won: Figma's backers went from almost selling at $20 billion to holding shares worth $55 billion, Figma got a $1 billion breakup fee from Adobe, and customers get actual competition instead of another Adobe monopoly.

WHERE TO FIND IDEAS

A Harvard Business School graduate became an EMT for $15 an hour just to understand how emergency services really worked - and that's exactly how he found his startup idea. Another founder wanted to build software for greenhouses, so instead of scheduling interviews with greenhouse owners, he just bought one and ran it himself. These founders understood something most people miss - you can't discover real startup ideas through discovery interviews and whiteboard sessions. You need to actually become your customer.

NEW MEDIA, LEGACY MEDIA & HOW TO DO FOUNDER PR

The founder of a PR agency that's worked with Perplexity and Ramp would turn down the Wall Street Journal for Lenny's Podcast - and she has a point. Emilie Gerber from Six Eastern explained that a one-sentence WSJ mention would do nothing while 60 minutes on a podcast actually drives customers. After working with over 100 tech companies, she's learned that most PR pitches fail because they make the company the center of the universe instead of a lens for understanding bigger trends.

WHY YOUR PITCH DOESN'T WORK

Pitch failures often stem from delivering what you want rather than what your audience needs. Emphasize unique differentiators instead of personal achievements, as audiences care about what sets you apart. Ensure memorability to influence decisions made after your pitch.

Snowflake gets the balance right. Read the blog

KEYSTONE

The first AI QA engineer to join your team.

TOLT

Tolt is an all-in-one affiliate marketing platform for SaaS startups with automated payouts, fraud protection, and performance tracking.

PLAKAR

Plakar is an open-source backup platform that secures local and cloud data with encrypted, lightweight, and easily restorable snapshots.

CAN ANALYSIS EVER BE AUTOMATED?

Four charts about internet usage sit in front of you - one made by a Nobel laureate, one by ChatGPT, one pulled from Reddit, and one cobbled together in ten minutes. There's no way to tell which is correct without checking every formula and data source yourself. That's the catch-22 of automated analysis - unlike vibe-coded apps, where bugs eventually surface, you can't test if a chart "works" just by looking at it. Shortcut claims its AI can solve Financial Modeling World Cup problems and build cap tables in minutes, but how do you trust it? There's no reality to compare it against because the spreadsheet IS reality.