Our site remains on Vercel despite push to Cloudflare, highlighting real-world tradeoffs in migrating hosting platforms.

Tom shares how Claude was tasked to fix his failing Vercel build, demonstrating AI’s role in real-time debugging and CI pipeline recovery.

Tom recounts an amusing conflict with the cursor agent that repeatedly downgraded his GPT-5 reference to GPT-4, despite his code comments insisting ot...Show more

Using Claude Code for code generation became expensive quickly, leading to caution in API usage for prototyping AI apps.

Cameron uses the decline of horseshoes and blacksmiths during industrialization as an anecdote to illustrate how technology displaces jobs but eventua...Show more

Klarna implemented AI in customer support and eliminated $40 million in operating expenses almost overnight, reducing support staff by 80%.

They recruited industry experts with a minimum of four years’ experience (average 14), requiring video interviews, background checks, training, and qu...Show more

Experienced professionals often face 20 rounds of back-and-forth with ChatGPT because they need to incrementally enrich context the model otherwise la...Show more

Tom Spencer observed that new graduates often write verbose ‘perfect’ emails which then go unread, highlighting the need for concise communication des...Show more

Cameron shared his experience of crafting an AI strategy for a primarily non-tech company and the need to make AI relatable and credible to non-techni...Show more

Tom recounts evolving the Groq desktop app by incrementally adding features based on hands-on usage feedback, illustrating an agile improvement cycle.

Three months ago I recommended rendering AI scenes down to React Three Fiber code for composability, but world models have since made this redundant.

Cameron created a custom car chase video on Ray models for a friend who visualizes himself racing through a city whenever he plays his favorite song.

Tom Spencer reflected on ‘eating his words’ live when testing Claude’s browser capabilities, highlighting the value of real-time experimentation.

When testing the BrowserComp benchmark prompt about a 1960s fictional character, many advanced models like Claude 4.5 and GPT-4.5 failed before the br...Show more

Tom had to disable LastPass because the Claude Chrome extension required direct credential input, highlighting extension compatibility challenges.

During a demo of Research Pro, Cameron envisioned that in 12 months developers would create streamable UI components, predating tools like Vercel AI S...Show more

Using screenshot-based UI automation can break if the user moves windows around, highlighting the need for more robust visual context strategies in ag...Show more

A rogue prompt hidden in a webpage's DOM could trick AI browser agents into unauthorized transactions, highlighting the need for robust agent security...Show more

Users cannot review previous prompts or code iterations in Claude Imagine because the conversation history is cleared on each update, emphasizing a li...Show more