If #Cursor loses #OpenAI does it matter? #chatgpt #agenticengineering #vibecoding

Agentic Engineering · 2026-08-31

This video discusses the potential impact of Cursor losing its direct access to OpenAI models in November. Surprisingly, Cursor states that OpenAI models account for only 5% of its user traffic, raising questions about the true nature of its product. The key question posed is whether the product's value lies in the underlying AI model or in the 'harness' (the integrated development environment and its features) that leverages various models.

How to use AI to become smarter #AI #productivity #AItools #futureofwork #criticalthinking

Nate B Jones · 2026-08-31

This video emphasizes the need for senior software engineers to find practical payoffs for AI, moving beyond hype and new models. The speaker advocates for developing skills that enable rapid AI tool selection and efficient work, minimizing time spent on tasks based on incorrect assumptions. The core message is to focus on tangible value and productivity improvements through strategic AI adoption and learning.

They lied...

Theo - t3.gg · 2026-08-31

Anthropic announced a `permanent 25% increase` to weekly limits for Claude Code on September 14th, but this actually translates to a 17% reduction in effective usage because a temporary 50% increase was in place. This confusing communication style, coupled with past changes and restrictions around their Fable model (which is now getting its separate weekly limit removed), highlights Anthropic's prioritization of business needs over clear developer communication. The speaker argues this approach is counterproductive and compares it unfavorably to OpenAI's directness in similar situations.

read more

Anthropic's recent announcement regarding changes to Claude Code weekly limits has been met with significant criticism, primarily due to misleading communication and perceived anti-developer sentiment. The core issue revolves around a tweet from `ClaudeDevs` stating a `permanent 25% increase` to standard weekly limits starting September 14th. However, this statement is deceptive because a `temporary 50% increase` has been active, meaning the `permanent 25% increase` actually represents a net 17% reduction in weekly limits compared to the current promotional levels.

The speaker highlights a pattern of poor communication from Anthropic, noting that the original announcement was `deleted` and a new version was posted with slightly modified wording, but still failing to clearly explain the reduction. This lack of transparency is seen as a deliberate attempt to obscure a negative change for users. The speaker, a heavy user of Claude Code with multiple accounts (some of which regularly hit their limits), expresses deep familiarity with Anthropic's rate limits and offers a developer's perspective on the impact of these changes.

The history of Claude Code's rate limits further exacerbates user frustration. In May, Anthropic announced a `doubling of 5-hour rate limits` and the removal of `peak hour limit reductions`, along with `substantially raising API rate limits for Opus models`. However, this May announcement didn't address weekly limits, which were a more pressing concern for many users. The speaker notes that at this time, there was already an existing 50% increase in weekly limits, which was later extended until July 13th, then August 19th, and finally to the current September 14th date.

A significant point of contention is the handling of Fable 5, a highly-regarded model for coding. Fable 5 was initially released with broad access but was later restricted. Its eventual reintroduction came with a 50% limit on usage within premium plans, meaning users could only allocate half of their weekly allowance to Fable 5, regardless of their overall plan limits. This was a direct downgrade from previous access and a major pain point for users who valued Fable 5 for its quality. The speaker notes that this forced users to consume their Fable 5 allowance quickly, leaving them with remaining quota for less-preferred models like Opus and Sonnet, which the speaker openly criticizes for their performance in coding tasks. The announcement on September 14th states that the `separate weekly limit for Fable will be removed`, allowing users to utilize their full limit with their preferred Claude model. While this sounds positive, it’s only a return to a previous, superior state of affairs, coinciding with a net reduction in overall limits.

The speaker criticizes Anthropic for its communication strategy, describing it as `absolutely baffling` and `fumble after fumble`. They argue that Anthropic's messaging is often geared towards their business partners and investors, rather than their developer community. This is exemplified by the contrast with a hypothetical good-will announcement format, where bad news is presented empathetically and clearly. The speaker points to OpenAI's communication regarding its breakup with Cursor (a partner of Anthropic as well) as a much better example of transparent and empathetic communication, where OpenAI acknowledged the impact on developers and provided maximum notice. In stark contrast, Anthropic's announcements often `pretend bad things are good` and `refuse to admit fault`.

The underlying issue, according to the speaker, is compute availability and Anthropic's business decisions. The subscription models are seen as a `loss leader` that restricts compute from being available for more lucrative API sales. This creates a situation where individual users with subscriptions are consuming valuable compute that could be sold at higher rates to enterprise customers via the API, particularly during peak business hours. The speaker suggests Anthropic's consistent `anti-user stance` and misleading communication will ultimately harm their long-term position, especially if competitors like OpenAI continue to deliver high-quality models with better user experiences and communication. The speaker expresses a strong preference for Fable 5 but laments having to defend Anthropic's questionable business practices to continue using it.

Understanding ChatGPT Work

Simon Willison · 2026-08-30 · 10 min read

TLDR: ChatGPT Work (Cloud) is essentially a persistent, internet-connected agent environment layered on top of ChatGPT — the key differentiators over regular Chat are unrestricted code execution with full internet access, a headless Chrome browser for web automation, and a shared filesystem that persists across sessions. Combined with the ability to deploy live websites (via Cloudflare Workers) and run parallel sub-agents, it's less a chat upgrade and more a cloud automation platform. It's only available to $20/month+ subscribers and draws from a separate usage allowance tied to the Codex quota.

Model stays unpredictable how do you build reliable #AI #agenticengineering #claude #gpt #vibecoding

Agentic Engineering · 2026-08-30

This video, drawing from Andrew Ng's latest AI Engineering skill map, argues that the core task of AI engineering is not to make models inherently predictable, but to build reliable systems around models that will remain inherently unpredictable. This involves a shift from traditional software engineering's direct reasoning from code to behavior, to an iterative process of building, inspecting, identifying failure patterns, and adjusting context, architecture, or tools. The key takeaway is the importance of disciplined evals and error analysis to understand which failures matter and how to measure them, moving beyond superficial demo performance.

If you've been at the same company 3+ years, you're already in a box

Beyond Coding · 2026-08-30

This video emphasizes that discomfort is crucial for career growth, suggesting that feeling uncomfortable in your job about 30% of the time indicates you're learning and not just sticking to what you know. It highlights that every career move is a trade-off, where you exchange specialized knowledge for new learning opportunities. The speaker advocates for understanding that systemic issues, rather than individual stupidity, often drive problematic architectural decisions, encouraging motivation to change while acknowledging that the system might be rigged against individual efforts.

alchemy-utils 0.1a1

Simon Willison · 2026-08-13 · 1 min read

Simon Willison released alchemy-utils 0.1a1, a library that brings sqlite-utils-style database manipulation to multiple database backends by building on top of SQLAlchemy. This matters because it extends a popular, developer-friendly data wrangling API beyond SQLite to any SQLAlchemy-supported database, with notable performance improvements for DuckDB exports and CSV imports.