Sponsored by
Welcome, AI enthusiasts
Inside OpenAI, the finance team now closes the month in hours, and sales can turn a first call into a working pitch by the next morning. The tool behind both just started reaching everyone else. It is ChatGPT, now able to take a goal and hand back the finished work. That capability is about to meet the ordinary tasks that fill most people's days. Let's dive in!
Last call! We start tomorrow: this is your final chance to join our live 2-Day Agentic AI Workshop and build real AI agents, MCP servers, and autonomous workflows alongside us. By Sunday evening, you’ll have working systems running on your own machine. Join now →
In today’s insights:
OpenAI wants ChatGPT to Work on Your Job
Claude's New Feature Wants You Using It Less
Elon Musk Is Building a Creator Machine
Read time: 4 minutes
LATEST DEVELOPMENTS
Evolving AI: penAI already runs nearly all its own teams on the agent it is now opening to everyone.
Key Points:
ChatGPT Work acts across your apps and files, turning one goal into finished sheets, slides, docs, and web apps.
At OpenAI it already cut finance month-end close from days to hours and turned a sales call into a pitch within 24 hours.
It is on macOS desktop for every plan today, with Windows, web, and mobile following in days.
Details:
ChatGPT Work runs on OpenAI's Codex, the agent technology that already draws more than 5 million users a week, over a million of them working outside software. It links to more than 1,400 apps, from Slack and Google Drive to most major CRMs. Before it starts, it lays out a plan for approval, then runs it on a schedule or stays on the job for hours.
Why It Matters:
Sam Altman is putting a finish-the-work agent in front of everyday users, so the busywork of pulling files together and formatting them can shift onto the agent. The judgment and the final call still sit with the person. For many workers, this is the first time a finished file arrives without them building it.
The best voice models, now across all channels
Most CX platforms do not own the voice. They orchestrate a workflow, then call a third party for speech and transcription. Every hop adds latency, cost, and another vendor to manage.
ElevenAgents is the opposite. They make the voice models the market builds on, and ElevenAgents puts full orchestration on top. Voice, transcription, text-based chat, and reasoning run in one vertically integrated pipeline, so responses come back in <400 milliseconds and sound human, not synthetic.
Plus, you keep full control. Plug in any LLM, integrate tools, webhooks, and MCP servers, and ground responses in your knowledge base. Get an agent live in minutes, then A/B test with Experiments, enforce Guardrails, and version every change.
The payoff: more human conversations, lower latency, and far less time stitching infrastructure together. You build on the models you already trust. Pricing is transparent and flat at $0.08 per minute.
ANTHROPIC
🪞 Claude's New Feature Wants You Using It Less
Source: Claude
Key Points:
AThe Reflect beta opened on July 9 to Free, Pro, and Max accounts with memory turned on.
Users can set quiet hours and schedule dismissible reminders to step back after a set stretch of use.
Reflect periodically asks what a person wants to keep doing themselves, even when Claude could do it faster.
Details:
Anthropic designed Reflect to break down the topics, tasks, and times behind someone's Claude use across the past 1, 3, 6, or 12 months on web and desktop, with a total-time view on the way and the phone app still to come. Reflect scores that against a fluency framework of delegation, description, discernment, and diligence, built with MIT Media Lab and Boston Children's Hospital. Incognito and health chats stay out of the report.
Why It Matters:
Claude is one of the first major assistants to give people a clear picture of their own use. For anyone leaning on it daily for writing, code, or decisions, Reflect makes a vague sense of over-reliance concrete enough to act on. Seeing which topics and tasks pile up gives someone a real basis to pull back before the habit hardens.
AlphaSignal's 8-person team automated 18 workflows. No developer.
AlphaSignal runs the most-read technical AI newsletter with an 8-person team. One salesperson installed Viktor for prospect research. Sixty-seven days later it runs 18 workflows across sales, ops, editorial, and finance: proposal builder, deal updates, competitor monitoring, P&L analytics. No developer hired.
Source: Meta
Evolving AI: Meta's Muse Spark 1.1 tops rivals on its own tool-use tests and is free in the Meta AI app.
Key Points:
Muse Spark 1.1 is Meta Superintelligence Labs' new agentic model, built to call tools, operate computers, and run multi-step jobs.
On Meta's own benchmarks it leads Opus 4.8 and GPT-5.5 on tool-use tests like MCP Atlas and JobBench, while trailing them on coding.
Anyone can use it free in Thinking mode in the Meta AI app, and developers pay through the new Meta Model API.
Details:
Muse Spark 1.1 is tuned to use tools, not just answer, calling functions and APIs to finish jobs and handing parts of a task to subagents. It is free to try in the Meta AI app's Thinking mode. Meta reports it beats Opus 4.8 and GPT-5.5 on tool-use tests it selected and ran, with a wide lead on JobBench, though it trails both on coding. Developers pay $1.25 per million input tokens and $4.25 per million output through the Meta Model API.
Why It Matters:
Meta is handing everyday users a tool-using agent for free, which puts real task automation in reach without a subscription or code. For readers that means chores like booking and form-filling can run on the model instead of by hand. The edge sits mostly on tests Meta ran itself, worth watching as independent scores land.
👀 Our tip
Last chance: our live Agentic AI Workshop starts tomorrow. In just one weekend, you’ll build working AI agents, your own MCP server, and autonomous workflows that run on your machine, with no prior experience required. Join us July 11–12 and leave with practical skills, reusable templates, and real systems you can keep building on.
👀 Click on the image you think is real
QUICK HITS
☁️ Kling AI's Revenue Quadrupled the Same Month OpenAI Killed Sora's Consumer App.
🏦SK Hynix's Nasdaq Debut Beats Alibaba's Record, Second-Largest US IPO Ever.
⚡ Cerebras Surges 12% as It and Flex Move to 7x Their California Chip Output.
🎙️ Nurix AI Acquires Verloop.io to Merge Voice and Chat Automation Across India and the Middle East.
📈 Trending AI Tools
🎞️ Guideless - AI tool that automatically turns your clicks and workflows into step-by-step video guides*.
🔗 Zapier - Connects thousands of apps to automate repetitive work.
📡 CiteRadar - Tracks whether ChatGPT, Claude, and Perplexity recommend your brand.
🎧 Podcastle - AI studio that records and edits podcasts from the transcript.
*partner link