The Infinite Loop

The two most common questions i get are “what’s the secret to success in the vibe coding era” and “if anyone can build now how can you have a moat” I enjoyed talking about it with @abracarioca from @RunInfiniteLoop and trying to give my 2 cents

Every SaaS company is now deciding how to become infrastructure for AI agents — or get replaced by them. The Infinite Loop sat down with @nikitabase to find out what’s changing, what keeps him up at night, and what will follow the age of agents. infiniteloop.media/p/the-rise-of-… 4/4

In the agentic era, AI isn’t just changing how software is built. It may dismantle SaaS itself. Shamgunov, for one, is preparing for the SaaS apocalypse. You have to be “extremely paranoid,” he says. 3/4

99% of infrastructure traffic is coming from agents. 2 years ago it was 10%. Pretty much all the code being produced is generated by agents now, says Nikita Shamgunov @nikitabase, VP at @databricks and founder of @neondatabase (now Lakebase). 1/4

The timeline — and even the very concept — of AGI are contested. The decisions being made in anticipation of it are not. Full piece by @annaheim: infiniteloop.media/p/the-philosop… 4/4

Henry Shevlin @dioscuri spent his career asking them from inside a university. Now he is being paid by the company that thinks it might build the answer first. 3/4

Demis Hassabis thinks AGI arrives around 2030. The CEO of one of the world's leading AI labs is not waiting to find out. Google DeepMind just hired a philosopher. With that actual job title. 1/4

@async_platform cut latency to under 100ms by fixing the pipeline, not the model. Voice AI fails not where it speaks, but where it hands off. @BerryWrites examines the infrastructure making voice AI work: infiniteloop.media/p/voice-makes-… 🧵5/5

@Fujitsu_Global didn't fix their contact center by replacing human agents with AI. They automated the 6 minutes of admin work agents were doing after every call. That's where the time was going. Voice AI delivers most of its value in the handoffs, not the conversations. 🧵4/5

If you’ve talked to an AI agent in customer service lately, you know how it is. Satisfying when the conversation goes well. Infuriating when it stalls. But it’s actually not the AI model at fault. 🧵1/5

Real-time crime prevention used to mean more police. Now it means better AI infrastructure. Full piece by @timmpsmith: infiniteloop.media/p/meet-the-ai-… 🧵4/4

Augur deliberately avoided facial recognition. It’s not just an ethics call. It’s also the better engineering decision. 3D spatial tracking works more reliably in real-world conditions where facial recognition often simply fails. 🧵3/4

AI is now watching stadiums, power plants, and transport hubs to prevent crime in real time. But the latest crime stopping technology doesn’t use facial recognition. 🧵1/4

The catch: the compute is really expensive. AI output gets triple-checked for accuracy. Huge amounts of data. Compliance requirements. AI systems won’t save money, but they might make insurance less painful for customers. Full piece by @timmpsmith: infiniteloop.media/p/ai-insurance… 🧵5/5

“Humans aren't instant,” said @nico_laqua at Corgi Insurance @UseCorgi. That’s the real limitation. The AI-native insurer is building their own full-stack AI system for a better customer experience. 🧵4/5

A fraudulent insurance claim. A photo of a damaged vehicle. It’s an AI deepfake that’s almost impossible to detect. But insurance companies are getting ahead of fraud with their own AI technology. 🧵1/5