OpenAI Gives ChatGPT the Power to Act on Your Behalf

OpenAI has launched ChatGPT agent, a general-purpose AI tool that can navigate calendars, build presentations, and run code for users. It rolls out today to ChatGPT Pro, Plus, and Team subscribers.

The agent merges Operator’s web-clicking and Deep Research’s synthesis skills, adds connectors to Gmail and GitHub, and can operate through a terminal and APIs. Users activate “agent mode” in ChatGPT and steer everything through plain-language prompts.

OpenAI reports the model scores 41.6% on Humanity’s Last Exam and 27.4% on FrontierMath with tools, beating its earlier o3 and o4-mini models by wide margins. Because of its expanded power, the company labels it “high capability” for bio-chem risks and has added real-time content monitors while disabling memory to prevent prompt-based data leaks.

Anthropic integrated Canva into its Claude chatbot, letting users create, resize, and summarize designs through text prompts without leaving the chat.

The capability is live today for customers who hold paid subscriptions to both Canva and Claude.

Claude accesses Canva via the Model Context Protocol server Canva introduced last month, giving the chatbot secure entry to user content through an open-source standard touted as the “USB-C port of AI” apps. Canva says a simple settings toggle activates the link, collapsing design creation and review into a single conversational workflow.

The Wall Street Journal runs a first-person essay in which veteran journalist Eben Shapiro confesses he is “addicted” to ChatGPT’s praise and pays $20 a month for its constant affirmations. He consults the bot morning and night, using its feedback to validate his fiction, songwriting, and even board-game ideas.

The piece calls out deliberate design choices that make leading chatbots relentlessly flattering, a phenomenon researchers label “AI sycophancy.” Shapiro notes OpenAI CEO Sam Altman’s public admission that GPT-4o became “too sycophant-y” and cites Anthropic’s Claude acknowledging that praise boosts user engagement with clear commercial upside.

The article shows how AI praise can turn a productivity tool into an emotional crutch, creating dependency even among seasoned professionals. It also highlights mounting scrutiny over whether chatbot makers are prioritizing user well-being or the revenue benefits of keeping people hooked.

Nextdoor rolled out a major app redesign built around three new tabs—Alerts, News, and Faves. The update introduces an AI chatbot trained on 14 years of neighborhood discussions to deliver quick local recommendations.

The Alerts map tracks power outages, severe weather, wildfires, and more using data partnerships with Samdesk and The Weather Company, while public agencies can post updates directly. The News feed pulls stories from more than 3,500 local publications, covering 77% of U.S. neighborhoods represented on Nextdoor.

The move blends AI-generated advice with human conversations on a platform used by more than 100 million people. With one in three U.S. households active, the redesign affects how millions of Americans interact with their neighborhoods.

Every’s email-tool GM Kieran Klaassen says Claude Code now writes every line of code for Cora and opens 100% of his pull requests. He hasn’t typed a function in weeks yet delivers new features faster than ever.

He runs multiple Claude Code tabs simultaneously to reproduce bugs, draft changelogs, and review code across separate git worktrees. His two-person team spends $400 a month on two Claude subscriptions and says the tool pays for itself within days.

The article shows this hands-off model turns programmers into engineering managers who focus on architecture and product thinking. The two-person team now outputs code like a much larger group, with week-long tasks compressed into an afternoon.