Benedict Evans used a Threads post to take aim at ChatGPT’s increasingly crowded work interface, questioning why the app now asks users to distinguish among projects, tasks and chats while treating those modes differently on screen.
Evans’s post, which Threads showed at 15.2K views, focused on basic product mechanics rather than model quality. He asked why chats appear in a floating window while tasks and projects do not, why selecting plugins appears to surface templates, and whether users can complete setup without connecting Slack or Google Drive.
That is the kind of complaint that lands because it is about the part of AI software users touch every day: the wrapper around the model. A strong model does not make a muddled interface coherent. If a user cannot tell whether they are starting a chat, creating a task, opening a project or configuring an integration, the software has made its own taxonomy the user’s problem.
The thread became a small pile-on from people who said OpenAI’s product structure is getting harder to read. One commenter, gmitt98, argued that this is what can happen when an engineering-led company decides product management and design can be collapsed into one role. Another, nomad.mike, framed it as “vibe-product-managing,” a jab at the broader AI-era habit of shipping fast and asking the information architecture to clean itself up later.
Several replies focused on the way ChatGPT appears to prioritize tasks over ordinary chats. User derickmoncado.jpg said chat should be a top-level option rather than feeling like an afterthought, and said previous projects still appeared in the web sidebar while being hard to find elsewhere. User edu723 described the emphasis on tasks over chats as a growth tactic they hoped would be reversed.
Other complaints were about platform and onboarding behavior. User dmccullum said they had returned from Claude because ChatGPT’s native Mac app had been better, only to see it replaced by what they described as an Electron app. User torusdev said a Slack login browser window appeared during onboarding without clear prompting or indication that it came from Codex.
Codex also came up as a sore point. User barefootinharvard said Codex had been one of OpenAI’s products that felt as if someone working with access needs had influenced it, then said it was gone. The thread does not include an explanation from OpenAI for the interface choices Evans and commenters criticized.
The criticism is narrower than the usual “AI is overhyped” argument. The commenters were largely accepting that the models are useful. Their complaint was that the surrounding product now mixes chat, task management, project organization, templates, plugins and third-party integrations without making the boundaries obvious. That is how a chat app starts behaving like a would-be super app, and how users end up debugging the menu system before they can ask the model anything.
This story draws on original reporting from Daring Fireball.
