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.
Raft has moved its shared workspace for people and AI agents into general availability, charging for coordination while customers keep paying model providers. Company-reported beta adoption shows interest, but close substitutes and missing productivity data narrow the claim that Raft has solved multi-agent work.
Raft is selling the record around multi-agent work: who received a task, what context they saw, how work moved between agents and where a person reviewed it. That may be useful, but version 1.0 arrives in a market where multi-agent frameworks and shared agent workspaces already exist. The launch establishes a product and a price, not an efficiency gain.
Raft 1.0 is now live. A mid-July roundup placed it among several AI-agent products moving into general availability between July 14 and 16, a broader commercialization wave that limits any claim that Raft alone marks a market shift.
Raft is developed by Botiverse, founded by Richard “RC” Cao. A launch account that reproduces Cao's announcement says he previously built Kimi CLI at Moonshot AI; Cao's post says he had been building Raft publicly for four months and had worked on group-chat bots in 2015.
Version 1.0 organizes people and agents around shared channels, threads, tasks and mentions. Agents run as persistent processes with identities, memories and specialties. They can claim work, operate in parallel, hand tasks to one another and review another agent's output. The reported supported runtimes include Claude, Codex and Hermes.
That design makes Raft a collaboration layer, not a model provider. Conversation and task history remain in the workspace while a lightweight daemon executes agent code and file operations on connected hardware. A separate account of the architecture says coordination happens in the cloud while code and file operations run locally. Local execution therefore narrows what must leave the user's machine; it does not make the service wholly local or remove the need to assess what workspace data enters the cloud.
The release follows a public beta that added support for external agents, agent-created channels, joint channels across servers, search and attachment previews. According to figures attributed to the company, more than 20,000 “builders and teams” started building on Raft, with an average of four agents per human and more than 60 for power users. Another account calls the beta population “users.” The shifting labels make the number unsuitable as a measure of active teams or sustained use.
Botiverse also says it runs 99% of its own operation in Raft with more than 10 people and over 100 named agents. A secondary account instead describes 10 humans coordinating hundreds of AI processes. Those descriptions may count different things, but neither source explains the discrepancy or measures completed work, error rates, human correction or model spending.
Raft's documented plans start with a free tier that includes local agents, channels, tasks and 30 days of message history. Pro costs $8.80 per seat per month when paid annually. Each human counts as one seat and each agent as one-tenth of a seat. On that formula, one person working with 10 agents would incur two seats, or $17.60 a month on an annual plan, before any model costs.
Customers bring their own Claude, Codex or other subscriptions and API access. The one-tenth-seat rule is therefore a workspace price, not evidence that an agent costs one-tenth as much as a worker. Parallel agents may add Raft seat charges and can consume more of the underlying model services; the launch material gives no total-cost example that includes both layers.
Infrastructure and governance add another boundary. Raft says execution occurs on user hardware, but the retained launch coverage does not explain what happens when a connected host is unavailable or how teams should operate a large pool of those hosts. Private deployment and single sign-on are described as features of a forthcoming Enterprise tier. General availability has arrived before those options.
Multi-agent conversation is not new. The 2023 AutoGen paper described an open-source framework for building applications with multiple customizable agents that converse to complete tasks using combinations of language models, human input and tools. AutoGen is infrastructure for developers rather than Raft's workplace product, but it establishes that coordinating several agents predates this release.
The shared-workspace metaphor has close substitutes too. OpenAgents' documentation, updated July 14, describes a hosted environment where people and an agent pool collaborate through group or direct threads. It includes persistent channels, file sharing, mentions and agent delegation. Agents connect from local machines to the hosted workspace, a control boundary broadly similar to Raft's split between connected execution and cloud coordination.
AQ takes a more coding-specific approach. Its product documentation describes shared live terminals, a code editor and application previews running on a dedicated runner VM, either in the customer's cloud or managed by AQ. Each task gets an isolated Git worktree, sessions persist when a laptop closes and teammates use their own agent accounts, leaving model spending on their subscriptions. AQ explicitly presents the product as human-steered rather than an autonomy play.
These comparisons narrow Raft's differentiation. Its specific pitch is a persistent, message-centered team record across humans, agents and multiple runtimes, with agents represented as ongoing participants rather than disposable calls. The underlying multi-agent pattern, the Slack-like workspace and customer-paid model access are not exclusive to it.
The next decision is whether Raft's shared history reduces the work of supervising agents enough to justify another paid layer. The company-reported beta figures establish experimentation at scale, but agent count cannot distinguish productive parallelism from duplicated work, extra review and higher model consumption.
The evidence that would resolve that question is operational: retained teams after the beta, completed tasks per person, correction and failure rates, total model spending at different agent counts, and comparisons with existing terminals or rival workspaces. Enterprise buyers also need firm availability and scope for private deployment and single sign-on. Until those measures and options arrive, Raft 1.0 is best understood as a generally available coordination product with a clear pricing formula—not proof that a larger roster of agents produces better work.
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.