A new MIT study suggests that AI coding tools can dramatically increase code volume, but far less of that output survives the full production funnel. According to the shared summary of the study, autonomous coding agents increased commits by 180%, while releases rose only 30%. Chamath Palihapitiya added the operator version of the lesson: coding with AI without clear intent upfront is how AI slop starts.

That is the real warning for the coding-agent wave. The scarce work is no longer typing more code. It is deciding what should exist, checking it, connecting it to the product, shipping it, and seeing whether users actually adopt it. AI can multiply output fast; the bottleneck moves to product judgment, review gates, and release discipline.

The Business Times reports that NVIDIA CEO Jensen Huang called the tech-stock sell-off a buying opportunity. He said the industry is still in the early stages of building the infrastructure for an AI-fueled future, and that a market pullback gives investors a chance to buy at a discount.

The surrounding context matters too: on the same day, NVIDIA and SK Hynix announced a multi-year agreement to design future generations of memory chips for AI. The point is not that the market will automatically follow NVIDIA. It is the gap between public-market fatigue and infrastructure conviction. While investors worry that the AI trade is overheating, NVIDIA is still framing AI infrastructure as an early buildout, closer to how the Internet became global infrastructure.

Apple's June 8 WWDC is expected to center on a Siri overhaul, including chat mode, personal context, developer extensions, and possible model choices from OpenAI, Anthropic, or Gemini. After years of Siri being seen as a lagging assistant, Apple needs to show that AI can live inside the OS and app ecosystem, not just inside a standalone chatbot.

If Apple pulls this off, consumer AI distribution could shift from chat apps into the workflows already sitting on the device. Siri would not just answer better. It would need to become the context-aware layer that helps users act across apps, preserves Apple's privacy posture, and makes AI feel useful in everyday device behavior.

Boris Cherny shares how to run Opus autonomously for hours or days: Boris Cherny says Opus is strong at long-running work and recommends auto mode for permissions, dynamic workflows, /goal or /loop, Claude Code in the cloud, and end-to-end self-verification through web, simulator, server, or service access. Autonomy in coding agents is becoming a permissions, orchestration, persistence, and verification problem.

Peter Steinberger says builders should design loops for agents: Peter Steinberger says builders should design loops that prompt coding agents instead of prompting agents directly. It is a short line, but it captures the direction clearly: agent work is starting to look less like prompt craft and more like control-system design.

US startups are routing more traffic to Chinese LLMs through OpenRouter: Rohan Paul says OpenRouter token-consumption data shows Chinese models becoming a larger growth engine for US AI startup app traffic from early 2026. If that pattern holds, the model market is moving away from brand loyalty and toward raw utility: whichever model does the job well, cheaply, and reliably gets routed into the product.

Aaron Levie says AI makes building easier, but enterprise GTM remains hard: Aaron Levie argues that AI has made software development easier, but it has not removed the hard parts of enterprise software: quality, security, differentiation, sales, support, implementation, and integration. When software gets cheaper and more abundant, distribution, trust, and GTM can become the bigger bottlenecks.

Notion temporarily disabled Anthropic models after a service disruption: TechCrunch reports that Notion temporarily disabled all Anthropic models after degraded performance in Opus 4.7 and 4.8, then restored access after Anthropic resolved the infrastructure issue. For AI-native productivity products, model-vendor reliability has become product reliability.

NVIDIA and LG are building an AI factory for physical AI: NVIDIA says it and LG Group are building an AI factory collaboration for physical AI, mobility, and AI infrastructure. AI factories are moving beyond cloud training into robotics, manufacturing, mobility, and real-world deployment loops.

Suraj Sharma points to OpenAI's Codex use-cases page, which includes workflows like drafting email, reviewing PRs, turning Figma designs into code, understanding large codebases, triaging bugs, querying spreadsheets, deploying apps from prompts, building Mac and iOS apps, creating slide decks, turning Slack threads into coding tasks, and using computer actions.

Today's practical move: do not read that list as a feature catalog. Turn it into a delegation menu for your team or for yourself. In 30 minutes, you can build a small table that tells you which tasks should go to Codex, which tasks Codex should only prepare, and which tasks still need a human gate before shipping.

Steps:

How to read the result:

Managing your cap table doesn’t have to be complex. Pulley simplifies equity management for Founders and CFOs with intuitive workflows, accurate, audit-ready reporting, and predictable pricing—so you can plan and scale without surprises. Onboard in days, not weeks, and rely on responsive, expert support every step of the way.  

From issuing grants to 409A valuations or ASC 718 reporting, Pulley gives you the clarity to manage equity, make decisions, and get back to work. Experience a platform built for business owners and finance teams: transparent, reliable, and designed to put your company first.

See Pulley in action

Lenny Rachitsky shared a conversation with Tony Fadell on opinion-based decisions for v1 products, why marketing matters as much as product, why taste may be the biggest moat in AI, and the risk of cognitive surrender when builders hand too much judgment to AI.

That lens fits the opening story well. If AI can increase production, then taste, intent, and the ability to refuse outsourced judgment may become what separates strong builders from average ones.

Microsoft argues that AI alone will not transform a business; the value sits in the system around agents: tools, actions, evals, traces, and interoperability. It is a useful lens for the whole issue.

A strong agent is not enough. Operators need systems for assigning work, tracking progress, checking outputs, limiting permissions, and connecting agents to real workflows. That is where AI output gets a chance to become shipped value.