The Daily Brief · Issue 6 · 21 August 2026ChatGPT can now read and send your iMessages on the Mac, and Slack turns channels into shared coding rooms for agents
Agents moved deeper into everyday tools: ChatGPT gained an Apple Messages plugin, Slack launched code channels with OpenAI, Anthropic and GitHub, and Google bundled Antigravity into Gemini Enterprise. Mistral showed a big accuracy jump from letting models search documents step by step.
AppsChatGPT for Mac adds an Apple Messages plugin
OpenAI's ChatGPT desktop app for macOS can now read and search your Messages chats and draft or send replies, 9to5Mac reports. It works with iMessage, SMS and RCS, on all ChatGPT plans, inside ChatGPT Work and Codex. The optional plugin runs locally on the Mac using AppleScript and Accessibility permissions, and it only looks at messages when you ask. By default ChatGPT sends nothing until you approve both the text and the recipients. OpenAI warns about granting permanent approval and notes that tasks set to 'Full access' may fail to show the confirmation needed to send.
Why it matters: Letting an AI agent send texts for you is useful but risky, so keep the default 'ask before sending' setting switched on.
Source: 9to5Mac
Developer toolsSlack launches Slack Code with OpenAI, Anthropic, GitHub, Cognition and Vercel
Slack introduced Slack Code, which adds 'code channels' where a team and its AI coding agents plan, write and review software together. Mention a coding agent and it spins up a channel, pulls in the right teammates and shows a code diff, planning doc or live HTML preview alongside the chat. When the work is done the channel archives itself and stays as an audit log. Launch partners are Anthropic, Cognition, GitHub, OpenAI and Vercel. Slack says over 70% of its own internal code channels open and close within a single day, from idea to merged pull request.
Why it matters: It moves agent coding out of private tabs and into shared view, which makes it easier for teams to catch agent mistakes early.
Source: Slack
Enterprise AIMistral's Agentic Search triples accuracy on financial filings in its tests
Mistral released Agentic Search for enterprise documents. Standard RAG (retrieval-augmented generation) grabs a few text chunks once and answers from them. Agentic Search instead lets the model search, open, navigate, read and grep documents in a loop until it can verify an answer. On FinanceBench, questions over 368 company filings, Mistral reports correctness rising from 26.7% to 86%. On the harder OfficeQA Pro test, GLM-5.2 went from 6.3% to 51.9%. Mistral also reports slow-case (p90) latency falling from 255 to 154 seconds. These are Mistral's own tests with default settings.
Why it matters: For long, table-heavy documents, how a model searches can matter as much as which model you use.
Source: Mistral AI
Enterprise AIGoogle bundles Antigravity into Gemini Enterprise and adds IDE extensions
Google's Antigravity coding agent is now included in eligible Gemini Enterprise Standard and Plus subscriptions, so companies no longer need a separate licence. Google also released Antigravity extensions for popular code editors, and developers sign in with their existing company identity instead of managing API keys. Admins get sandbox and tool-permission limits, monthly budget caps with shared token pools, and one-click audit logging of prompts and agent actions. Google says customer code is never used to train its base models. It cites AirAsia, where Antigravity now writes over 50% of production QA code.
Why it matters: Bundling a coding agent into an existing enterprise plan is a strong way to win companies away from Cursor and Claude Code.
Source: Google
ChipsMicron commits $10 billion to an AI memory research lab in Boise
Memory maker Micron announced a $10 billion, decade-long investment in a research lab in Boise, Idaho, focused on memory technology and compute systems. AI chips depend on high-bandwidth memory (HBM), stacks of fast memory placed right next to the processor, and shortages have become a bottleneck for AI hardware. The lab is expected to work on HBM4, next-generation DRAM and designs that do some computing inside memory. Micron has not given hiring or construction timelines. Rivals Samsung and SK Hynix are also expanding research and packaging sites.
Why it matters: Memory shortages are raising AI server prices, so more supply and better designs could eventually lower the cost of running models.
Source: insideAI News
Model watchAnonymous 'Ox Alpha' coding model appears on OpenRouter
OpenRouter listed Ox Alpha, a 'stealth' model whose maker chose to stay anonymous during a free preview. It is described as a reasoning model for coding, long agent tasks and production work, including tasks that mix text with images, and it has a context window of 1,048,576 tokens (the amount of text it can consider at once). Under OpenRouter's stealth terms, the unnamed provider keeps prompts and completions, though not for training. Developers quickly began testing it on coding leaderboards.
Why it matters: Stealth previews let labs test models on real users before launch, but anything you send is kept by an unknown company.
Source: OpenRouter
PolicyOpenAI launches a Strategic Futures team to study AI and concentration of power
OpenAI launched Strategic Futures, a small team with a blog called Intelligence Age. Its central question is how free societies can protect individual rights and agency as advanced AI arrives. The team argues that power has always depended on human labour, soldiers, police and taxpayers, and that AI and robots could let states or companies hold power without broad public support. It calls this concentration of power the largest and hardest long-term AI risk. The launch post sets out the problem rather than specific policy proposals.
Why it matters: It signals OpenAI wants a voice in debates about who controls AI, just as it prepares for an IPO and faces safety criticism.
Source: OpenAI