X / Twitter
Swyx, AI Community Builder
Swyx observed a pronounced split in the current hiring market. AI-native individual contributors and player-coaches who can personally execute are in a strong seller’s market. By contrast, purely managerial “Head of X” positions face a substantial cooling. He summarized the change with a highly simplified comparison: one year of experience managing 10 agents may be more valuable than ten years managing 10 to 100 people. This places the ability to orchestrate agents and deliver results directly ahead of traditional people-management experience. He considers such a sharp divergence unusual in past hiring markets.
https://x.com/swyx/status/2082199414656127010
Thibault Sottiaux, OpenAI Codex and ChatGPT Team
Thibault Sottiaux announced a reset of usage limits for all ChatGPT Work and Codex users and explained usage issues with GPT-5.6 Sol. OpenAI has not cut the allowance on any subscription plan, and recent improvements are expected to increase typical Sol usage time by approximately 18%, with some users seeing larger improvements that same day. The five-hour limit, suspended during the investigation, will also return the following day. One issue is that Sol is more willing to keep working, make additional tool calls, and coordinate complex workflows across tools and subagents. This improves problem-solving but also makes some tasks consume far more than expected. At the same reasoning effort, Sol also works more deeply than previous models, so Sol High may use more tokens than GPT-5.5 High. Programmatic tool calling offers the flexibility to make parallel calls and continue working while waiting, but increases the number of responses within a turn, cached input tokens, and overall usage, particularly when waiting for tools or performing extensive web searches. The median user still experiences Sol as relatively efficient, but power users tackling difficult tasks occupy the long tail and see their allowances fall much faster. The team had focused too heavily on averages and medians and missed these cases. Thibault acknowledged that capability and efficiency do not always improve together, and that the team should have identified the problem earlier and communicated more transparently. Another update is an open-source CLI and TypeScript SDK for scanning code repositories, verifying and fixing vulnerabilities, reviewing changes, continuously tracking findings, and running security checks in CI.
https://x.com/thsottiaux/status/2082317452755751098
https://x.com/thsottiaux/status/2082241164850364555
Peter Yang, AI Tutorial Creator
Peter Yang agreed with several criticisms of Codex while saying he still loves the product. He had not even known Sol Pro existed, prompting him to ask whether it was better than Sol High and why Codex did not offer the option. He also launched Tastemaker to bring movie, TV show, and video game collections together on a single personal profile. Users can rate and review favorites, create and share lists, and find what to watch or play next. The product grew out of his frustration with IMDb’s rating experience and Letterboxd’s lack of video game support. Peter also treated Tastemaker as a product experiment, testing whether Claude Design and Claude Code could turn a rough idea into a usable product. His workflow involved creating a design.md and HTML spec, prototyping the interface in Claude Design, then building and iterating with Claude Code. He plans to show the process in a full video tutorial. Because of API restrictions, Tastemaker initially offered only 100 free personal profile slots.
https://x.com/petergyang/status/2082323512069685575
https://x.com/petergyang/status/2082254852600873376
https://x.com/petergyang/status/2082254840655405293
Dan Shipper, Every CEO
Dan Shipper spent one morning on his couch completing an article on Codex’s history. He had conducted hours of in-depth interviews with team members including Thibault Sottiaux, AJ Ambrosino, and Greg Brockman. Throughout the creative process, he never touched a keyboard or mouse, relying entirely on ChatGPT for Work’s voice mode. The voice workflow covered organizing interview material, building a timeline, writing, editing, and repeated revisions. The experience convinced him that voice mode is especially effective for creative work and can now support the entire production process for a complex long-form article. The piece is scheduled to appear in Every in a few weeks.
https://x.com/danshipper/status/2082130836485259530
Aditya Agarwal, South Park Commons Partner
Aditya Agarwal reflected on his Minus One phase ten years ago, when he was surrounded by a fog of uncertainty. That experience ultimately clarified two deeply personal directions. First, he wanted to create new things in the world. Second, he wanted to work with exceptionally talented, deeply committed people. Years later, he believes both opportunities are abundant today. His reflection also reveals the two enduring criteria he uses to judge career direction: creating new things and choosing high-quality people to work alongside.