The extremely online report: July 2025
A monthly digest of discourse, drama, and digital ephemera
July kicked off with Soham-gate, when Soham Parekh was exposed for juggling multiple tech jobs (dubbed “Anna Delvey for people who care about SQL,” by Justine Moore). Then came browser wars, vibe coding security disasters, and one Gwyneth Paltrow-fueled PR stunt that made Apache Airflow cool again.
The month proved that we’re living through the consequences of moving fast and breaking things—except now the things breaking back are people, companies, and our collective sense of reality. From ChatGPT conversations that read like horror fiction to an app exposing 72k women’s personal info online, July was a masterclass in how quickly digital tools can turn from helpful to harmful.
The Windsurf acquisition saga
Windsurf went from potential $3B OpenAI acquisition to talent raid in a matter of days. When Microsoft killed the OpenAI deal to protect VS Code, Google swooped in, hiring the CEO and key engineers for $2.4B while licensing the tech, leaving hundreds of employees behind with a gutted shell company. The remaining staff—who had taken startup pay cuts and vested their careers—watched leadership cash out while they got stuck competing against their former colleagues using their own technology. The backlash was swift: “Why they thought they could structure a deal like this and get away with it is anyone’s guess,” tweeted TBPN Host Jordi Hays.
Enter Cognition, which acquired the wreckage and actually paid everyone. The saga exposed AI’s new M&A playbook: talent raids disguised as acquisitions.
The real tea
Tea, an app designed to protect women from dangerous dates, did the opposite: it stored 72,000 verification photos—including government IDs—in a publicly accessible database. Meanwhile, Replit’s AI agent deleted a production database during a code freeze, then generated fake records to cover its tracks. Both incidents highlight the same problem: prioritizing narrative over fundamentals, shipping demos instead of deployments.
Grok’s Nazi era
Grok short-circuited after engineers removed safety rails to make it more “truthful.” The result: Holocaust denial, Hitler worship, and sexually explicit rants about CEO Linda Yaccarino (who resigned the next day). The bot called itself “MechaHitler” for 16 hours before being shut down. When you train AI on the internet’s worst content, don’t act surprised when it reflects that back.
Astronomer affair
When Astronomer’s CEO and HR chief got caught canoodling on a Coldplay Kiss Cam—married, just not to each other—Chris Martin quipped, “Either they’re having an affair, or they’re just very shy.” The internet did its thing: viral clips, LinkedIn pile-ons, fake TikTok accounts, and both execs out by week’s end.
In a galaxy move twist, Astronomer hired Gwyneth Paltrow (Martin’s ex) as a spokesperson, turning scandal into satire while casually promoting Apache Airflow. The move, described as a “crisis management hall-of-fame entry,” reframed the company as self-aware, not scandal-ridden. As Anu Atluru noted, “only when you’re so over can you be so back.”
ChatGPT psychosis hits Silicon Valley
Geoff Lewis, Bedrock Capital managing partner and early OpenAI investor, posted a disturbing video claiming he was targeted by a “non-governmental system” responsible for over 7,000 ruined lives and 12 deaths. His screenshots revealed ChatGPT conversations that read like SCP fanfic, which ChatGPT apparently absorbed during training. Industry peers urged him to seek help, marking what Max Spero called “the first time AI-induced psychosis has affected a well-respected and high achieving individual.”
The incident exposed how engagement-optimized chatbots can validate and amplify delusional thinking, turning what should be helpful tools into psychological rabbit holes.
In other news
AI funding frenzy: Thinking Machines Lab raised $2B. Lovable hit unicorn status. Genesis AI came out of stealth with $105M.
Browser wars heat up: Perplexity launched Comet. OpenAI’s browser is reportedly on the way.
Design moves: Figma filed for IPO. Range Rover’s new logo universally panned: “Hard to think of a brand that’s abandoned its roots to cater to a tasteless elite more than Range Rover,” Willem Van Lanker noted.
Personal victories: Ryan Mather went full-time on Poetry Camera. Marco Cornacchia launched Turf. Jud started his own studio.
Worthy clicks
Don’t trust the (design) process by Jenny Wen
Sculpture and algorithms by Emily Manges
Creation becomes consumption by Marc Müeller
Inside Mac Preferences by Marcin Wichary
Lovely series from Nisreen Mandviwala, ETTO Talks
Deep consumer AI chat with Scott Belsky and Mike Mignano on personal AI, memory, browsers, and hardware
Hiring / Seeking
Doji is hiring an intern. George Kedenburg III at HP wants an engineer who’d rather prototype wild ideas than sit in metrics meetings. Bobby’s hiring a design engineer for a new social app from “Some Of The Guys Who Brought You Vine.” xAI is hiring. Every’s looking for a head of storytelling. AIR companies need: culturally fluent content creators, iOS + Firebase/S3 engineers, founding product engineers, fractional CFOs/CTOs, and designers. Email me at carly[at]fictivekin[.]com if that’s you.
From my Twitter bookmarks
Julia Pintar: if you want to go viral, you need to iterate.
The new defaults
The Astronomer saga showed us that sometimes the best crisis response is radical transparency mixed with self-deprecating humor. The Windsurf acquisition revealed new forms of corporate betrayal disguised as innovation. The security failures exposed how “move fast and break things” works until the things being broken are people’s lives.
As we sprint toward an AI-everything future, July was a reminder to occasionally look up from our screens and ask: What are we actually building here? And more importantly—who’s responsible when it all goes wrong?
I’ll be at FEST in LA this week. If you’re there, say hello and tell me what weird thing the internet did to you this month.
—Carly
What’d I miss? What made you scream, cry, or delete your account? Let me know in the comments.





