GPT-5 can now code for 7 hours straight
📖6 min read
Hey Snackers,
OpenAI just dropped GPT-5-Codex, a coding model that can now "think" about your broken code for up to seven hours straight - which is approximately 6 hours and 59 minutes longer than most human developers before they rage-quit and blame the compiler. The new version is rolling out to developers who presumably have the patience to wait for an AI to contemplate their spaghetti code longer than it takes to fly from New York to London. Meanwhile, Alphabet casually joined the $3 trillion market cap club yesterday, riding high on AI hype and a court ruling that basically said "yeah, you're a monopoly, but the stock market doesn't care." (source)
Speaking of trillion-dollar companies doing trillion-dollar things, Google launched something called the Agent Payments Protocol (AP2) - essentially teaching AI how to spend your money for you. Because what could possibly go wrong with giving a hallucination-prone language model access to your credit card? (source)
SILICON VALLEY HUNGER GAMES
Fiverr fires 30% of humans to become "AI-first"
In what might be the most on-the-nose example of AI eating jobs since the invention of the self-checkout kiosk, freelance marketplace Fiverr just laid off 250 employees - that's 30% of its entire workforce - in a bid to become an "AI-first company." (source)
CEO Micha Kaufman delivered the news with the kind of corporate speak that would make ChatGPT proud, explaining the company needs to be "leaner and faster" - which is executive-speak for "we're replacing you with Python scripts."
The irony here is delicious: a platform built on connecting human freelancers with clients is now firing its own humans to automate everything with AI. It's like watching a snake eat its own tail, except the snake has a Stanford MBA and stock options.
- The company claims this will help them "leverage AI to automate systems" - translation: they're building bots to manage the bots that will eventually replace the freelancers.
- This comes as Fiverr's stock has been struggling, down significantly from its pandemic highs when everyone thought they'd become a TikTok influencer.
- The remaining 70% of employees are presumably updating their LinkedIn profiles with "AI prompt engineer" as we speak. (source)
THE TAKEAWAY
Fiverr's move is the canary in the coal mine for the gig economy. If a company whose entire business model depends on human creativity and skills is going all-in on AI, what does that say about everyone else? We're speedrunning towards a future where the only jobs left will be "AI whisperer" and "person who turns the AI off and on again when it breaks."
NOTHING VENTURED, NOTHING GAINED
Smartphone startup raises $200M to build an "AI-first" device nobody asked for
Remember Nothing? The smartphone company founded by OnePlus co-founder Carl Pei that makes phones with blinking lights on the back? Well, they just convinced investors to hand over $200 million at a $1.3 billion valuation to build an "AI-first device." (source)
The company plans to launch this mysterious AI gadget next year, promising an operating system with "deep AI integration" - which sounds suspiciously like every other smartphone released in 2025, except this one will probably have RGB lighting.
- This comes after the spectacular failure of Humane's AI Pin, which proved that consumers don't actually want to wear a $700 AI brooch that projects a laser onto their palm.
- Nothing's current phones are basically Android devices with a transparent back and some LEDs - revolutionary stuff.
- The company claims they're building something "fundamentally different" but won't say what, which is the hardware equivalent of "trust me bro."
Investors are apparently still drunk on AI hopium, willing to throw money at anything with "AI-first" in the pitch deck. At this rate, someone could raise a Series A for an AI-powered paperclip. (source)
THE TAKEAWAY
The fact that Nothing can raise $200 million for vaporware while actual AI companies struggle to find product-market fit tells you everything about the current state of tech investing. We're in the "throw money at anything with AI in the name" phase of the bubble, which historically ends well for exactly nobody except the founders who cash out early.
CORPORATE SHOPPING SPREE
Workday drops $1.1B on AI firm because buying innovation is easier than building it
HR software giant Workday just announced it's acquiring Sana, an AI automation company, for $1.1 billion - proving once again that when big tech companies need innovation, they just open their checkbooks.
Sana builds "AI agents to automate tasks," which is the most generic description possible but apparently worth a billion dollars in 2025. The acquisition is part of a feeding frenzy where every enterprise software company is desperately buying AI startups to sprinkle some machine learning fairy dust on their decades-old codebases.
- This follows similar acquisitions by SAP, Oracle, and basically every other enterprise software dinosaur trying to stay relevant.
- Workday's stock barely moved on the news, suggesting investors are as excited about this as employees are about mandatory HR training.
- The real winners here are Sana's VCs, who turned a few million into a billion faster than you can say "synergistic AI transformation." (source)
THE TAKEAWAY
The enterprise software M&A market has become a game of musical chairs where everyone's scrambling to buy an AI company before the music stops. At this rate, there won't be any AI startups left to acquire by 2026 - just a bunch of bloated enterprise companies with hastily integrated chatbots calling themselves "AI-native."
THE BEST THING WE READ TODAY
OpenAI accidentally reveals ChatGPT is mostly used for cheating on homework
OpenAI released a massive study on how people actually use ChatGPT, and surprise surprise - it's not for building AGI or solving climate change. Most people use it for "getting information" (aka Googling with extra steps) and about 30% of consumer use is work-related (aka making their boss think they're productive).
The study found that the early gender gap has "nearly closed," which is corporate speak for "everyone's equally lazy now." The most telling stat? People are using it for "everyday practical tasks" - translation: nobody wants to write their own emails anymore.
What's fascinating is that this comprehensive look at ChatGPT usage confirms what we all suspected: AI isn't revolutionizing human intelligence; it's just making it easier to fake having any. (source)
What else we're Snackin'
- YouTube is giving Shorts creators access to Google's Veo 3 text-to-video model because apparently humans making content is so 2024 (source)
- YouTube also launched AI tools for podcasters to create viral clips, solving a problem nobody knew existed (source)
- Anthropic's new report shows AI adoption is concentrated in rich countries, shocking absolutely no one (source)
- YC-backed Rulebase raised $2.1 million to build an "AI coworker" for fintech compliance, the most boring use of AI yet (source)
- "The Morning Show" is tackling deepfakes this season, with Jennifer Aniston and Reese Witherspoon playing themselves but concerned about AI (source)
- The Agent Payments Protocol has Mastercard and PayPal on board, because teaching AI to spend money is definitely what humanity needs right now (source)
- Studies show 77% of business AI use is for automation rather than collaboration, confirming that companies prefer robots to humans (source)