AI Clambake — Week of April 27, 2026

AI tools like ChatGPT and Claude often produce homogenous responses, which hurts productivity, which is the cough-cough main reason to use AI. The article cites a number of ways creatives are attempting to address the issue. Some of them sound interesting, and pleasingly weird (e.g. an Ursula K. Le Guin bot), if not effective.

OpenAI is giving Business, Enterprise, Edu, and Teachers plans access to new cloud-based ‘workspace agents’ in ChatGPT. These agents can supposedly do things like find product feedback and send a report to Slack, or draft follow-up emails in Gmail. The launch follows competition from rivals like Claude Cowork. OpenAI describes the new agents as an ‘evolution’ of its custom GPTs and will eventually provide a way to convert GPTs into agents.

OpenAI has released ChatGPT Images 2.0, an update to its AI image generator. For subscribers to the Plus, Pro, Business, and Enterprise plans, the tool can now search the web for information, generate up to eight images in a series while maintaining character and style consistency, and create visual explainers from uploaded files. Updates for all users include image generation up to 2K resolution and more aspect ratios.

Heygen recently released Hyperframes, “an open-source video rendering framework that lets you create, preview, and render HTML-based video compositions” with support for AI agents. Reviews on Reddit suggest it’s easier to use and creates better vids than Remotion, but it may also be more token-hungry.

Humwork claims to have created a marketplace where developers can find experts to solve problems their AI agents can’t solve. You’re using Claude Code to build some freaky engineering or medical product, it chokes, you send it to an expert, expert fixes it in your session, then sends it back. No idea if it works as advertised but interesting insight into the new kinds of businesses and services AI will generate.

Open AI recently released a biology-specific LLM named Rosalind Franklin trained on jargon and hooked up to databases relevant to that field. OpenAI claims that that model has been tuned to avoid LLM’s sycophancy and overenthusiasm. The LLM could theoretically help identify new treatments.

Tim Davis is the president and cofounder of Modular. He wrote an interesting piece about what the rise of coding agents means (more coding mistakes). Also about the impact of AI on the daily routines and careers of coders. I think the trends he mentions will likely apply to many white collar workers.