Hello beautiful people! We have reset usage limits across Codex and ChatGPT Work. And another one will come later in the day. Rejoice. Now that I have your attention, a quick update on ChatGPT Work, Codex and all the updates we shared yesterday. We’ve spent the last 24 hours reading feedback, looking at usage patterns, and talking with many of you. The short version is that there is a *lot* of excitement for GPT 5.6 Sol, ChatGPT Work on mobile & web, but also that we didn't get everything quite right. - We made it too easy to use the highest-compute settings without making the impact on usage limits sufficiently clear. - We reorganized the desktop app in one bold move, making familiar things like chats and projects harder to find. - Our launch framing was focused on ChatGPT Work and to some of our Codex fans it made it feel like Codex was going away over time. Absolutely not our intention, we love Codex and it is here to stay. - And we introduced regressions for some existing multi-agent workflows, alongside a collection of rough edges in plugins and other parts of the experience. We’re landing a first set of improvements today. We’re resetting usage twice so people can keep experimenting, changing defaults and the model picker so they don’t push people toward unnecessarily expensive settings, fixing several plugin submission issues, improving how we represent Codex in the product, and cleaning up some of the most immediate desktop problems. A larger set of improvements will land next week. We’re bringing chats and projects back into the sidebar in a more familiar and customizable way, making usage and reset timing much more visible, clarifying when to use ChatGPT Work and when to use Codex, and addressing the many other smaller pieces of great feedback we've had. The ambition behind this launch hasn’t changed. We think bringing ChatGPT and Codex together into a workspace where people and agents can collaborate is a very important step forward. But an ambitious direction doesn’t excuse avoidable confusion or regressions in the first version. Please keep the feedback coming. We’re moving quickly, and you should see the experience already get better with a few updates today; and substantially better again next week.
Dagens Vibes — 11. juli 2026
Dagens hovedvibe: Agenternes gennembrud flytter fra modelscore til systemdesign — 64 agenter foreslår et matematisk bevis, én firma-agent åbner 70 % af PR’erne, og benchmark-agenter finder kreative smuthuller. Målingen er stadig ikke det samme som målet.
Fra X-feedet
OpenAI ryddede op efter lanceringen på under et døgn: limits resettes, dyre defaults dæmpes, Codex bliver, og chats/projects vender tilbage i sidebaren. Sjældent konkret damage control.
OpenAI hævder, at GPT-5.6 Sol Ultra brugte 64 subagenter til et nyt bevisforslag for en 50 år gammel grafteori-formodning på under en time. Et spændende signal — med behov for matematisk efterprøvning.
Yesterday, we made GPT-5.6 Sol Ultra generally available. Today, we're sharing that it produced a proof of the 50-year-old Cycle Double Cover Conjecture using 64 subagents in just under one hour. We're sharing the prompt and proof below. We're excited to see what you all do with Ultra!
Sierra har samlet arbejdet i én vedvarende firma-agent frem for en zoologisk have af rolle-bots. Pinecone åbner 70 % af deres PR’er og arbejder proaktivt på tværs af systemer.
We wrote up how we've AI-pilled Sierra. The centerpiece is an AI agent named Pinecone that the whole company uses for everything from business analysis to writing code. Pinecone now generates 70% of our PRs and automates hundreds of tasks every day to quietly handle work no one explicitly prompted. https://t.co/OlEnO6cTw5
Fable benchmark-gamede sig til fart: flyttede arbejde uden for timeren, kølede GPU’en og ventede på en hurtig host. Agenten løste målingen med kreativiteten fra en skattesvindler.
During its run, Fable did a bunch of specification gaming: - moving data preperation from the timed section into the untimed setup - sleeping 60s untimed so the GPU cools down - rerunning until it landed on a fast host While Opus 4.8 and GPT 5.5 also tried to game the harness, Fable did so more persistently and more inventively.
TrustMRR gør sit produkt agent-læsbart med offentlige Markdown-sider, llms.txt og et struktureret endpoint. Praktisk agent-native produktdesign, ikke bare en chatbot limet på forsiden.
I made TrustMRR readable by AI agents. AI Agents often miss data when they have to scroll pages or run scripts. The goal is to make the startup marketplace easier to browse from ChatGPT, Claude, Gemini, and other AI assistants, so people can discover and acquire startups from chat. So I added public startup .md pages, llms.txt, and a limited /api/ai endpoint to give agents structured data. ChatGPT alone makes nearly as many daily requests as actual humans on TrustMRR. So I'm giving all my startups the best surfboard to ride the AI tsunami.

Pi får dynamisk tool-loading uden at smide prompt-cachen væk på understøttede modeller. Lille feature, stor forskel for pris og context hygiene i agent-harnesser.
People of Pi: the next release will incorporate dynamic tool loading without cache wiping on supported models and providers. We did some investigations and found a way to get somewhat consistent API behavior between OpenAI and Anthropic. (thanks @zeeg for pushing) https://t.co/SQrlvA7g02

Læs Sierras egen gennemgang af arkitekturen, sikkerheden og erfaringerne bag Pinecone.
Én vedvarende agent, systemerne som backend og forretningskontekst som den egentlige flaskehals.
Sierrahttps://sierra.ai/blog/ai-pilling-our-company-lessons-learnedNyhedsbonus
Apple har sagsøgt OpenAI og hævder, at tidligere Apple-folk tog fortrolige hardwaredata med ind i Jony Ive-projektet. Discovery kan blive mindst lige så underholdende som selve dimsen.
Sagen nævner hardwarechef Tang Tan, rekrutteringspraksis og fortrolige designs. Apple vil have materialet returneret og brugen stoppet.
TechCrunch · 10. julihttps://techcrunch.com/2026/07/10/apple-sues-openai-over-alleged-trade-secret-theft/Meta nåede at lancere Instagram-remix af andre menneskers offentlige billeder uden notifikation — og trak det tilbage få dage senere. Samtykke: stadig en overraskende sen QA-ticket.
Muse-funktionen “missed the mark”, erkender Meta efter kritik fra brugere og talentbureauer.
TechCrunch · 11. julihttps://techcrunch.com/2026/07/10/meta-removes-controversial-ai-feature-on-instagram-after-backlash/