OpenAI has put conversations and Projects back in its redesigned ChatGPT desktop app and enabled cloud Work threads to move across devices, correcting the launch's biggest usability failures without merging local Work or Codex histories.
OpenAI has restored conversation history and Projects to its redesigned ChatGPT desktop app one week after launching a client that put those core ChatGPT features out of easy reach. The repair makes the app usable again for ordinary chat, but it does not create one history, one execution environment, or one availability model across Chat, Work, and Codex.
That distinction matters because the redesign is also OpenAI's distribution point for ChatGPT Work, a longer-running agent aimed at professional tasks. The corrected sidebar brings users' existing conversations and project context beside the new product; the underlying boundaries still determine which work travels, which files an agent can reach, and which usage limits apply.
OpenAI says in its July 16 release notes that the desktop changes are live for all plans on macOS and Windows. A global switcher selects ChatGPT or Codex. Within ChatGPT, a second control selects Chat or Work.
The new sidebar combines Chat and Work threads in Recents, with controls to sort, filter, and pin them. Existing ChatGPT Projects also appear there, and users can begin either a Chat conversation or a Work thread with a Project's context.
Those are corrections to the initial release, not merely refinements. When OpenAI introduced the combined client on July 9, regular Chat opened in a floating window without direct access to existing history or Projects. Users objected on OpenAI's forums and social media, and OpenAI product leader Thibault “Tibo” Sottiaux promised to restore those features, according to an account of the rollout and response.
Sottiaux later said the company had not gotten its first attempt entirely right. In a translated account of his update, he said OpenAI would continue fixing smaller problems and improving performance, reliability, and efficiency.
The initial version also had a functional gap beyond navigation: cloud Work conversations were not visible in the desktop client, according to a contemporaneous account. The new release closes that gap by synchronizing cloud Work across web, mobile, and desktop.
The app now presents a clearer hierarchy, but each mode still follows different rules.
| Surface | What now appears together | What remains separate |
|---|---|---|
| Chat | Chat threads and Work threads share Recents; Projects can supply context to either | Chat is still a distinct mode inside ChatGPT |
| Work | Cloud threads sync across web, mobile, and desktop | Local threads, files, and outputs stay on the originating computer unless moved or shared |
| Codex | Codex sits behind the app's global switcher | Its workflows and history do not join ChatGPT Recents |
OpenAI's product guidance says desktop Work can use local folders, files, and applications when the user's plan and workspace allow it and the user grants permission. Work on web and mobile cannot directly reach files on that computer. A local thread stays on the machine, while a cloud Work thread can be continued on another surface.
Codex is a separate view for repositories, terminals, local folders, tests, commands, and code review. It is not selectable on the web or mobile. Supported desktop Codex threads can be reached through the Remote tab in the ChatGPT mobile app, but remote access does not turn them into ordinary web or mobile chat history.
The distinction is also commercial. The July 16 interface update is available on every desktop plan, but the July 9 release notes describe Work on web and mobile as rolling out to paid plans other than Free and Go, with Pro, Pro Lite, Enterprise, and Edu first and Plus and Business following. OpenAI says Work uses the same capacity-and-credit structure as Codex; actual consumption varies by task. A universal app update therefore does not mean universal Work access or identical usage economics on every surface.
Users arrive at the new app through different paths. Updating the former Codex app converts it into the new ChatGPT desktop app and should preserve existing Codex chats and Projects, according to OpenAI's migration instructions. Existing Codex users continue to open into Codex by default and can keep its icon.
Users of the previous ChatGPT desktop client instead follow a prompt to download the new application. The old client may remain installed as ChatGPT Classic, which OpenAI says will continue receiving model updates, bug fixes, security patches, and support for existing Enterprise capabilities. New agent features may be limited to the replacement.
That supported fallback limits the cost of the redesign for users who do not need Work or Codex, but it leaves OpenAI maintaining two ChatGPT desktop experiences. The migration can also be confusing: one critical account measured the new Mac app at 1.5 GB and documented an update dialog that failed to reinstall Classic for someone who had already tried the new client. Restoring the sidebar does not resolve either complaint.
The redesign gives OpenAI a prominent route from familiar chat history into heavier agent work. Work can research and analyze information, use connected applications, create documents and spreadsheets, build presentations and Sites, and run scheduled tasks. Codex remains the specialist environment for software development.
That positioning follows Anthropic's earlier Claude Cowork launch. Both products are attempts to make capabilities associated with coding agents available to non-programmers, while both companies pursue enterprise customers. An interview-based launch report said OpenAI and Anthropic were also preparing for possible public offerings, adding financial stakes to the competition for professional users.
The same report shows why price claims need a narrow comparison base. OpenAI officials said the new offerings would be cheaper and more broadly available than rivals' products. Max Weinbach of Creative Strategies said the smallest of OpenAI's three GPT-5.6 variants could complete a task about as well as the largest at one-fifth of the cost. That is an analyst's comparison for a task between two OpenAI model sizes—not evidence that Work is always one-fifth the price of Claude Cowork, or that every Work job has equivalent quality.
The repaired sidebar may reduce the friction of trying Work because users' existing chats and Projects are visible beside it. It does not establish a technical moat: the headline achievement is the restoration of expected ChatGPT behavior after a poor launch, while the competitive product remains subject to plan access, task-dependent credits, local permissions, and a direct substitute from Anthropic.
OpenAI has fixed the most damaging discoverability and continuity problems. The unresolved question is whether users can understand the remaining boundaries before they start a task, grant local access, or assume a thread will be available on another device.
The evidence to watch is operational: whether cloud Work threads and Project context remain consistent across web, mobile, macOS, and Windows; whether the interface clearly distinguishes cloud Work from local Work and remote Codex; and whether the new client's performance improves enough that Classic is a fallback rather than the more dependable choice.
For now, OpenAI has consolidated entry points and repaired navigation. Whether it has consolidated a product will depend on reliable cross-device behavior, transparent usage economics, and permissions that make the three modes' differences clear rather than surprising.
Meta reportedly plans to put departing AWS compute executive Dave Brown to work on its data-center buildout, adding hyperscale operating experience while leaving any customer-facing cloud business conditional and undefined.
China has paired a five-year AI training offer for developing countries with cooperation centers, a weather-warning rollout and a new 29-country organization. The package gives Beijing a platform for influence, but no budget, selection rules or delivery timetable has been published.
RoboTTT extends a robot policy’s visuomotor context to 8,000 timesteps and raises its average rubric score from 42% to 79%, but the evidence comes from three author-run assembly tasks and the longest task was completed in only two of ten trials.
Zhipu reportedly reached $1 billion in annual recurring revenue in July, roughly four times a March estimate, but the unconfirmed run rate is not annual sales and still sits far ahead of recognized cloud revenue while margins remain thin.
Nebius has arranged its first senior secured facility against an operating GPU deployment and one customer’s cash flows. The deal adds project-level debt to its expansion toolkit, but an unnamed customer and a larger rival financing limit what it proves about the rest of Nebius’s backlog.
Cursor says automatically running a repository-root git.exe on Windows does not meet its criteria for patching, while the researcher calls it an untrusted-search-path defect and separate research shows the same weakness across several competing AI coding tools.
Huawei publicly displayed a 16-cabinet Atlas 950 configuration rated at 1 EFLOPS in FP8, providing tangible evidence of its system-scale AI strategy while leaving price, power use and sustained workload performance undisclosed ahead of the full system's planned fourth-quarter release.
CIA Director John Ratcliffe said US intelligence is consistent with an estimate that Russian recruits last 20 to 30 minutes on Ukraine’s battlefield, but the public trail leads to an unsourced claim about assault troops and does not establish a representative average.
UK testing places leading open-weight models four to seven months behind selected closed-model cyber results, yet longer attack chains, U.S. benchmarks and mixed cost comparisons show why that interval is a warning signal rather than a universal capability clock.
TSMC reached the top of its second-quarter revenue guidance and raised its 2026 outlook, but a one-off investment gain boosted profit growth while the company committed more capital to 2-nanometer production, advanced packaging and an undated Arizona expansion.
Moonshot AI's Kimi K3 added to a global technology selloff with near-frontier performance, but unreleased weights, mixed cost comparisons and a recommended 64-accelerator deployment leave its effect on chip demand unresolved.
Moonshot AI has made Kimi K3 available through its apps and API, pairing a 2.8-trillion-parameter architecture with early frontier-level results, but the model's open-weight claim cannot be tested until its weights and technical report arrive.
Two binding EU decisions require Google to give rival AI services comparable access to 11 Android features and offer eligible search competitors a restricted, anonymized dataset, but phased deadlines, certification, pricing and privacy safeguards leave the competitive effect unproven.
Databricks has signed a term sheet for a Coatue-led financing at a $188 billion valuation, while unidentified sources put the round at $3 billion. The proposed capital would deepen its push into AI governance, data agents and operational databases, but the transaction remains open and the company supplied no new operating figures.
Netflix says generative AI workflows were used on roughly 300 titles in 2026, but its only quantified example tied to that disclosure covers 17 minutes and does not establish how much finished material, spending or labor changed across the slate.
Gemini 3.5 Pro missed its expected June rollout. An anonymous-source account says a late-June data change fell short of Google's coding goals, but Google has confirmed only partner testing—not the reported cause, a new date, or public results and pricing.
Google is bringing Gemini Omni editing and a reusable face-and-voice avatar to Vids, but the sharper distinction is account-level identity across Vids and Gemini rather than a new category of AI video; Vids already offered Veo generation and customizable avatars, while specialist rivals already sell digital presenters.
Intel plans to adopt Gemini Enterprise across engineering, supply-chain and corporate work while extending chip-simulation capacity into Google Cloud, but the companies supplied no rollout schedule, contract value or performance baseline.
The Navy has approved an immediate department-wide framework for turning data into operational effects, but the public rollout leaves budgets, owners, deadlines, performance measures and assurance rules to a promised implementation roadmap.
Runta has raised a $20 million seed round led by Andreessen Horowitz for an execution layer that governs agents at the operating-system and network levels, entering a market where cloud runtimes and sandbox providers already offer overlapping controls.